May 6, 2025 • 47 mins
AI is no longer experimental—it's foundational.
In this explosive episode, PrimeVP Founder & Partner Shripati Acharya and Principal Gaurav Ranjan unpack how AI is reshaping B2B software—and what that means for evaluating and investing in early-stage startups.
What you’ll learn:
⚙️ The rise of Enterprise AI and its real-world applications
📈 Prime’s investment thesis in AI-first SaaS startups
🧠 What separates hype from defensible value in AI
🚀 Advice for founders building in vertical SaaS, AI agents, and more
Timestamps:
00:00 – Introduction
02:00 – AI’s impact on startup scaling and team size
03:20 – Faster sales cycles with enterprise AI
07:26 – Will AI unlock India’s SaaS market?
13:36 – Has AI neutralised India’s cost advantage?
20:19 – Models vs APIs: Where’s the real business?
25:52 – Why UI still wins in enterprise software
27:27 – Building moats through workflow depth
34:38 – How to price AI products right
39:43 – Prime’s investment themes in enterprise AI
44:58 – Final thoughts
💡 Key Takeaways:
Where AI is driving real enterprise value today
Prime's active investment thesis in AI + SaaS
The rise of function-specific AI agents across verticals
Shifts in software buying behaviour—from IT to HR budgets
This is your tactical guide to the next generation of enterprise innovation.
📌 Follow us:
LinkedIn: https://www.linkedin.com/company/primevp
Twitter: https://twitter.com/primevp_in
Website: https://primevp.in
🔔 Subscribe for more founder-centric content from Prime
#AIRevolution #EnterpriseAI #SaaSInnovation #FutureOfWork #AIEcosystem #StartupGrowth #FoundersFirst #VCInsights #TechLeadership #AIInvesting #B2BInnovation #IndiaStartups #AIxSaaS #PrimeVenturePartnersPodcast
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
Zero to a couple of million dollars of ARR, even
tens of millions of dollars ofARR in a couple of months.
In some cases, maybe a year orso.
What do you think has changed?
I think that puts Indiancompanies at some disadvantage.
We can build the same qualityof applications whether you're
sitting in Bangalore or Bhopalor somewhere in Europe or US.
So in this world then, how doyou differentiate your product?
Speaker 3 (00:21):
The nuances required to build a successful enterprise
software are oftenunderappreciated.
Hello everyone, welcome to thisepisode of Prime in Japan.
I'm excited about are.
Hello everyone, Welcome to thisepisode of Prime Venture
Partners podcast.
Speaker 1 (00:51):
In today's episode, shripati and I will be
discussing the implications ofAI in this world of B2B software
.
I mean, every day we are seeingnew things happening with AI
new models being dropped, newbusiness models being cracked,
new companies being formed andscaled from literally zero
million to like a couple ofmillion dollars of ARR in a few
(01:12):
months.
So we'll cover a lot of thatand also how this impacts the
way we think about evaluatingand investing in early stage
startups.
Welcome, shripati, to the showAbsolutely.
Speaker 3 (01:25):
Thank you, Karan.
Speaker 1 (01:26):
So the first thing that we'd like to know from you,
shripati, is that traditionally, we've seen SaaS companies
scaling from inception to amillion ARR in a couple of years
, and then, from there on, theylike triple and triple and
double year on year.
In this new world, we arehearing about a lot of companies
going from literally zero to acouple of million dollars of ARR
(01:48):
, even tens of millions ofdollars of ARR, in a couple of
months, in some cases, maybe ayear or so.
What do you think has changed?
I mean, my assumption is thebuyer.
My understanding is the buyerpersona remains same, the buying
behavior more or less remainssame.
So what has changed?
Because of which companies areable to scale so fast?
Speaker 3 (02:08):
only a couple of comments.
One thing is that remember thatthe we only hear about the
companies which are goingthrough these breathtaking, uh
growth rates.
I'd hear of cursor going fromzero to 100 in like a year right
, stuff like that or two years,whatever the number is and, you
know, competing with some of theother companies which have gone
(02:28):
since the product launch.
So Glean I think the number wasthree years, and this is faster
than Glean is what I recall,200 million ARR.
So, yes, we hear about thefastest of the fast-growing
companies, but we should notconclude from that that that is
the norm for every company,right?
I'll just make that comment.
That said, some things arechanging and I think one of the
(02:51):
most striking features of thenew generation of B2B SaaS
companies is that the size ofthe company is smaller.
That's like number one, and theanswer for that is like a more
simpler answer, which is thatyou require a smaller team to
generate the same quality ofcode, thanks to, you know,
co-pilots for coding and soforth, and also the use of AI in
(03:15):
all the other aspects of aproduct.
Right, you have sales, you havemarketing, you have insights,
sdrs all of these aspects aremade efficient and faster and
better with the use of AI, right, so that's why the companies
are smaller and that's actually,I think, a systemic trend.
The other thing which I feelwhich is different about this
(03:37):
and where the sales cycles aregoing to be faster in many
perhaps not all, but manyenterprise businesses is that
the time to wow, which is thetime it takes for a customer to
look at a product and say that,hey, this is going to actually
really be driving meaningfulvalue to me as a business, as an
(03:59):
increase in top line or asignificant cost saving is lower
, so that gap is just lower.
So if you just take a step backand look at how the cloud
adoption, for instance, happenedin the enterprise, it took a
while for that to happen,because it's not like these
large enterprise let's just take, you know why, wall Street
Enterprise, or even like a largeuniversity or any of those
(04:19):
things which had a deepdeployment of software.
They had on-prem software right, or their own captive data
centers, and then, when that,when cloud came in, they had to,
like, make a decision, likeokay, how much benefit am I
going to get?
Is there a significant changein the way I'm doing things?
What will I do with my existinginfrastructure.
And, of course, over a periodof time, they realized that, you
(04:40):
know, I can scale seamlesslyand it's more OpEx versus CapEx
and so on and so forth.
But in the case of AI, I feelthat the decision makers are
able to see that much moreclearly.
So, for instance, they might beable to say wait a minute, this
particular piece of software isgoing to enable me to respond
(05:01):
to my customer faster, developmore products faster or actually
reduce my headcount in customerservice, which is like our
support which is, you know, inthe news all over the place.
So I feel that that time tovalue and the leap of faith
which the enterprise decisionmaker has to make has come down
(05:23):
significantly, which means thesales cycle are smaller,
companies are smaller, theability to create a product is
faster and the sales cycle aresmaller.
But that doesn't mean that theaverage company would actually
go to 10 million of year one.
But I do think that the totalrevenue ramp for successful
(05:47):
companies is going to be higherthan the pre-AI generation of
companies.
What do you think I mean like?
Does that make sense?
Speaker 1 (05:54):
I mean you covered everything.
One thing that we also see.
I mean we do speak to a lot ofpotential AI customers the way,
like 15, 20 years back, everyenterprise had a digital
transformation budget.
These days, everybody has an AIbudget, so they're open to
trying new things and, as yousaid, like with these AI
(06:14):
solutions, the time to wow islike very fast, you can quickly
see the value, so they have abudget for it.
Once they start investing in it, they see the value very fast
and then, from there on, it'svery easy for them to adopt.
So that's the other factor thatwe have seen, at least in large
enterprises, where they have adedicated budget for AI, which
at least creates a room forexperimentation pilots and from
there on it converts to a realcustomer.
Speaker 3 (06:37):
Yeah, and if you look at, like the history of like
really the various you like,really the various stages or
errors in software, like we hadthe PC software, then we had the
client server computing andthen we had the cloud and mobile
and so forth, in all of thesecases, if you think about the
person on the other side, thiswas not a product which they
(06:59):
were familiar with, right, theywere not using client server
computing in their day-to-daystuff.
They were not using cloud intheir day-to-day uh stuff, or
even if it is there, it was veryabstract.
But I feel also that now everydecision maker, if not them,
their children, are definitelyusing, uh, you know, chat, gpt
or equivalent or gemini, uh andcloud and so forth, right, so
(07:22):
they actually have a firsthandimpression and a firsthand
experience with really the magicof AI, because it's quite
magical, the user experience,and that also helps because now
they're able to correlate, likethis is happening and this demo
is actually delivering value,which I can see.
(07:42):
And perhaps there's the otherthing which is also happening
that competitive dynamic herethey can also see that they're
competitors.
There's so much noise aboutindustries adopting it.
So I feel that all of this isleading to a significant
compression in the cycles.
I think it's a good.
It's an exciting time from thataspect because one of the
biggest issues in enterprisesales as you and I know from our
(08:03):
direct experience investing inthis field is a sales cycle like
.
Speaker 1 (08:07):
It takes time to sell and a time to deploy yeah, the
other thing we're seeing some ofthat impact even in the indian
market.
Traditionally, like indiamarket, has been not very
attractive when it comes to sascompanies.
Like going after india market.
Now, with ai coming in, atleast we're seeing some early
signs where enterprises arewilling to adopt AI solutions,
(08:28):
are willing to pay for that.
Do you think they'll have anychange in the India software
market?
Like with AI coming in, do youthink things will change and the
market will become more?
Speaker 3 (08:37):
interesting.
So India software, indiaenterprise SaaS has always been
a challenging market, in myopinion, from a market size
perspective.
So, honestly, the threecompanies we all keep talking
about which have a billiondollars or more of revenue are,
from an enterprise softwarestandpoint, salesforce, which
has talked about it's more thana billion dollars in India, and
SAP and Oracle, which haven'tdisclosed the India-specific
(08:59):
revenues, but likely quitesignificant.
They are in the same zone.
But if you think of these,these are like one very
foundational pieces of softwareand it's taken a while, with
their heft and brand the globalbrands to actually get to that
level.
So until now we have at leastnot seen a significant total
(09:23):
addressable market forenterprise SaaS in India.
I feel that AI is really goingto open that up.
Now how big and how fast itopens remains to be seen, but
I'm very optimistic about it andthe way I'm thinking about it,
gaurav, is that AI is coming inwith a software solution in
areas where there was actuallyno software at all.
(09:45):
So, for instance, think about aform filling thing.
You're actually filling I don'tknow your compliance audits.
Think of just take anyparticular area like that which
you traditionally think about.
Wait a minute, it's a lot ofpeople, a lot of form, filling,
a lot of, you know, readingregulations, this, that, and
filling and submitting and soforth, and a back and forth
(10:05):
going on with the serviceprovider, with some government
website and this and that andthe other.
Just putting software thereseems like a complete nightmare.
Right Come AI.
Now you can actually visualizeand you're already beginning to
see, at least in other markets,folks actually delivering
software for that, which canlook at all your existing
(10:26):
internal documentation, whichcan correlate with external
documentation, come up with therecommendations and so forth,
and then also help you in theworkflows there.
So I feel that software now isentering areas where previously
the substitute was humans doingpaperwork, humans using
(10:46):
essentially spreadsheets, emailor a word processor like
Microsoft Word or something likethat to do their work Like, if
you ask, even like lawyers,there are two big pieces of
software which they use areMicrosoft Word and email.
Yes, right, so that's what youare really displacing, right, if
you look at the larger legaltech companies that are coming
(11:07):
up, especially in the US.
So I feel that we have thatopportunity now in India, which
is that it's not likeenterprises are not doing
workflows, just that theworkflows are entirely manual.
So now software can actually goand deliver value.
There and in our earlier comment, there are time to wow for that
which is value, realization ismore obvious.
(11:27):
So thing which remains to beseen is that what is the average
order value or average contactacvs of these kind of things?
I think that some discovery hasto happen there.
Obviously it won't be as largeas you know some of some of the
ACVs we are seeing in AI SaaScompanies in the US, but I
believe that you know it's goingto start becoming a significant
(11:49):
market now, precisely for thatreason.
Certainly, that's my belief.
Speaker 1 (11:53):
Yeah, I totally agree with that.
Like new use cases have come upwhich has not previously been
addressed by software.
In fact, one of our portfoliocompanies which is in the QC
space, like computer visionbased QC, yeah, quality control
you mean.
Yeah, quality control Onassembly lines.
You had people looking at stuffcoming out of assembly line and
looking at that and discardingthe bad stuff.
(12:13):
Now with computer vision, ai,you can do that in real time
with the model sitting on theedge and things happen very fast
.
Yeah, so this use case was notpreviously possible with like a
simple software.
Computer vision models werethere around but the models were
not great, you could not deploythat on the edge.
Speaker 3 (12:29):
Now, with lighter models, faster models, you can
do that quite easily yeah, andin one sense, this actually
opens up fairly large and you,you said, right, the AI
transformation budget.
Yes, ai transformation budget.
So this is a conversation whichis not just happening in the
CEO's office, it's happening inthe boardrooms of these
companies, right?
So if you're an investor andthe first thing which we ask all
(12:53):
our portfolio companies whichwe are involved with is, hey,
what is your plan?
And we were asking this back in23 when GPT came out, right,
and this is a conversation whichis in all large enterprise
boardrooms as well which is howare you going to leverage AI?
So this is top of mind for Fools, so it is not something where
(13:13):
you have to, as a startup, goand sell that.
Hey, look, you need to deployAI.
And the question is can youactually have a demonstrable
return on investment,demonstrable ROI for them?
Yes, right, that's the key interms of these things.
So I'm hopeful.
I think that this definitely isa new era.
It definitely is going to be alarger market than our existing
(13:34):
TAM.
Whether it will openmulti-billion dollar TAMs within
India, I don't know, but we'llsee.
I think it will definitely leadto more successful startups in
India, targeting Indiaenterprise.
Speaker 1 (13:46):
Yeah, I agree.
I mean, while we hope that theIndia market opens up for
software, at least with aforcible feature, we do see US
as a large market for companiesbuilding out of India.
And so far, one advantage thatwe had was the cost advantage.
Whether it was in terms ofdevelopment, testing, sales
support, all of that India hadan inherent cost advantage.
Whether it was in terms ofdevelopment, testing, sales
support, all of that india hadan inherent cost advantage
(14:07):
compared to a us native company.
Now, with ai coming insomething that we were talking
about earlier the size of theteams have come down.
Ai will do heavy lifting interms of sales and support, with
ai agents taking over humanagents or like working together
human agents.
Do you think that puts Indiancompanies at some disadvantage
or takes away the advantage thatthey had before when it comes
(14:29):
to the cost structures andthey'll have to compete at par
with a US company which willhave a similar cost structure as
an Indian company?
Speaker 3 (14:37):
So, as the, I mean that's definitely is a very
significant aspect which is atplay here in the sense that the
coding teams are smaller and thesales force is smaller, because
now the outbound calling withall the agent, API software and
so forth, the support sales opsoperation, for example, is
(15:00):
smaller and customer supportteams are smaller and so forth.
So if the total size of thecompany has shrunk, then
definitely the arbitrage youmight have from a development
team in India versus adevelopment team in the US
becomes definitely significantlysmaller.
I think that aspect you'reright, Gaurav significantly
smaller.
(15:20):
I think that aspect you'reright, Gaurav.
I am also seeing, though, thatthere is a potentially
countervailing force here, whichis that a lot of a new
opportunity we believe onenterprise SaaS is service as a
software or outcome as a service, or however you want to call it
, wherein the end customer isgetting service and it's just
(15:44):
like there might even be a humaninterface.
Think about it as a contactcenter or a customer support,
where you're actually talking toa customer agent and you're a
medium or a large enterprise,and I feel that a lot of the
non-tech companies will not wantto in-house this.
So, for instance, you knowthere's a lot of talk about
clarna.
Companies will not want toin-house this.
So, for instance, there's a lotof talk about Klarna, which is
this fintech wherein they wentahead and essentially reduced
(16:08):
their customer support by 90% ormore and have the same NPS,
same customer side by goingentirely into an agent tech bot
which is actually doing bothvoice and text support.
Right, but if you think about aregular enterprise, they both
voice and text support right,but if you think about a regular
enterprise, they're not goingto go ahead and develop that
software in-house first of all.
So the transition we see inthose cases is those enterprises
(16:30):
telling their service providersthat hey look, we want you to
use AI on your thing.
We should translate into betterand cheaper service for us as
an enterprise.
So, which means that there willbe pressure on the service
providers who are currentlythere to actually go and provide
that kind of service, which isessentially powered
significantly by AI, whichactually opens up opportunities
(16:52):
for, in two fronts one, newcompanies who can now come in
and do that better than existingincumbents, which might be in
India or worldwide, anywhereelse in the world.
It also opens up places whereservices were not, you know,
quite as accepted as and notbeing offered as a service,
(17:13):
right.
So I feel that that particularaspect can actually create
opportunity.
For instance you could thinkabout I'm just kind of like, you
know, doing some blue skythinking here we have legal tech
companies like Harvey, right,which are providing software
(17:34):
solutions for law firms to usewithin their firms.
Right, they use it for allkinds of cases, right, and
workflows within the firm.
But you could think of a firmwhich is, say, an Indian company
which is targeting US companies, offering legal services which
(17:54):
are entirely outsourced, but nowthese legal services are
powered predominantly by AI, butthey also have a legal team
here which is providing that.
So now a whole bunch ofcompanies which previously did
not have access to this level ofhigh quality legal work can now
think about doing those kind ofthings.
(18:16):
So I feel that more areas willopen up where services can be
powered by AI and where therewill be new opportunities there.
So that, I will say, is likeanother thing which kind of
favors the Indian companies,because we understand tech, we
understand services very well interms of just the level of
experience which is there in theworkforce here.
So there could be newopportunities there.
(18:37):
So I think that the nature ofwhat a successful global SaaS
coming in company coming out ofIndia looks like will definitely
change.
It will morph, but the realsize of the opportunity you know
who knows might actuallyincrease.
Speaker 1 (18:53):
Yeah, I agree with that.
The other thing which alsoplays to the advantage is now,
with access to tools,technologies and AI models, even
Indian developers can build aworld-class software, which was
always the question.
That is now, with access totools, technologies and AI
models, even Indian developerscan build a world-class software
, which was always the questionthat the quality of software of
a US company is better than theIndian software.
So maybe there also you'd havea level playing field where the
(19:16):
Indian software is at par within terms of quality and
experience, at par with what aUS company would build or a
European company would build.
Speaker 3 (19:23):
I think it's a fair point that we have a fairly deep
bench of talent on SaaS rightThanks to the early success
stories in India from both Zohoand Freshworks and so forth.
So we have actually folks whohave worked there, have started
companies since then, are in thesecond or third startups, or
folks have just worked there,have started companies since
then, are in the second or thirdstartups, or folks have just
(19:43):
worked there and have deep levelof experience.
So I think that the talent inIndia definitely can compete
with the global talent.
So long as the overallopportunity side is increasing,
I think India will get its fairshare, just that the cost the
traditional levers ofcompetitive advantage might
shift here.
Speaker 1 (20:02):
The other thing, which was, I mean, where a lot
of thinking and opinion was thatIndian companies will typically
or Indian opportunity lies inthe application layer, with
DeepSeek coming up with a modeland showing that you don't need
billions and billions of dollarsto build a world-class model.
Do you think we should look atopportunities or indian founders
(20:24):
should look at building, uh,vertical specific or sector
specific models?
Is there value in buildingmodels?
Speaker 3 (20:31):
so I feel that even if you look at the companies
which have models like, uh, likeopen, ai or a clod, they are
actually not so.
Yes, there there is a bunch ofrevenue which comes from tokens
right, per million tokens orwhatever.
By the way, the price permillion tokens have fallen by
100x in two years.
Right, it's gone from about 60dollars per million tokens to
(20:53):
like 60 cents per million tokensright now.
So it has fallen quite rapidly.
But if you think about it, theway their business models are,
their business models areactually APIs.
Right, you build applicationson top of my thing and of course
, they are going ahead and willbuild a few of those themselves,
but primarily it is that right,so, providing a platform for
others to build on.
And in the case of Google andAmazon, they have models, but
(21:20):
their monetization is really thecloud services.
So it's GCP and AWS and Azurewhich are the monetization
models.
So I feel that pure modelbuilding is going to, in the
limiting case, just becomeequivalent to the cost of the
hardware which is going to runon it, because there are so many
competing models which aregoing to run and, to your point,
(21:40):
if you can actually make afairly sophisticated model at a
lower price point, which reallymeans that more open sources is
going to come as well, becausenow open now think about a
university researcher.
One of their big complaintshave been that hey look, we
actually don't have the money todo this kind of like big bang
research and create new models.
(22:01):
That open ai does now.
Something like deep seek, justdefinitely opens the door,
although you know, curiously,deep seek actually used open as
model to train its models, butnevertheless, a lot of
innovation which has happened inthe process of developing deep
seek, which is obviously goingto be taken forward in many of
these places, which means thatyou're going to see a lot more
open source models.
(22:21):
So point I was getting to isthat what DeepSeek itself means
is that we will have more opensource models coming up, which
are going to be very good theywill never be as good as the
latest open AI model, butthey'll probably just a shade
behind that and which means thatit gets commoditized in terms
(22:41):
of how much you can charge, andthe big boys like Google and
Amazon are also going to keeptheir cost point low in order to
get market share on their side.
So, yes, building models andtrying to sell them is not going
to be the business model, butthose who build models will
probably have other businessmodels.
(23:01):
A business model, but those whobuild models will probably have
other business models, and thatmeans that they have to either
build their own verticalapplication stack on top of the
models they have, or they willhave to have a developer
ecosystem which leverages theirmodels and they can charge on a
per API basis, or what have youcharge on a per API basis, or
what have you In the terms ofapplications that we are likely
(23:22):
to see, you know, as venture?
In most cases, I would thinkthat startups would be
developing vertical apps andhave an abstraction layer which
enables them to, you know,choose the right models for
right things, and it's not goingto be unusual for the
architecture to have severalmodels for specific workflows.
(23:43):
They might use one model forsomething, but if it's marrying
vision, which might be adiffusion model along with some
LLM for text and data analysisand so forth.
So you might actually becombining that and I feel that
open source will end up playinga big part, but I don't think
that just model as a productwill probably be there.
(24:07):
It's already not there.
Speaker 1 (24:08):
Yeah, makes sense.
Now the point is, I mean, withopen source models coming in,
with cost of launching newmodels getting down and what you
alluded to earlier, the tokenpricing has come down right, so
everybody has access to the samesort of models and infra to
build applications.
Yeah, so in theory, everybodycan build the same quality of
(24:31):
applications, uh, whether you'resitting in bangalore or bhopal
or somewhere in europe or us.
Uh, right, uh, so in this world, then, how do you differentiate
your product?
Right, and you'll see more andmore competition, because
instead of like a 20 developerteam, you can do this with a two
developer team, so you'll havemore and more products coming
out trying to solve the same setof problems.
So it's going to be hypercompetitive.
At least you are seeing that.
(24:52):
For example, the customersupport space, ai based customer
support we see like hundreds,like thousands of companies in
that.
How do you builddifferentiation in such a
crowded market?
Like thousands of companies inthat, right, how do you build
differentiation?
Speaker 3 (25:02):
uh, in such a crowded market.
I think that, uh, this willrequire people.
In one sense, I feel that theimportance or, like the nuances
required to build a successfulenterprise software are often
underappreciated, because I feelthat if you, we are like so
plugged into what the latestnews coming out of it and what's
going on in the Valley, and soforth, if you look at a regular
(25:24):
enterprise company which mightbe in oil and gas or even in
financial services or whateverit is, or in a regulated
industry like healthcare, theyare not like hopping from one AI
news to another, to another, toanother right their care abouts
are going to look somethinglike this, which is OK.
What is the security of thisthing?
(25:45):
Are you going to take my dataand do something and put me in
violation of compliance andregulation?
That's a big problem.
No, that's an existential riskto the business.
What is the security?
Where is this data actuallygoing?
How are you actually crunchingit, and so on and so forth.
Third would be okay how muchchange does it require to my
(26:06):
existing workflows?
Because I have all these peoplewho are doing real work.
So, if you're an insurancecompany, you're looking at
claims and so forth.
Yes, you want AI to make thatprocess better and more accurate
and reduce the overhead anderrors.
Better and more accurate andreduce the overhead and errors,
but you don't want to actuallychange that workflow itself
(26:28):
right Now.
If you have to actually goahead and retrain your entire
workforce, it's not going towork right.
So to be successful inenterprise, one has to realize
that AI is a tool.
It's a very powerful tool, butit needs to address a very
specific and nuanced businessproblem for the enterprise,
(26:49):
which means you need tointegrate with their software
and their workflows.
It does not mean that you cango and dump a chat box as your
interface.
I mean that might be great forconsumers because now you're
able to use natural language todo it instead of being
constrained by that one singlesearch box of Google.
Right, but in enterprise, whenyou have dropdown and logins and
(27:13):
this and that and filters,those might actually be quite
fast.
Imagine trying to do that witha prompt and then having three
other prompts after.
It gives you the incompleteanswer.
Instead, I'm fine actuallydoing some dropdowns and
clicking a few things.
So we have to understand in thisparticular case, the solution
(27:33):
providers startups have tounderstand what is the problem
they are trying to solve andwhat makes it easiest for the
enterprise to adapt thissolution, all right.
So I feel that all of thoseaspects will be very important
in order, and that willdifferentiate the solutions.
So AI is part of it, but AI isnot going to be all of it.
(27:55):
The overall will be thisoverall solution, wherein all of
these, the user interfacematters, the integration with
the various software matters,the right workflow understanding
matters, and so in all of thesecases, it is the same old
things which will be required tobuild a business, which is you
need to have deep understandingof the customer behavior.
How are they using software,what is the problem they have
(28:17):
and how can we demonstrateundeniable ROI to them in the
shortest period of time?
So I feel that just taking AIand like plonking it on top of
the customer's heads is notgoing to work and that's
certainly not going to be asuccessful enterprise startup.
Speaker 1 (28:31):
I totally agree.
I mean, the fundamentals ofbuilding a business does not
change with AI, which is who isthe customer, what are the
problem and how am I going tosolve that in a 10x better way
with AI?
So that still remains right.
Of course, we'll use AI to doit in a much better, faster way.
The other thing that I totallyagree is that in all of these
(28:52):
enterprises they'll use a bunchof tools, right.
So AI will solve part of theproblem, but then you'll have to
build the integrations and theworkflow that give you like an
initial differentiation or amode, and, with time, as you get
access to proprietary data,maybe that is what will help you
build the long-term mode for aspecific industry or for a
specific vertical.
Speaker 3 (29:11):
It's a great point.
I think that models are only asgood as the data they are
trained on, right, they cannotcome up with information which
is outside that universe ofinformation that they're trained
on.
They cannot come up withinformation which is outside
that universe of informationthat they're trained on.
And the more relevant thatinformation it is, and less
extraneous information that isthat it is trained on, the
(29:32):
faster and cheaper and moreaccurate that answer is going to
be, and I feel that a lot ofthe data, or enterprise data, is
actually not in the publicdomain.
Obviously, the informationabout my customers and what they
are buying and their behaviorand where they're logging in
from and their demographic etc.
(29:53):
Is particular to me, theenterprise.
The more the software solutionactually relies on non-public
information to actually go anddeliver that value to the
customers actually, the moredifferentiated and sticky it
actually gets.
Speaker 1 (30:13):
Do you think in that case, the incumbents will have
advantage over new players?
Let me just think of Salesforce.
Right, there are tons of dataabout companies, customers,
sitting in the CRM.
Now, if I have to build a salesenablement tool using AI or AI
sales enablement platform,Salesforce has access to the
proprietary data for therespective customer or the
(30:34):
vertical.
You think it'd be easier forthem to build against, say, a
new company trying to do that?
Speaker 3 (30:39):
So I feel that, at least initially I felt is
obviously yes, but now I feelthat maybe not.
And this is the reason, which isthat they clearly have a
distribution of advantage whichis compelling.
So the enterprise customer hasthe Salesforce sitting there and
they'll say like, why do Iactually want to take it out and
put something else in its place?
And besides, I have all thisdata which Salesforce has.
(31:04):
So obviously those models areactually going to be better.
So I feel there are a couple ofthings.
One is that there is this uhadvent of synthetic data, so
models can take some set of realdata and then simulate a whole
lot of synthetic data which justlooks like that data.
It's similar in in many ways,which can make the model itself.
So one model can create thesynthetic data and make your
(31:26):
main model actually a lotsmarter and in that sense, from
a model performance perspective,it can be quite close to what
something which is trained onall volumes of data is.
So that's one point to beunderstood.
So I feel that the dataadvantage is probably not there.
(31:49):
So then it comes to the productside of things, and I just feel
that this is the innovator'sdilemma thing, right, which is a
hungry young startup, will justbe much faster in their
iterations than someone likeSalesforce usually thinking, hey
look, if I actually go and dothis, I have to think about it.
(32:10):
You know, how is it going toimpact my current stuff.
I cannot go ahead and offersomething at a dramatically
lower price.
It's going to cannibalize allmy existing sales.
The sales guys will yell at me,my stock price will drop
because it's going to impact myrevenues, and so forth.
A new startup is like,completely unfettered by any of
these things.
(32:30):
They'll just go and addressthat particular thing.
So that's one thing, one aspect.
The second thing I also thinkabout is that you know the
quality of talent which is goingto be available to startups
versus what is available toSalesforce.
So, in one sense, this is allnew, right?
Ai, you know, if you look atchat, gpt, gpt-3 as the start of
(32:54):
this era, right, so that iswhat November 22, or something,
right, so November 22.
So, essentially, anybody who isactually, you know, using AI or
so so forth, has a veteran, hasa grand total of two years of
experience, right, two and ahalf years of experience.
And, of course, there arepeople who are working on ai
since 2017 18 and know aboutthey're probably the original
(33:15):
researchers in open air and soforth, but from an application
standpoint, right.
So it's not like the pool oftalent that sales has is
actually that much larger and,if you think about it, the most
ambitious and cutting edgetalent would probably want to do
the startup.
So I think that they'll have atalent advantage.
(33:35):
So there is no datadisadvantage, there is a talent
advantage and there's a productadvantage, but they have a
distribution massivedisadvantage.
So I feel that this is going tobe an interesting thing out
there.
But if I were a startup, I wouldnot be looking at something
which goes directly at the heartof something like salesforce.
Right, because a whole host isa system of record, the whole
(33:57):
host of processes you'd like.
You should probably try and dothings which cannot be done by
sales and the Salesforce doesnot have an existing product and
you might want to at that point, talking to our integration,
things actually integrate withSalesforce.
Salesforce has APIs, it has gota marketplace of applications
and so forth, so you actuallyleverage that to start providing
(34:18):
services and I think that willbe an attractive enough market
and then go from there.
So I feel that it will becompetitive with Salesforce.
It doesn't mean the demise ofSalesforce, by any stretch of
imagination.
They're also a fairlyfast-moving company.
But I just feel it opens up alot of opportunities for
startups.
Speaker 1 (34:33):
Yeah, makes sense.
And just for audience, I meanSalesforce we have used here
just as a placeholder incumbent,it could be, Microsoft, it
could be Oracle, it could be SAP.
Yeah, exactly, got it.
Okay, now that brings to thenext question, which is around
pricing.
Right, and I mean pricingmodels are evolving.
Every day we meet a startup,they come with a new pricing
model.
But the core of the questionhere is that with AI, the
(34:57):
promise is that you can do a lotmore with fewer resources,
right, so I'm actually helpingyou reduce the headcount in some
ways.
In most of the use cases, thetraditional software has been
priced on a per user, per seatbasis, at least most of it.
Now, with AI coming, you'll saythat, okay, earlier you were
having like 20 sales reps orlike 100 customer support reps.
(35:20):
Now, with AI coming in, you canbring down a customer support
rep by like 80.
So how do you price it right?
Will you be like undercuttingthe existing offerings?
Are you like analyzing thepricing here?
Like, how should one thinkabout pricing here?
Is the price pool, is the?
Is the profit pool coming downwith ai coming in, or the market
(35:40):
size coming down?
How should one think about it?
Speaker 3 (35:43):
so I feel that the foundation of a good pricing is
that it should be simple.
It should be fair andtransparent to the, to the
customer right and of coursethat should be should make sense
from an roi perspective.
So these things will continueto remain the same.
It has been like that beforeuntil it continues to be here
(36:05):
now.
So while there might be atemptation to come up with a
newfangled pricing model, itshouldn't be so complicated that
the customer is confused aboutwhat's going on.
It needs to be fairly simplefor them to understand.
Like, okay, if I do this, thisis how much I pay.
If I use this much, this is howmuch I pay, and so on and so
forth.
And that's one part.
If I use this much, this is howmuch I pay, and so on and so
(36:25):
forth.
And that's one part.
The second part I would say thatenterprises are used to
thinking in a certain way.
So if I were providing anenterprise software and thinking
about a pricing framework, Iwill try to go and see where the
customer is today and try tomove from there.
Customer is today and try tomove from there.
(36:47):
So if they are actually used toa per seat pricing, then there
is no issue with providing a perseat pricing yourself.
Right, because now it is a lesscognitive load on the customer
to decide how to do this.
Because one of the things whichcustomers would hate is not
knowing how much they'll end upgetting charged.
Right at the end of the theysign up for something, at the
end of the quarter you come upwith a different set of bill.
(37:08):
It's a problem for them in abig way, right, so it needs to
be predictable.
So if you are changing thepricing model and you want to
make it an outcome-based pricing, it has to be very clear and
obvious to the customer aboutwhat their comparable
alternative is today and how itis a better thing from there.
So in some cases it does work.
So, for instance, if we say incustomer support you're talking
(37:29):
about earlier, instead of havinga per person, I would actually
do a per ticket, okay, thatbroadly makes sense, right?
And they will say, hey look,yeah, this I used to spend so
much, I have this much tickets.
I used to spend so much, I havethis much tickets, I used to
spend so much.
Now I can actually go ahead andgive it to you per ticket and
kind of make it seems fair,because, instead of doing it on
a per person basis, you're usingAI, and that's probably right,
(37:50):
but I wouldn't try to shoehornit in all cases.
Right, you have to kind of likesee what the customer is.
My framework here is and try toadapt to that, but ultimately,
it should be very easy for thecustomer to calculate a return
on investment for themselves.
That's what I would say is thesimplest thing out there.
(38:10):
Now there's this question aboutan implicit in your question was
this piece about okay, if I'mreplacing people and there's a
lot of chatter about okay, nowmy ECUs are going to be
significantly higher because I'mpaying somebody.
Let's pick a number, $50,000and you are able to reduce
headcount by one person.
You should, in theory, be ableto charge one dollar less than
(38:31):
fifty thousand dollars, right,but the fallacy there is that,
um, the competitive environmentwill make you not not do that,
right, right, it is.
So I think that that's one.
The second thing, of course,that the incremental value which
you're saving to the customershould also will need to.
There'll be certain expectationof what percentage of that
(38:51):
value addition you're able tocharge.
Right, it'll be maybe 10, 20percent.
So I think that there's a lotof price discovery which is
going to happen and that's whythe rest of the pieces which
make the total solution areimportant.
So the more the total solutionis, you know, looking end-to-end
for the customer, then they'renot just looking for like one
piece and just thinking,comparing you with the other
(39:12):
software, but they're able tolook at the total value at which
you have provided, which mightmake it fairly more
differentiated than the othercompetitors.
The more differentiated it is,the more you can actually
justify having that piece, andthat piece remains the same.
Speaker 1 (39:27):
The other interesting thing which we see could play
out is so far the softwarebudget.
I mean the software expenseused to come from the IT budget.
We do hope that with AIreplacing a lot of humans, some
of that human capital budgetwill also go into this.
Of course not like for like,but at least some part of that
(39:48):
will go here.
So if I'm able to replace theheadcount by, say, 20%, I may be
able to charge, like, say, 5%of that reduced headcount as my
pricing.
Speaker 3 (39:58):
That is one part of it, gaurav.
I would say that that's thecost part of it, but in a lot of
cases, AI is going to driveincremental revenues.
Yes, right, and I feel that theproducts in general which drive
more revenues will have accessto a larger time, larger market.
Right, and you also have abetter pricing, more pricing
flexibility there, and I feelthat there'll be a lot of
(40:18):
opportunity for that.
Speaker 1 (40:19):
Yeah, makes sense.
Finally, what are you excitedabout in this new world of B2B
SaaS being supercharged by AI?
Speaker 3 (40:29):
So I feel that it's a very exciting time for software
.
I think this is one of the mostseminal inventions wherein you
can make.
What we are saying is that, heylook, intelligence is now
available to everybody, and soforth.
But my feeling is that is notavailable to everybody and so
(40:49):
forth.
But my feeling is that you'reseeing intelligence everywhere
available to everybody isprobably a little bit incorrect,
in the sense that what aimodels are is that they take a
huge amount of information, theycompress it and then they give
you an output.
So, if you think about it, ifyou think you know, even if you
think about, like, if you ask it, some piece of information,
right, okay, so what are the?
You know?
Five places I should see inIndia, whatever it is right.
Or five places to visit inParis right, it might give you
(41:12):
five things, but it is not likethose are the five definitive
things.
It's really a mishmash of whatis there in, of what is there in
Wikipedia, blogs, twitter,youtube, all of the social media
put together, and someprobabilistic answer of that.
Right, if you just go a leveldeeper, that is what the model
(41:35):
is saying and, unfortunately,the model is saying it in a very
definitive way.
So you feel like, okay, this isgospel, right, but if you, as
technical folks, would knowthat's what's really going on
underneath, gospel right.
But if you, you know, astechnical folks, would know,
that's what's really going onunderneath it, right.
It's a probabilistic machinewhich is generating these
answers and the probabilitydepends on the data which has
(41:57):
been ingested.
And the data which has beeningested is the net, yes, and it
is not like and, of course, thefolks at OpenAI or wherever
would be clever enough to say,hey look, I should overweight
wikipedia because it's a morevalidated information, also
versus tiktok or or twitter.
But ultimately, that's what'shappening.
So it's not saying that, heylook, you know, everybody will
(42:19):
become intelligent, isn't interms of, like, their deep and
understanding of what's going onis not entirely true, but at
any rate, I think that aside, itdoes make a lot of.
It's a very significantinflection point.
To answer your question, areaswhich we at Prime are excited
about I'm excited about arevertical AI, wherein we are
(42:42):
looking at certain domains.
So you're looking atmanufacturing, you might be
looking at retail, you might belooking at accounting, you might
be looking at security, thesespecific domains in which you're
actually going and creating anapplication, because it
increases the surface area ofadding more value to the
enterprise by doing moreintegration, by looking and
(43:02):
understanding deeply theworkflows, by looking at what is
the domain-specific UI whichneeds to be there or the
domain-specific considerationslike security and compliance.
You talk about manufacturingthings have to be on-prem there,
for instance, and things likethat.
So vertical AI is one big thing.
The second area which we talkedabout, which is as far as the
(43:24):
horizontal pieces, are like moreservices software, which we
talked about, wherein you canthink about delivering services
in an extremely cost-effectiveand very differentiated price
point and with a very highquality.
And a third area which I'mreally hoping that we'll start
seeing some more startups out ofIndia we're definitely seeing
that in other places is physicalAI, and what I mean by that is
(43:48):
automation.
I'm not talking about autonomoussystems.
I probably should differentiatebetween the two, because
autonomous systems are you know,there's a robot which is going
around in the thing serving chaiand then doing this and
cleaning the windows at the sametime as vacuuming the floor and
greeting the visitor.
Okay, that's like autonomous.
You know completely autonomousstuff, right, and there's a lot
(44:09):
of things which make it bothmechanically complicated and the
software of it fairlycomplicated in terms of just the
wide range of things they needto do.
But automation, I think, is adifferent thing and I feel that
we will see a lot of innovationautomation.
So automation is very specific,narrow area in which you are
(44:32):
actually functioning, and theexample you gave earlier about
manufacturing right that isbeing automated.
So the robot that is justtaking something and putting it
there, right, but for instance,just to use that, the vision
systems and the advancement invision, which means that we can
actually dramatically reduce thecost of automation, because
what it takes to train a robotlike that has come down.
(44:55):
How flexible that robot can bein terms of the range of
activities it can do has comedown, which means that you can
provide automation in industrieswhich previously wouldn't have
been possible at all because theprice points were higher.
In one sense, you'redemocratizing automation.
So that is happening and thereason that is possible is that
(45:16):
the models the SLMs, not theLLMs, not the large language
models, but the small languagemodels are now getting more and
more powerful.
Actually, just as DeepSeq is asmaller version of the larger
models, but the small languagemodels are now getting more and
more powerful.
Actually, just as DeepSeq is asmaller version of the larger
models, it's yet very powerful.
Think of even smaller modelswhich will have very
domain-specific things which weare very, very good at and they
are very fast at, and they arefairly cheap and can be embedded
(45:39):
inside a robot in a small formfactor.
So that just opens upopportunities for automation.
So I'm very excited about thatand I feel that we will see, you
know, just new greenfieldopportunities there.
So three things which I'mlooking at, but I'm sure that
more will probably come up.
Speaker 1 (45:56):
Yeah, I'm sure somebody is building an AI VC
while we're discussing software.
Ai will eat software.
Somebody might be building a AIVC will eat traditional VCs
yeah, you're absolutely right.
Speaker 3 (46:08):
Until then, we'll keep our day job.
Speaker 1 (46:12):
Okay, Thank you, Duthi.
Thank you everyone.
I hope you'll enjoy thediscussion that we had.
Do share your comments andfeedback.
Speaker 3 (46:20):
Thanks, gaurav, pleasure talking to you.
Speaker 2 (46:25):
Podcasts in Apple Podcasts, Spotify, CastBox or
however you get your podcasts,Then hit subscribe and if you
(46:47):
have enjoyed the show, we wouldbe really grateful if you leave
us a review on Apple Podcasts.
To read the full transcript,find the link in the show notes.
Zero to a couple of million dollars of ARR, even
tens of millions of dollars ofARR in a couple of months.
In some cases, maybe a year orso.
What do you think has changed?
I think that puts Indiancompanies at some disadvantage.
We can build the same qualityof applications whether you're
sitting in Bangalore or Bhopalor somewhere in Europe or US.
So in this world then, how doyou differentiate your product?
Speaker 3 (00:21):
The nuances required to build a successful enterprise
software are oftenunderappreciated.
Hello everyone, welcome to thisepisode of Prime in Japan.
I'm excited about are.
Hello everyone, Welcome to thisepisode of Prime Venture
Partners podcast.
Speaker 1 (00:51):
In today's episode, shripati and I will be
discussing the implications ofAI in this world of B2B software
.
I mean, every day we are seeingnew things happening with AI
new models being dropped, newbusiness models being cracked,
new companies being formed andscaled from literally zero
million to like a couple ofmillion dollars of ARR in a few
(01:12):
months.
So we'll cover a lot of thatand also how this impacts the
way we think about evaluatingand investing in early stage
startups.
Welcome, shripati, to the showAbsolutely.
Speaker 3 (01:25):
Thank you, Karan.
Speaker 1 (01:26):
So the first thing that we'd like to know from you,
shripati, is that traditionally, we've seen SaaS companies
scaling from inception to amillion ARR in a couple of years
, and then, from there on, theylike triple and triple and
double year on year.
In this new world, we arehearing about a lot of companies
going from literally zero to acouple of million dollars of ARR
(01:48):
, even tens of millions ofdollars of ARR, in a couple of
months, in some cases, maybe ayear or so.
What do you think has changed?
I mean, my assumption is thebuyer.
My understanding is the buyerpersona remains same, the buying
behavior more or less remainssame.
So what has changed?
Because of which companies areable to scale so fast?
Speaker 3 (02:08):
only a couple of comments.
One thing is that remember thatthe we only hear about the
companies which are goingthrough these breathtaking, uh
growth rates.
I'd hear of cursor going fromzero to 100 in like a year right
, stuff like that or two years,whatever the number is and, you
know, competing with some of theother companies which have gone
(02:28):
since the product launch.
So Glean I think the number wasthree years, and this is faster
than Glean is what I recall,200 million ARR.
So, yes, we hear about thefastest of the fast-growing
companies, but we should notconclude from that that that is
the norm for every company,right?
I'll just make that comment.
That said, some things arechanging and I think one of the
(02:51):
most striking features of thenew generation of B2B SaaS
companies is that the size ofthe company is smaller.
That's like number one, and theanswer for that is like a more
simpler answer, which is thatyou require a smaller team to
generate the same quality ofcode, thanks to, you know,
co-pilots for coding and soforth, and also the use of AI in
(03:15):
all the other aspects of aproduct.
Right, you have sales, you havemarketing, you have insights,
sdrs all of these aspects aremade efficient and faster and
better with the use of AI, right, so that's why the companies
are smaller and that's actually,I think, a systemic trend.
The other thing which I feelwhich is different about this
(03:37):
and where the sales cycles aregoing to be faster in many
perhaps not all, but manyenterprise businesses is that
the time to wow, which is thetime it takes for a customer to
look at a product and say that,hey, this is going to actually
really be driving meaningfulvalue to me as a business, as an
(03:59):
increase in top line or asignificant cost saving is lower
, so that gap is just lower.
So if you just take a step backand look at how the cloud
adoption, for instance, happenedin the enterprise, it took a
while for that to happen,because it's not like these
large enterprise let's just take, you know why, wall Street
Enterprise, or even like a largeuniversity or any of those
(04:19):
things which had a deepdeployment of software.
They had on-prem software right, or their own captive data
centers, and then, when that,when cloud came in, they had to,
like, make a decision, likeokay, how much benefit am I
going to get?
Is there a significant changein the way I'm doing things?
What will I do with my existinginfrastructure.
And, of course, over a periodof time, they realized that, you
(04:40):
know, I can scale seamlesslyand it's more OpEx versus CapEx
and so on and so forth.
But in the case of AI, I feelthat the decision makers are
able to see that much moreclearly.
So, for instance, they might beable to say wait a minute, this
particular piece of software isgoing to enable me to respond
(05:01):
to my customer faster, developmore products faster or actually
reduce my headcount in customerservice, which is like our
support which is, you know, inthe news all over the place.
So I feel that that time tovalue and the leap of faith
which the enterprise decisionmaker has to make has come down
(05:23):
significantly, which means thesales cycle are smaller,
companies are smaller, theability to create a product is
faster and the sales cycle aresmaller.
But that doesn't mean that theaverage company would actually
go to 10 million of year one.
But I do think that the totalrevenue ramp for successful
(05:47):
companies is going to be higherthan the pre-AI generation of
companies.
What do you think I mean like?
Does that make sense?
Speaker 1 (05:54):
I mean you covered everything.
One thing that we also see.
I mean we do speak to a lot ofpotential AI customers the way,
like 15, 20 years back, everyenterprise had a digital
transformation budget.
These days, everybody has an AIbudget, so they're open to
trying new things and, as yousaid, like with these AI
(06:14):
solutions, the time to wow islike very fast, you can quickly
see the value, so they have abudget for it.
Once they start investing in it, they see the value very fast
and then, from there on, it'svery easy for them to adopt.
So that's the other factor thatwe have seen, at least in large
enterprises, where they have adedicated budget for AI, which
at least creates a room forexperimentation pilots and from
there on it converts to a realcustomer.
Speaker 3 (06:37):
Yeah, and if you look at, like the history of like
really the various you like,really the various stages or
errors in software, like we hadthe PC software, then we had the
client server computing andthen we had the cloud and mobile
and so forth, in all of thesecases, if you think about the
person on the other side, thiswas not a product which they
(06:59):
were familiar with, right, theywere not using client server
computing in their day-to-daystuff.
They were not using cloud intheir day-to-day uh stuff, or
even if it is there, it was veryabstract.
But I feel also that now everydecision maker, if not them,
their children, are definitelyusing, uh, you know, chat, gpt
or equivalent or gemini, uh andcloud and so forth, right, so
(07:22):
they actually have a firsthandimpression and a firsthand
experience with really the magicof AI, because it's quite
magical, the user experience,and that also helps because now
they're able to correlate, likethis is happening and this demo
is actually delivering value,which I can see.
(07:42):
And perhaps there's the otherthing which is also happening
that competitive dynamic herethey can also see that they're
competitors.
There's so much noise aboutindustries adopting it.
So I feel that all of this isleading to a significant
compression in the cycles.
I think it's a good.
It's an exciting time from thataspect because one of the
biggest issues in enterprisesales as you and I know from our
(08:03):
direct experience investing inthis field is a sales cycle like
.
Speaker 1 (08:07):
It takes time to sell and a time to deploy yeah, the
other thing we're seeing some ofthat impact even in the indian
market.
Traditionally, like indiamarket, has been not very
attractive when it comes to sascompanies.
Like going after india market.
Now, with ai coming in, atleast we're seeing some early
signs where enterprises arewilling to adopt AI solutions,
(08:28):
are willing to pay for that.
Do you think they'll have anychange in the India software
market?
Like with AI coming in, do youthink things will change and the
market will become more?
Speaker 3 (08:37):
interesting.
So India software, indiaenterprise SaaS has always been
a challenging market, in myopinion, from a market size
perspective.
So, honestly, the threecompanies we all keep talking
about which have a billiondollars or more of revenue are,
from an enterprise softwarestandpoint, salesforce, which
has talked about it's more thana billion dollars in India, and
SAP and Oracle, which haven'tdisclosed the India-specific
(08:59):
revenues, but likely quitesignificant.
They are in the same zone.
But if you think of these,these are like one very
foundational pieces of softwareand it's taken a while, with
their heft and brand the globalbrands to actually get to that
level.
So until now we have at leastnot seen a significant total
(09:23):
addressable market forenterprise SaaS in India.
I feel that AI is really goingto open that up.
Now how big and how fast itopens remains to be seen, but
I'm very optimistic about it andthe way I'm thinking about it,
gaurav, is that AI is coming inwith a software solution in
areas where there was actuallyno software at all.
(09:45):
So, for instance, think about aform filling thing.
You're actually filling I don'tknow your compliance audits.
Think of just take anyparticular area like that which
you traditionally think about.
Wait a minute, it's a lot ofpeople, a lot of form, filling,
a lot of, you know, readingregulations, this, that, and
filling and submitting and soforth, and a back and forth
(10:05):
going on with the serviceprovider, with some government
website and this and that andthe other.
Just putting software thereseems like a complete nightmare.
Right Come AI.
Now you can actually visualizeand you're already beginning to
see, at least in other markets,folks actually delivering
software for that, which canlook at all your existing
(10:26):
internal documentation, whichcan correlate with external
documentation, come up with therecommendations and so forth,
and then also help you in theworkflows there.
So I feel that software now isentering areas where previously
the substitute was humans doingpaperwork, humans using
(10:46):
essentially spreadsheets, emailor a word processor like
Microsoft Word or something likethat to do their work Like, if
you ask, even like lawyers,there are two big pieces of
software which they use areMicrosoft Word and email.
Yes, right, so that's what youare really displacing, right, if
you look at the larger legaltech companies that are coming
(11:07):
up, especially in the US.
So I feel that we have thatopportunity now in India, which
is that it's not likeenterprises are not doing
workflows, just that theworkflows are entirely manual.
So now software can actually goand deliver value.
There and in our earlier comment, there are time to wow for that
which is value, realization ismore obvious.
(11:27):
So thing which remains to beseen is that what is the average
order value or average contactacvs of these kind of things?
I think that some discovery hasto happen there.
Obviously it won't be as largeas you know some of some of the
ACVs we are seeing in AI SaaScompanies in the US, but I
believe that you know it's goingto start becoming a significant
(11:49):
market now, precisely for thatreason.
Certainly, that's my belief.
Speaker 1 (11:53):
Yeah, I totally agree with that.
Like new use cases have come upwhich has not previously been
addressed by software.
In fact, one of our portfoliocompanies which is in the QC
space, like computer visionbased QC, yeah, quality control
you mean.
Yeah, quality control Onassembly lines.
You had people looking at stuffcoming out of assembly line and
looking at that and discardingthe bad stuff.
(12:13):
Now with computer vision, ai,you can do that in real time
with the model sitting on theedge and things happen very fast
.
Yeah, so this use case was notpreviously possible with like a
simple software.
Computer vision models werethere around but the models were
not great, you could not deploythat on the edge.
Speaker 3 (12:29):
Now, with lighter models, faster models, you can
do that quite easily yeah, andin one sense, this actually
opens up fairly large and you,you said, right, the AI
transformation budget.
Yes, ai transformation budget.
So this is a conversation whichis not just happening in the
CEO's office, it's happening inthe boardrooms of these
companies, right?
So if you're an investor andthe first thing which we ask all
(12:53):
our portfolio companies whichwe are involved with is, hey,
what is your plan?
And we were asking this back in23 when GPT came out, right,
and this is a conversation whichis in all large enterprise
boardrooms as well which is howare you going to leverage AI?
So this is top of mind for Fools, so it is not something where
(13:13):
you have to, as a startup, goand sell that.
Hey, look, you need to deployAI.
And the question is can youactually have a demonstrable
return on investment,demonstrable ROI for them?
Yes, right, that's the key interms of these things.
So I'm hopeful.
I think that this definitely isa new era.
It definitely is going to be alarger market than our existing
(13:34):
TAM.
Whether it will openmulti-billion dollar TAMs within
India, I don't know, but we'llsee.
I think it will definitely leadto more successful startups in
India, targeting Indiaenterprise.
Speaker 1 (13:46):
Yeah, I agree.
I mean, while we hope that theIndia market opens up for
software, at least with aforcible feature, we do see US
as a large market for companiesbuilding out of India.
And so far, one advantage thatwe had was the cost advantage.
Whether it was in terms ofdevelopment, testing, sales
support, all of that India hadan inherent cost advantage.
Whether it was in terms ofdevelopment, testing, sales
support, all of that india hadan inherent cost advantage
(14:07):
compared to a us native company.
Now, with ai coming insomething that we were talking
about earlier the size of theteams have come down.
Ai will do heavy lifting interms of sales and support, with
ai agents taking over humanagents or like working together
human agents.
Do you think that puts Indiancompanies at some disadvantage
or takes away the advantage thatthey had before when it comes
(14:29):
to the cost structures andthey'll have to compete at par
with a US company which willhave a similar cost structure as
an Indian company?
Speaker 3 (14:37):
So, as the, I mean that's definitely is a very
significant aspect which is atplay here in the sense that the
coding teams are smaller and thesales force is smaller, because
now the outbound calling withall the agent, API software and
so forth, the support sales opsoperation, for example, is
(15:00):
smaller and customer supportteams are smaller and so forth.
So if the total size of thecompany has shrunk, then
definitely the arbitrage youmight have from a development
team in India versus adevelopment team in the US
becomes definitely significantlysmaller.
I think that aspect you'reright, Gaurav significantly
smaller.
(15:20):
I think that aspect you'reright, Gaurav.
I am also seeing, though, thatthere is a potentially
countervailing force here, whichis that a lot of a new
opportunity we believe onenterprise SaaS is service as a
software or outcome as a service, or however you want to call it
, wherein the end customer isgetting service and it's just
(15:44):
like there might even be a humaninterface.
Think about it as a contactcenter or a customer support,
where you're actually talking toa customer agent and you're a
medium or a large enterprise,and I feel that a lot of the
non-tech companies will not wantto in-house this.
So, for instance, you knowthere's a lot of talk about
clarna.
Companies will not want toin-house this.
So, for instance, there's a lotof talk about Klarna, which is
this fintech wherein they wentahead and essentially reduced
(16:08):
their customer support by 90% ormore and have the same NPS,
same customer side by goingentirely into an agent tech bot
which is actually doing bothvoice and text support.
Right, but if you think about aregular enterprise, they both
voice and text support right,but if you think about a regular
enterprise, they're not goingto go ahead and develop that
software in-house first of all.
So the transition we see inthose cases is those enterprises
(16:30):
telling their service providersthat hey look, we want you to
use AI on your thing.
We should translate into betterand cheaper service for us as
an enterprise.
So, which means that there willbe pressure on the service
providers who are currentlythere to actually go and provide
that kind of service, which isessentially powered
significantly by AI, whichactually opens up opportunities
(16:52):
for, in two fronts one, newcompanies who can now come in
and do that better than existingincumbents, which might be in
India or worldwide, anywhereelse in the world.
It also opens up places whereservices were not, you know,
quite as accepted as and notbeing offered as a service,
(17:13):
right.
So I feel that that particularaspect can actually create
opportunity.
For instance you could thinkabout I'm just kind of like, you
know, doing some blue skythinking here we have legal tech
companies like Harvey, right,which are providing software
(17:34):
solutions for law firms to usewithin their firms.
Right, they use it for allkinds of cases, right, and
workflows within the firm.
But you could think of a firmwhich is, say, an Indian company
which is targeting US companies, offering legal services which
(17:54):
are entirely outsourced, but nowthese legal services are
powered predominantly by AI, butthey also have a legal team
here which is providing that.
So now a whole bunch ofcompanies which previously did
not have access to this level ofhigh quality legal work can now
think about doing those kind ofthings.
(18:16):
So I feel that more areas willopen up where services can be
powered by AI and where therewill be new opportunities there.
So that, I will say, is likeanother thing which kind of
favors the Indian companies,because we understand tech, we
understand services very well interms of just the level of
experience which is there in theworkforce here.
So there could be newopportunities there.
(18:37):
So I think that the nature ofwhat a successful global SaaS
coming in company coming out ofIndia looks like will definitely
change.
It will morph, but the realsize of the opportunity you know
who knows might actuallyincrease.
Speaker 1 (18:53):
Yeah, I agree with that.
The other thing which alsoplays to the advantage is now,
with access to tools,technologies and AI models, even
Indian developers can build aworld-class software, which was
always the question.
That is now, with access totools, technologies and AI
models, even Indian developerscan build a world-class software
, which was always the questionthat the quality of software of
a US company is better than theIndian software.
So maybe there also you'd havea level playing field where the
(19:16):
Indian software is at par within terms of quality and
experience, at par with what aUS company would build or a
European company would build.
Speaker 3 (19:23):
I think it's a fair point that we have a fairly deep
bench of talent on SaaS rightThanks to the early success
stories in India from both Zohoand Freshworks and so forth.
So we have actually folks whohave worked there, have started
companies since then, are in thesecond or third startups, or
folks have just worked there,have started companies since
then, are in the second or thirdstartups, or folks have just
(19:43):
worked there and have deep levelof experience.
So I think that the talent inIndia definitely can compete
with the global talent.
So long as the overallopportunity side is increasing,
I think India will get its fairshare, just that the cost the
traditional levers ofcompetitive advantage might
shift here.
Speaker 1 (20:02):
The other thing, which was, I mean, where a lot
of thinking and opinion was thatIndian companies will typically
or Indian opportunity lies inthe application layer, with
DeepSeek coming up with a modeland showing that you don't need
billions and billions of dollarsto build a world-class model.
Do you think we should look atopportunities or indian founders
(20:24):
should look at building, uh,vertical specific or sector
specific models?
Is there value in buildingmodels?
Speaker 3 (20:31):
so I feel that even if you look at the companies
which have models like, uh, likeopen, ai or a clod, they are
actually not so.
Yes, there there is a bunch ofrevenue which comes from tokens
right, per million tokens orwhatever.
By the way, the price permillion tokens have fallen by
100x in two years.
Right, it's gone from about 60dollars per million tokens to
(20:53):
like 60 cents per million tokensright now.
So it has fallen quite rapidly.
But if you think about it, theway their business models are,
their business models areactually APIs.
Right, you build applicationson top of my thing and of course
, they are going ahead and willbuild a few of those themselves,
but primarily it is that right,so, providing a platform for
others to build on.
And in the case of Google andAmazon, they have models, but
(21:20):
their monetization is really thecloud services.
So it's GCP and AWS and Azurewhich are the monetization
models.
So I feel that pure modelbuilding is going to, in the
limiting case, just becomeequivalent to the cost of the
hardware which is going to runon it, because there are so many
competing models which aregoing to run and, to your point,
(21:40):
if you can actually make afairly sophisticated model at a
lower price point, which reallymeans that more open sources is
going to come as well, becausenow open now think about a
university researcher.
One of their big complaintshave been that hey look, we
actually don't have the money todo this kind of like big bang
research and create new models.
(22:01):
That open ai does now.
Something like deep seek, justdefinitely opens the door,
although you know, curiously,deep seek actually used open as
model to train its models, butnevertheless, a lot of
innovation which has happened inthe process of developing deep
seek, which is obviously goingto be taken forward in many of
these places, which means thatyou're going to see a lot more
open source models.
(22:21):
So point I was getting to isthat what DeepSeek itself means
is that we will have more opensource models coming up, which
are going to be very good theywill never be as good as the
latest open AI model, butthey'll probably just a shade
behind that and which means thatit gets commoditized in terms
(22:41):
of how much you can charge, andthe big boys like Google and
Amazon are also going to keeptheir cost point low in order to
get market share on their side.
So, yes, building models andtrying to sell them is not going
to be the business model, butthose who build models will
probably have other businessmodels.
(23:01):
A business model, but those whobuild models will probably have
other business models, and thatmeans that they have to either
build their own verticalapplication stack on top of the
models they have, or they willhave to have a developer
ecosystem which leverages theirmodels and they can charge on a
per API basis, or what have youcharge on a per API basis, or
what have you In the terms ofapplications that we are likely
(23:22):
to see, you know, as venture?
In most cases, I would thinkthat startups would be
developing vertical apps andhave an abstraction layer which
enables them to, you know,choose the right models for
right things, and it's not goingto be unusual for the
architecture to have severalmodels for specific workflows.
(23:43):
They might use one model forsomething, but if it's marrying
vision, which might be adiffusion model along with some
LLM for text and data analysisand so forth.
So you might actually becombining that and I feel that
open source will end up playinga big part, but I don't think
that just model as a productwill probably be there.
(24:07):
It's already not there.
Speaker 1 (24:08):
Yeah, makes sense.
Now the point is, I mean, withopen source models coming in,
with cost of launching newmodels getting down and what you
alluded to earlier, the tokenpricing has come down right, so
everybody has access to the samesort of models and infra to
build applications.
Yeah, so in theory, everybodycan build the same quality of
(24:31):
applications, uh, whether you'resitting in bangalore or bhopal
or somewhere in europe or us.
Uh, right, uh, so in this world, then, how do you differentiate
your product?
Right, and you'll see more andmore competition, because
instead of like a 20 developerteam, you can do this with a two
developer team, so you'll havemore and more products coming
out trying to solve the same setof problems.
So it's going to be hypercompetitive.
At least you are seeing that.
(24:52):
For example, the customersupport space, ai based customer
support we see like hundreds,like thousands of companies in
that.
How do you builddifferentiation in such a
crowded market?
Like thousands of companies inthat, right, how do you build
differentiation?
Speaker 3 (25:02):
uh, in such a crowded market.
I think that, uh, this willrequire people.
In one sense, I feel that theimportance or, like the nuances
required to build a successfulenterprise software are often
underappreciated, because I feelthat if you, we are like so
plugged into what the latestnews coming out of it and what's
going on in the Valley, and soforth, if you look at a regular
(25:24):
enterprise company which mightbe in oil and gas or even in
financial services or whateverit is, or in a regulated
industry like healthcare, theyare not like hopping from one AI
news to another, to another, toanother right their care abouts
are going to look somethinglike this, which is OK.
What is the security of thisthing?
(25:45):
Are you going to take my dataand do something and put me in
violation of compliance andregulation?
That's a big problem.
No, that's an existential riskto the business.
What is the security?
Where is this data actuallygoing?
How are you actually crunchingit, and so on and so forth.
Third would be okay how muchchange does it require to my
(26:06):
existing workflows?
Because I have all these peoplewho are doing real work.
So, if you're an insurancecompany, you're looking at
claims and so forth.
Yes, you want AI to make thatprocess better and more accurate
and reduce the overhead anderrors.
Better and more accurate andreduce the overhead and errors,
but you don't want to actuallychange that workflow itself
(26:28):
right Now.
If you have to actually goahead and retrain your entire
workforce, it's not going towork right.
So to be successful inenterprise, one has to realize
that AI is a tool.
It's a very powerful tool, butit needs to address a very
specific and nuanced businessproblem for the enterprise,
(26:49):
which means you need tointegrate with their software
and their workflows.
It does not mean that you cango and dump a chat box as your
interface.
I mean that might be great forconsumers because now you're
able to use natural language todo it instead of being
constrained by that one singlesearch box of Google.
Right, but in enterprise, whenyou have dropdown and logins and
(27:13):
this and that and filters,those might actually be quite
fast.
Imagine trying to do that witha prompt and then having three
other prompts after.
It gives you the incompleteanswer.
Instead, I'm fine actuallydoing some dropdowns and
clicking a few things.
So we have to understand in thisparticular case, the solution
(27:33):
providers startups have tounderstand what is the problem
they are trying to solve andwhat makes it easiest for the
enterprise to adapt thissolution, all right.
So I feel that all of thoseaspects will be very important
in order, and that willdifferentiate the solutions.
So AI is part of it, but AI isnot going to be all of it.
(27:55):
The overall will be thisoverall solution, wherein all of
these, the user interfacematters, the integration with
the various software matters,the right workflow understanding
matters, and so in all of thesecases, it is the same old
things which will be required tobuild a business, which is you
need to have deep understandingof the customer behavior.
How are they using software,what is the problem they have
(28:17):
and how can we demonstrateundeniable ROI to them in the
shortest period of time?
So I feel that just taking AIand like plonking it on top of
the customer's heads is notgoing to work and that's
certainly not going to be asuccessful enterprise startup.
Speaker 1 (28:31):
I totally agree.
I mean, the fundamentals ofbuilding a business does not
change with AI, which is who isthe customer, what are the
problem and how am I going tosolve that in a 10x better way
with AI?
So that still remains right.
Of course, we'll use AI to doit in a much better, faster way.
The other thing that I totallyagree is that in all of these
(28:52):
enterprises they'll use a bunchof tools, right.
So AI will solve part of theproblem, but then you'll have to
build the integrations and theworkflow that give you like an
initial differentiation or amode, and, with time, as you get
access to proprietary data,maybe that is what will help you
build the long-term mode for aspecific industry or for a
specific vertical.
Speaker 3 (29:11):
It's a great point.
I think that models are only asgood as the data they are
trained on, right, they cannotcome up with information which
is outside that universe ofinformation that they're trained
on.
They cannot come up withinformation which is outside
that universe of informationthat they're trained on.
And the more relevant thatinformation it is, and less
extraneous information that isthat it is trained on, the
(29:32):
faster and cheaper and moreaccurate that answer is going to
be, and I feel that a lot ofthe data, or enterprise data, is
actually not in the publicdomain.
Obviously, the informationabout my customers and what they
are buying and their behaviorand where they're logging in
from and their demographic etc.
(29:53):
Is particular to me, theenterprise.
The more the software solutionactually relies on non-public
information to actually go anddeliver that value to the
customers actually, the moredifferentiated and sticky it
actually gets.
Speaker 1 (30:13):
Do you think in that case, the incumbents will have
advantage over new players?
Let me just think of Salesforce.
Right, there are tons of dataabout companies, customers,
sitting in the CRM.
Now, if I have to build a salesenablement tool using AI or AI
sales enablement platform,Salesforce has access to the
proprietary data for therespective customer or the
(30:34):
vertical.
You think it'd be easier forthem to build against, say, a
new company trying to do that?
Speaker 3 (30:39):
So I feel that, at least initially I felt is
obviously yes, but now I feelthat maybe not.
And this is the reason, which isthat they clearly have a
distribution of advantage whichis compelling.
So the enterprise customer hasthe Salesforce sitting there and
they'll say like, why do Iactually want to take it out and
put something else in its place?
And besides, I have all thisdata which Salesforce has.
(31:04):
So obviously those models areactually going to be better.
So I feel there are a couple ofthings.
One is that there is this uhadvent of synthetic data, so
models can take some set of realdata and then simulate a whole
lot of synthetic data which justlooks like that data.
It's similar in in many ways,which can make the model itself.
So one model can create thesynthetic data and make your
(31:26):
main model actually a lotsmarter and in that sense, from
a model performance perspective,it can be quite close to what
something which is trained onall volumes of data is.
So that's one point to beunderstood.
So I feel that the dataadvantage is probably not there.
(31:49):
So then it comes to the productside of things, and I just feel
that this is the innovator'sdilemma thing, right, which is a
hungry young startup, will justbe much faster in their
iterations than someone likeSalesforce usually thinking, hey
look, if I actually go and dothis, I have to think about it.
(32:10):
You know, how is it going toimpact my current stuff.
I cannot go ahead and offersomething at a dramatically
lower price.
It's going to cannibalize allmy existing sales.
The sales guys will yell at me,my stock price will drop
because it's going to impact myrevenues, and so forth.
A new startup is like,completely unfettered by any of
these things.
(32:30):
They'll just go and addressthat particular thing.
So that's one thing, one aspect.
The second thing I also thinkabout is that you know the
quality of talent which is goingto be available to startups
versus what is available toSalesforce.
So, in one sense, this is allnew, right?
Ai, you know, if you look atchat, gpt, gpt-3 as the start of
(32:54):
this era, right, so that iswhat November 22, or something,
right, so November 22.
So, essentially, anybody who isactually, you know, using AI or
so so forth, has a veteran, hasa grand total of two years of
experience, right, two and ahalf years of experience.
And, of course, there arepeople who are working on ai
since 2017 18 and know aboutthey're probably the original
(33:15):
researchers in open air and soforth, but from an application
standpoint, right.
So it's not like the pool oftalent that sales has is
actually that much larger and,if you think about it, the most
ambitious and cutting edgetalent would probably want to do
the startup.
So I think that they'll have atalent advantage.
(33:35):
So there is no datadisadvantage, there is a talent
advantage and there's a productadvantage, but they have a
distribution massivedisadvantage.
So I feel that this is going tobe an interesting thing out
there.
But if I were a startup, I wouldnot be looking at something
which goes directly at the heartof something like salesforce.
Right, because a whole host isa system of record, the whole
(33:57):
host of processes you'd like.
You should probably try and dothings which cannot be done by
sales and the Salesforce doesnot have an existing product and
you might want to at that point, talking to our integration,
things actually integrate withSalesforce.
Salesforce has APIs, it has gota marketplace of applications
and so forth, so you actuallyleverage that to start providing
(34:18):
services and I think that willbe an attractive enough market
and then go from there.
So I feel that it will becompetitive with Salesforce.
It doesn't mean the demise ofSalesforce, by any stretch of
imagination.
They're also a fairlyfast-moving company.
But I just feel it opens up alot of opportunities for
startups.
Speaker 1 (34:33):
Yeah, makes sense.
And just for audience, I meanSalesforce we have used here
just as a placeholder incumbent,it could be, Microsoft, it
could be Oracle, it could be SAP.
Yeah, exactly, got it.
Okay, now that brings to thenext question, which is around
pricing.
Right, and I mean pricingmodels are evolving.
Every day we meet a startup,they come with a new pricing
model.
But the core of the questionhere is that with AI, the
(34:57):
promise is that you can do a lotmore with fewer resources,
right, so I'm actually helpingyou reduce the headcount in some
ways.
In most of the use cases, thetraditional software has been
priced on a per user, per seatbasis, at least most of it.
Now, with AI coming, you'll saythat, okay, earlier you were
having like 20 sales reps orlike 100 customer support reps.
(35:20):
Now, with AI coming in, you canbring down a customer support
rep by like 80.
So how do you price it right?
Will you be like undercuttingthe existing offerings?
Are you like analyzing thepricing here?
Like, how should one thinkabout pricing here?
Is the price pool, is the?
Is the profit pool coming downwith ai coming in, or the market
(35:40):
size coming down?
How should one think about it?
Speaker 3 (35:43):
so I feel that the foundation of a good pricing is
that it should be simple.
It should be fair andtransparent to the, to the
customer right and of coursethat should be should make sense
from an roi perspective.
So these things will continueto remain the same.
It has been like that beforeuntil it continues to be here
(36:05):
now.
So while there might be atemptation to come up with a
newfangled pricing model, itshouldn't be so complicated that
the customer is confused aboutwhat's going on.
It needs to be fairly simplefor them to understand.
Like, okay, if I do this, thisis how much I pay.
If I use this much, this is howmuch I pay, and so on and so
forth.
And that's one part.
If I use this much, this is howmuch I pay, and so on and so
(36:25):
forth.
And that's one part.
The second part I would say thatenterprises are used to
thinking in a certain way.
So if I were providing anenterprise software and thinking
about a pricing framework, Iwill try to go and see where the
customer is today and try tomove from there.
Customer is today and try tomove from there.
(36:47):
So if they are actually used toa per seat pricing, then there
is no issue with providing a perseat pricing yourself.
Right, because now it is a lesscognitive load on the customer
to decide how to do this.
Because one of the things whichcustomers would hate is not
knowing how much they'll end upgetting charged.
Right at the end of the theysign up for something, at the
end of the quarter you come upwith a different set of bill.
(37:08):
It's a problem for them in abig way, right, so it needs to
be predictable.
So if you are changing thepricing model and you want to
make it an outcome-based pricing, it has to be very clear and
obvious to the customer aboutwhat their comparable
alternative is today and how itis a better thing from there.
So in some cases it does work.
So, for instance, if we say incustomer support you're talking
(37:29):
about earlier, instead of havinga per person, I would actually
do a per ticket, okay, thatbroadly makes sense, right?
And they will say, hey look,yeah, this I used to spend so
much, I have this much tickets.
I used to spend so much, I havethis much tickets, I used to
spend so much.
Now I can actually go ahead andgive it to you per ticket and
kind of make it seems fair,because, instead of doing it on
a per person basis, you're usingAI, and that's probably right,
(37:50):
but I wouldn't try to shoehornit in all cases.
Right, you have to kind of likesee what the customer is.
My framework here is and try toadapt to that, but ultimately,
it should be very easy for thecustomer to calculate a return
on investment for themselves.
That's what I would say is thesimplest thing out there.
(38:10):
Now there's this question aboutan implicit in your question was
this piece about okay, if I'mreplacing people and there's a
lot of chatter about okay, nowmy ECUs are going to be
significantly higher because I'mpaying somebody.
Let's pick a number, $50,000and you are able to reduce
headcount by one person.
You should, in theory, be ableto charge one dollar less than
(38:31):
fifty thousand dollars, right,but the fallacy there is that,
um, the competitive environmentwill make you not not do that,
right, right, it is.
So I think that that's one.
The second thing, of course,that the incremental value which
you're saving to the customershould also will need to.
There'll be certain expectationof what percentage of that
(38:51):
value addition you're able tocharge.
Right, it'll be maybe 10, 20percent.
So I think that there's a lotof price discovery which is
going to happen and that's whythe rest of the pieces which
make the total solution areimportant.
So the more the total solutionis, you know, looking end-to-end
for the customer, then they'renot just looking for like one
piece and just thinking,comparing you with the other
(39:12):
software, but they're able tolook at the total value at which
you have provided, which mightmake it fairly more
differentiated than the othercompetitors.
The more differentiated it is,the more you can actually
justify having that piece, andthat piece remains the same.
Speaker 1 (39:27):
The other interesting thing which we see could play
out is so far the softwarebudget.
I mean the software expenseused to come from the IT budget.
We do hope that with AIreplacing a lot of humans, some
of that human capital budgetwill also go into this.
Of course not like for like,but at least some part of that
(39:48):
will go here.
So if I'm able to replace theheadcount by, say, 20%, I may be
able to charge, like, say, 5%of that reduced headcount as my
pricing.
Speaker 3 (39:58):
That is one part of it, gaurav.
I would say that that's thecost part of it, but in a lot of
cases, AI is going to driveincremental revenues.
Yes, right, and I feel that theproducts in general which drive
more revenues will have accessto a larger time, larger market.
Right, and you also have abetter pricing, more pricing
flexibility there, and I feelthat there'll be a lot of
(40:18):
opportunity for that.
Speaker 1 (40:19):
Yeah, makes sense.
Finally, what are you excitedabout in this new world of B2B
SaaS being supercharged by AI?
Speaker 3 (40:29):
So I feel that it's a very exciting time for software
.
I think this is one of the mostseminal inventions wherein you
can make.
What we are saying is that, heylook, intelligence is now
available to everybody, and soforth.
But my feeling is that is notavailable to everybody and so
(40:49):
forth.
But my feeling is that you'reseeing intelligence everywhere
available to everybody isprobably a little bit incorrect,
in the sense that what aimodels are is that they take a
huge amount of information, theycompress it and then they give
you an output.
So, if you think about it, ifyou think you know, even if you
think about, like, if you ask it, some piece of information,
right, okay, so what are the?
You know?
Five places I should see inIndia, whatever it is right.
Or five places to visit inParis right, it might give you
(41:12):
five things, but it is not likethose are the five definitive
things.
It's really a mishmash of whatis there in, of what is there in
Wikipedia, blogs, twitter,youtube, all of the social media
put together, and someprobabilistic answer of that.
Right, if you just go a leveldeeper, that is what the model
(41:35):
is saying and, unfortunately,the model is saying it in a very
definitive way.
So you feel like, okay, this isgospel, right, but if you, as
technical folks, would knowthat's what's really going on
underneath, gospel right.
But if you, you know, astechnical folks, would know,
that's what's really going onunderneath it, right.
It's a probabilistic machinewhich is generating these
answers and the probabilitydepends on the data which has
(41:57):
been ingested.
And the data which has beeningested is the net, yes, and it
is not like and, of course, thefolks at OpenAI or wherever
would be clever enough to say,hey look, I should overweight
wikipedia because it's a morevalidated information, also
versus tiktok or or twitter.
But ultimately, that's what'shappening.
So it's not saying that, heylook, you know, everybody will
(42:19):
become intelligent, isn't interms of, like, their deep and
understanding of what's going onis not entirely true, but at
any rate, I think that aside, itdoes make a lot of.
It's a very significantinflection point.
To answer your question, areaswhich we at Prime are excited
about I'm excited about arevertical AI, wherein we are
(42:42):
looking at certain domains.
So you're looking atmanufacturing, you might be
looking at retail, you might belooking at accounting, you might
be looking at security, thesespecific domains in which you're
actually going and creating anapplication, because it
increases the surface area ofadding more value to the
enterprise by doing moreintegration, by looking and
(43:02):
understanding deeply theworkflows, by looking at what is
the domain-specific UI whichneeds to be there or the
domain-specific considerationslike security and compliance.
You talk about manufacturingthings have to be on-prem there,
for instance, and things likethat.
So vertical AI is one big thing.
The second area which we talkedabout, which is as far as the
(43:24):
horizontal pieces, are like moreservices software, which we
talked about, wherein you canthink about delivering services
in an extremely cost-effectiveand very differentiated price
point and with a very highquality.
And a third area which I'mreally hoping that we'll start
seeing some more startups out ofIndia we're definitely seeing
that in other places is physicalAI, and what I mean by that is
(43:48):
automation.
I'm not talking about autonomoussystems.
I probably should differentiatebetween the two, because
autonomous systems are you know,there's a robot which is going
around in the thing serving chaiand then doing this and
cleaning the windows at the sametime as vacuuming the floor and
greeting the visitor.
Okay, that's like autonomous.
You know completely autonomousstuff, right, and there's a lot
(44:09):
of things which make it bothmechanically complicated and the
software of it fairlycomplicated in terms of just the
wide range of things they needto do.
But automation, I think, is adifferent thing and I feel that
we will see a lot of innovationautomation.
So automation is very specific,narrow area in which you are
(44:32):
actually functioning, and theexample you gave earlier about
manufacturing right that isbeing automated.
So the robot that is justtaking something and putting it
there, right, but for instance,just to use that, the vision
systems and the advancement invision, which means that we can
actually dramatically reduce thecost of automation, because
what it takes to train a robotlike that has come down.
(44:55):
How flexible that robot can bein terms of the range of
activities it can do has comedown, which means that you can
provide automation in industrieswhich previously wouldn't have
been possible at all because theprice points were higher.
In one sense, you'redemocratizing automation.
So that is happening and thereason that is possible is that
(45:16):
the models the SLMs, not theLLMs, not the large language
models, but the small languagemodels are now getting more and
more powerful.
Actually, just as DeepSeq is asmaller version of the larger
models, but the small languagemodels are now getting more and
more powerful.
Actually, just as DeepSeq is asmaller version of the larger
models, it's yet very powerful.
Think of even smaller modelswhich will have very
domain-specific things which weare very, very good at and they
are very fast at, and they arefairly cheap and can be embedded
(45:39):
inside a robot in a small formfactor.
So that just opens upopportunities for automation.
So I'm very excited about thatand I feel that we will see, you
know, just new greenfieldopportunities there.
So three things which I'mlooking at, but I'm sure that
more will probably come up.
Speaker 1 (45:56):
Yeah, I'm sure somebody is building an AI VC
while we're discussing software.
Ai will eat software.
Somebody might be building a AIVC will eat traditional VCs
yeah, you're absolutely right.
Speaker 3 (46:08):
Until then, we'll keep our day job.
Speaker 1 (46:12):
Okay, Thank you, Duthi.
Thank you everyone.
I hope you'll enjoy thediscussion that we had.
Do share your comments andfeedback.
Speaker 3 (46:20):
Thanks, gaurav, pleasure talking to you.
Speaker 2 (46:25):
Podcasts in Apple Podcasts, Spotify, CastBox or
however you get your podcasts,Then hit subscribe and if you
(46:47):
have enjoyed the show, we wouldbe really grateful if you leave
us a review on Apple Podcasts.
To read the full transcript,find the link in the show notes.
Popular Podcasts
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.