October 26, 2022 • 29 mins
In this episode, Vin Vashishta (AI Strategy Leader) and Andreas Welsch discuss how you can define an AI strategy and put it into action. Vin shares examples on business and technology goals, and provides valuable insights for listeners looking to define their own AI strategy.
Key topics:
- Define the key aspects of an AI strategy
- Measure your AI strategy's effectiveness
- Describe which business aspects are influenced by an AI strategy
Listen to the full episode to hear how you can:
- Make your AI strategy actionable
- Lead with business impact over technology
- Create enormous amounts of value for the business
Watch this episode on YouTube: https://youtu.be/EdRJ_gh53N0
Questions or suggestions? Send me a Text Message.
***********
Disclaimer: Views are the participants’ own and do not represent those of any participant’s past, present, or future employers. Participation in this event is independent of any potential business relationship (past, present, or future) between the participants or between their employers.
Level up your AI Leadership game with the AI Leadership Handbook:
https://www.aileadershiphandbook.com
More details:
https://www.intelligence-briefing.com
All episodes:
https://www.intelligence-briefing.com/podcast
Get a weekly thought-provoking post in your inbox:
https://www.intelligence-briefing.com/newsletter
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Andreas Welsch (00:00):
Today we'll talk about making your AI strategy
actionable.
And who better to talk to aboutit than someone who helps
leaders do just that.
Hey, how are you?
Vin Vashishta (00:12):
I'm good.
Thanks for having me on.
How are you doing today?
Andreas Welsch (00:15):
Doing all right.
Hey, why don't you tell us alittle bit about yourself, who
you are and what you do.
Vin Vashishta (00:20):
Sure.
Vin Vashishta, I've been intechnology for over 25 years.
Been working in the data sciencemachine learning AI space for
the last 10, doing strategywork, AI strategy, data
strategy, data, product strategyfor just about eight years now
because as I started V squaredin 2012.
(00:40):
And almost immediately realizedI could not do any interesting
projects unless I got C-levelbuy-in unless the ROI was there.
And I began to see all of theproblems that were stacked up
behind asking for buy-in.
And that's where I got involvedin strategy because you can't do
this without holistic strategy,which is what we're talking
(01:02):
about today.
So I've helped clients, Fortune100 clients, SMEs, startups.
It's been interesting over thelast five years how interested
startups and SMEs are suddenlyin doing this in a more rigorous
way and doing this in a way thatreturns value to the business
instead of this exploratorypilot way.
So that's my story.
(01:23):
The only funny part about it wasI tried to get into AI in the
early nine or early midnineties.
When that first boom happenedand no one wanted to hire me, I
went to college for machinelearning.
I think I was gonna graduate andgo straight into Microsoft.
Yeah, I had to wait almost 15years for it to come around.
Andreas Welsch (01:44):
See, that's what I like about the industry.
There are lots of secondchances.
Awesome.
I've, been following you onLinkedIn for quite some time and
I'm sure many of of you in theaudience are familiar with you
and your content as well.
So I'm really, looking forwardto learning more about defining
an AI strategy from you.
And, to those of you in theaudience, if you're just joining
(02:06):
the stream, drop a comment inthe chat.
What do you think an AI strategyfor your business needs to
answer?
And if you have any otherquestions, please do put them in
the chat as well.
I want to make sure that we haveenough time to answer them and
keep this interactive.
Vin Vashishta (02:21):
Yeah, take advantage of the free time if
you got a question.
Definitely throw it into thechat because this is a good
opportunity.
Andreas Welsch (02:28):
Exactly.
Doesn't get better than.
At least not cheaper.
Vin Vashishta (02:33):
I promise it'll be worth every penny.
Andreas Welsch (02:37):
I'm, really curious.
I, posted a poll on, LinkedInearlier this week, and I would
love to get your perspective.
So I asked the community howshould organizations start their
AI journey, start with strategyor start with projects.
What do you say?
Vin Vashishta (02:53):
I think you should start with opportunity
discovery.
It's the same place that youshould start any sort of
strategy planning is, andinstead of saying the technology
enables these opportunities, sowe should do them I take a more
traditional strategic approach,which is, here are the
opportunities that are presentedto the business.
Here are the opportunities thebusiness is best suited to go
(03:16):
after.
Just from a basic core strategystandpoint.
Not even thinking about thetechnology at this.
And then from there we have yourbusiness strategy and then
business model, operating model.
And you should have a thirdpiece, which is the technology
model.
And that explains why thebusiness uses technology to go
after the opportunities thatit's decided on, that it's
(03:37):
discovered, that it's gonethrough the process of saying,
this is what we should be doing,not what we could be.
And when you do that, you havestrategy and value driving
technology versus technologysaying, this is possible, so why
don't we do it?
Yes.
That's the big mistake upfrontthat I think most, clients that
(03:59):
I get brought in to help, mostbusinesses that I see struggling
with data science and machinelearning, just because it's
possible, they think it'sprobably something that should
be done, and there's a lot ofFOMO.
There's a lot of what if thecompetition gets there first?
It's way more important to saythese are the best opportunities
for the business.
Technology agnostic.
(04:19):
How do we use data, analytics,AI to be better at capitalizing
on those opportunities.
And if you use the discoveryframework, you end up with a
better AI strategy because it'sthrough to the technology model.
It's tethered to your businessmodel, your operating model, and
(04:40):
core strategy.
It isn't trying to change corestrategy or drive core strategy,
it's just an extension.
Andreas Welsch (04:48):
The point that really resonates with me is what
you mentioned is it a solutionlooking for a problem, right?
We could be using AI, so let'sthrow it at a problem or let's
let's use AI and let's find aproblem that we can apply to it.
To your point, I feel we stillsee that quite a lot in
businesses today, and it's greatto hear from you really think
(05:09):
about the opportunity first,what do you really want to
accomplish?
And then think about your AI andtechnology strategies and
ancillary support.
Vin Vashishta (05:18):
Yeah.
And it forces us to justifyourselves.
I think that's one of thecritical pieces of it, is now
we're not assuming there's valuenow.
Now we're making the case.
And when you make the case, wetalk about buy-in, we talk about
stakeholders, keeping themengaged.
If they've already decided thisis what they need to do, this is
what they should be doing, and Icome along and say, I got great
(05:42):
news for you.
I can help you do this in a waythat generates more value, where
you'll be more productive or ina way that'll cost you less.
That's engagement right there.
I'm not pitching them my petproject, I'm pitching them their
pet project made better withthis technology.
Now I'm a partner.
Andreas Welsch (06:01):
That's, that's awesome.
And I think, also just the lastpiece you mentioned, partnership
is really important in this,too.
And seeing on eye level.
Now strategy can, be a prettybig term.
And I'm sure even if you do workwith C-level executives,
strategy obviously it's veryimportant.
(06:21):
But as you cascade it down inthe organization, it's still a
big term and it might be alittle fuzzier or nebulous.
So what would you say is thefirst thing that the leaders
should really consider whenworking on their AI strategy?
I know you already mentionedlook for opportunities and then
see what other strategies cansupport it or might need to be
adjusted.
But what's the first thingleaders should take a look at?
Vin Vashishta (06:44):
I would actually say three things.
I think one is the concept ofcontinuous transformation.
When you look at where thebusiness is today, they're not
going to go from level onematurity to level 30 in one
step.
So looking at it as continuous,I think that's really important
(07:04):
because you are going to startmaking decisions that amplify
and support your goals next yearand in three years.
And when you start looking attransformation as a continuous
process your decision makingchanges.
And I think that's a criticalcomponent of defining strategy.
(07:25):
Strategy must inform decisionmaking across the enterprise.
That's the implementation pieceof it, which often gets lost.
We have this great strategy,this great thesis of value
creation and justification forwhy we're using AI data
analytics.
That's awesome.
But do people downstream havethe capability to make better
(07:48):
decisions now than they couldbefore with the presence of this
strategy?
Did this do anything?
Did it actually improve?
And I think we overlook that AIdata strategy analytics strategy
isn't just for the dataorganization.
It lives throughout the entirebusiness, every business unit,
and all the way down tofrontline employees.
(08:09):
They're making decisions aboutdata, about what software to
buy, about what functionality toenable, how to spend their time.
What literacy really means.
These are all decisions that arehappening anywhere from
mid-level management to thefront line, and if strategy
doesn't inform decision making,it's useless.
(08:31):
It also has to create thisalignment.
If I'm in two different businessunits say a supply chain and
marketing, if I'm making adecision about buying a piece of
technology and that decision issomething that could impact
marketing, then my strategyneeds to be able to inform me
that, hey, I need to talk toother business units.
(08:53):
Because there could be someimpacts.
They might already havesomething.
And this happens so frequentlywhere there's a tool that can
just be repurposed.
Yes.
Where all you have to do is buya couple of seats and you're
done.
There's no adoption.
There's also the concept of, ohcould we be centralizing?
Oh, so now instead of havingfive places where data exists,
(09:13):
we can keep it to one place.
Oh, that'll help the data teamtoo.
It'll make it less expend and,so these decisions are what I
think is critical for strategyto inform.
Andreas Welsch (09:24):
That's fantastic insight.
Yeah.
Especially I think in largerorganizations like you
mentioned, there's a focus onwhat's right in front of your
nose or what's in your area ofinfluence where there could be
so many opportunities across thedifferent business units.
Fantastic.
So hey, why don't we take aquick look at the chat.
(09:44):
I see there's a question fromMike Nash.
So Mike is asking (09:47):
When using a discovery framework, how do you
weigh up one opportunity overanother?
Do you use a form of metric ormeasure?
Vin Vashishta (09:57):
I think that's just traditional strategy.
Whatever the business iscurrently using for KPIs should
be the beginning of where wetrace value back to.
So I don't want to introduce toplevel new metrics to begin with
and new KPIs to begin with.
I want to dovetail into whateverthe business is already using
(10:18):
because this is hard enough.
I don't want them to changeovernight.
I'm going to provide decisionsupport systems.
I'm going to begin to evolveKPIs and improve them.
Decisions with outcomes usingKPIs and get to the point where
KPIs are causal.
But I'm not going to do thattomorrow.
(10:39):
I'm not going to try to do thatas step one.
So no matter what the businessis using right now, I'm just
going to dovetail with that.
And my message is always, soyou're discovering opportunities
this way.
Is this how we should bediscovering opportunities?
Maybe it is.
Maybe you're right and I'm gonnahelp you prove.
Now we're gonna have data tosupport these KPIs that you're
(11:01):
already using.
And part of the discoveryframework and discovering
opportunities is discoveringopportunities to improve the
numbers, the data points that weuse to measure success and what
we're watching.
that helps us make betterdecisions.
And so opportunity discoveryisn't just products, how we're
(11:23):
gonna use data to generaterevenue.
It's also how are we gonna usedata to optimize your business
model and your operating model.
And so that becomes part of theDiscovery framework.
Andreas Welsch (11:34):
Fantastic.
Thanks, thanks for sharing that.
I'm, looking at the chat again.
It seems that also coversCynthia's question, who asked:
Do you have a standard way ofidentifying the value of a new
initiative?
So not just increased revenue orhard savings, but cost
avoidance, increasedproductivity, and so on.
So maybe if you can expand alittle bit on that or if you
(11:56):
have an example.
Vin Vashishta (11:58):
Opportunity scoping is number one, it's
critical.
Number two, it's badly done.
Product management is used totraditional software products.
Data products are a completelynew type of asset, and so we
have to measure the value of adataset differently.
We have to measure the value ofresearch artifacts differently
(12:19):
because they can be reusedmultiple times.
And so when you create software,you create it for an internal
use.
And it doesn't generalize.
You have to make significantmodifications to it in order for
it to be used again.
But a data set can be used totrain.
Five models, a hundred models,the value of that dataset.
(12:41):
When you try to capture that,you know the, what is the ROI of
capturing this particulardataset.
You have what's right in frontof you, but then you have
everything else that it opens uplater.
So this concept of assessing thecomplete and total picture of
ROI very, difficult.
So what I come into this doingis I'm trying to figure out what
(13:06):
are the most valuable processesthat we have right now?
When you look at the valuestream connected to the
workflow, what are the mostvaluable processes?
Those are going to create themost valuable data sets.
And so that's where I want tobegin.
Are we capturing data aboutthat?
And so KPIs that I'll capture,things like what percentage of
(13:27):
our workflow is producing data,and I want to start with the
highest value Generat.
Areas of the workflow.
And that's not just internally,but that's also customer
products.
How much of their workflow isbeing captured?
How much data do we have?
And those are the beginnings.
If you have nothing if you'restarting at zero and you don't
(13:48):
have any metrics to capturevalue, that's where I begin.
And then you can extend forwardfrom there.
Andreas Welsch (13:55):
Building on that when you say you look at,
processes is there a bias tolook more at the the processes
that impact your top line?
So from lead to cash if youwill, or is it more at the tail
end of this part-Finance.
(14:16):
How, can I optimize my business?
Where do you typically seeopportunities or where do you
see a bigger impact if you getyour strategy right?
Vin Vashishta (14:26):
Products.
I want the team to generaterevenue as fast as possible.
At the very earliest stage.
And yes, process mining, processdiscovery, process mapping,
process mining those are theearliest.
I go from transparent to opaque.
We want transparent.
Most of all the business doesright now is opaque and so we
(14:49):
need to start gathering dataabout that.
And so I think when you askwhere do we begin, where is
that?
If you're at zero, where's thebias for your first few
initiatives?
Yes.
And I like to sit down and findpeople's biggest.
because it's not so much ROIthat gains trust and that gets
those coalitions built and getsyou advocates for next
(15:12):
initiative.
Next initiative.
At the very early beginning, ifI can take pain away from
someone.
I have a friend for life.
That person will go to bat forme and every time I need a
recommendation, some sort ofsocial proof, I can go to them
and say, Hey, can I get a quotefrom you?
Can I send somebody over to talkto you about what this did for
you?
(15:33):
And so that's not the highestROI at that level.
But when you look at it from along-term perspective, the
allies are sometimes almost morevaluable because, now, I can
justify more initiatives.
I can get to those highest valueuse cases now.
Because I have someone who'sproven that I'm okay to touch
(15:56):
this really important thing.
It's okay.
I'm not gonna break it.
I'm not gonna mess with thegolden goose or I'm not gonna do
anything bad.
This will actually help.
And I have proof now and that'sreally the hesitation, is if I
try to go straight for thehighest value use case.
In some organizations, everyonesays yes, that's awesome.
(16:17):
We've scoped a very high valueuse case.
It's connected to eithercustomer value or some sort of
internal high value.
High need for automation usecase, high need for scaling, and
those are good criteria if Ineed.
If scale is a problem, that'susually a good high value use
case, but more times than not, Ican't touch that because no one
(16:39):
trusts that this data person orthis new data team can touch
this product because oh, what ifyou
Andreas Welsch (16:48):
And I, remember seeing that as well in my roles
or early projects, right?
It's building that clout, thatreputation.
And, to your point that's socialproof.
Just goes to show thattechnology is just this tiny
little piece in this.
Vin Vashishta (17:03):
It's culture whole thing.
So yeah, it really is.
I think what's unique with wherebusinesses are right now and
probably will be for the nextthree years is it's interesting.
The need requires people who aretechnical strategists that
understand data science andmachine learning at a
practitioner level, but alsounderstand strategy at a
(17:27):
practitioner and a planninglevel.
And when you don't have oneperson or a group that you have
people in that group who areboth, that have that hybrid
capability set, it doesn't.
I can't explain how valuabletechnical strategists are right
now in these early to mid phasesof data and AI maturity.
(17:53):
There are people who are in thedata science field right now who
are looking for a next step.
Some of'em are looking toproduct management, and that's a
great gateway into strategy.
Some of them are looking atgoing straight into the
strategist role.
These are career paths and datascientists don't have career
paths.
They just keep putting stuff infront of your data scientist
(18:13):
title, and eventually the datascientist leaves because there's
no opportunity for growth.
And if you open up roles likeproduct management, if you open
up roles like strategists, yourtechnical strategists, your AI
strategist, and datastrategists, if you open those
career paths to them, that'shuge when it comes to retention.
You don't lose your best people.
Andreas Welsch (18:34):
I think that's actually a good segue.
I'll pick another question fromthe audience.
Carly says (18:39):
Hey I'm, hoping to learn more about CoEs and
whether the recommendations areto centralize machine learning
teams or to embed them onproduct development teams to be
closer to stakeholders.
Vin Vashishta (18:50):
It's an arc.
And this is and I think we'veseen, I'm not gonna name the
company, but there was a bigtech company that recently
completed their arc.
Phase one is everything'sdecentralized, everything's
chaos.
You have platforms all over theplace.
You have people all over theplace.
You have stakeholders all overthe place.
There's nothing.
There's nothing that's undercontrol.
(19:10):
There's nothing that's reallysupporting or uniform, and so
you have to centralize.
These are essentials.
You have to go through thecenter of excellence model.
You have to create a singleprocess for data analytics, data
science, and eventuallyresearch.
You have to create a single setof tools so that you can.
(19:33):
Instead of having 18 differentcapability sets scattered
throughout the organization,really you have to build the
central, the center ofexcellence so that your business
learns how to do data science.
It becomes reproducible.
You have ways of getting thingsinto production, integrating
with existing product lines,developing and deploying new
product lines.
(19:54):
All of these things have tohappen.
And if you have eight groupsthat are separated, it's way
less efficient.
But once that happens, and thisis the transition that most
people aren't aware of, then youdecentralize, you take all of
your resources and you put themin either use internal user
(20:15):
facing, internal usersupporting, or product teams,
external product, customerfacing teams, and you have an
innovation team.
Because if you lose that,decentralization has that
massive threat.
When you centralize innovation,initiatives get prioritized
because you know the techiescontrol it.
(20:36):
Hey, we're gonna do all thiscool stuff, and so you end up
with a lot of innovationinitiatives.
Create the need to have aframework for monetizing
innovation, keeping innovationconnected to business value.
And so that's critical in theCoE so that when you distribute,
now, those innovationinitiatives are still included
(20:58):
because each team and eachproduct team knows how to
monetize innovation.
And so that gets continued.
Andreas Welsch (21:08):
That's an awesome answer.
I think that ties it nicelytogether.
And, also the, evolution.
I've been seeing a lot morequestions about that lately,
too.
When is a good time to evolveyour CoE or to transition it
into the business if, you can,to that point.
Thanks for, answering.
Vin Vashishta (21:25):
I think when you've got a track record of
success if it was gonna be onething that I would say is when
you have a history of success ofinternal initiatives that do
cost savings and improvedproductivity, and you have a
track record of successdelivering incremental new
features to existing productlines, and you figured out how
to deliver new product.
(21:45):
When you've got those running,the processes are documented,
repeatable, then you can startsaying, okay, now what can we
start decentralizing?
Andreas Welsch (21:55):
Good, point around that decentralization and
making sure that it also staysconnected, too.
Maybe to a nucleus that remainsso you know what is going on in
the different areas so that youcan support if needed.
I know early on in this episode,you mentioned there are
different kinds of strategiesthat are influenced or that you
(22:18):
should look at.
How many dimensions of strategyare there and, how do you make
your AI strategy actionable?
Or how do bring all of that totogether and get it to a point
where the rubber meets the road?
Vin Vashishta (22:32):
I think actionable, again, it's back to,
it informs decision making, andif you look at what decisions
must be made across theorganization, this is an
extension of process mapping.
and your process discovery valuestream mapping, you're also
mapping decision chains, andwhen you map decision chains,
you begin to understand how yourdecision support, your internal
(22:56):
platform is going to be created.
When you understand whatdecisions are being made about
data science, machine learning,just data in general across the
organization, you have a betterunderstanding of what your
strategy needs to cover.
What are the business needsaround.
What will they become in orderto support those opportunities
(23:17):
that you've decided the businessshould be pursuing?
What will be the implications ofthose changes?
And you can see how just massivethis thing gets, how fast this
gets to just mind blowingproportions.
And so I think the biggestcomponents of your business
model, your operating model, andthere is now this third
(23:39):
construct, which is a technologymodel.
Because technology is anintegral part of how the
business creates value and howit delivers value to customers.
And so it needs to be that.
Third foundational column incore strategy, and then
technology can branch out tocover continuous transformation.
(24:02):
You can cover your cloudstrategy, your AI strategy, your
data strategy, your analyticsstrategy, your 5G strategy.
It can cover your IoT strategy.
You can cover your quantummachine learning strategy.
Whatever comes this concept ofthe technology model connects
it.
And so that's where you begin tohave this umbrella for of the
(24:23):
complexity.
It is just as complicated asyour operating model and just as
complicated as your businessmodel.
And so it needs that new pillar.
Andreas Welsch (24:33):
Looking at these different kinds of strategies
and parts of the business thatare impacted who, should be
involved in developing acompany's AI strategy?
Is it just the data team or justthe AI team, or just the head of
AI, head of AI CoE?
Who do you work with most?
Who's, involved in, developingthat strategy?
Vin Vashishta (24:52):
Obviously I should own it.
You should bring me in and haveme take over and I'll help you.
I'll take care of this.
I think the best person to ownthe implementation is whoever's
in charge.
Your C level data leader, yourorganizational data leader, and
(25:13):
even in a startup down to whereyour, one data scientist
essentially is your CDO.
That person, even at that seedstage owns it.
They own implementation and sothey are a strategic leader, but
they also lead strategy.
This is an important conceptthat.
(25:34):
Data leaders and dataorganizational leaders don't,
bring into the process is thatyou are going to own the
implementation.
But when it comes to planningand creating the AI strategy,
that's going to happen incollaboration with everyone else
because they have expertknowledge about what they need,
(25:58):
what customers need.
They are your connection tovalue.
And so if the AI strategydoesn't have that, that
tethering to value creation, ifit doesn't support their
decision making, you failed.
If it lives in a silo and no oneknows what it is, are they
supposed to use it?
(26:19):
You can almost put a KPI aroundthis.
What percentage of the businesswould be able to tell you what
the AI strategy is?
What percentage of the businesscan answer the question?
Why do we use data?
If that's a low percentage,you're in trouble.
It's a bad strategy.
It's not effective.
Andreas Welsch (26:36):
Can you maybe summarize the, top three
takeaways for, our audiencetoday?
I see we're getting close to theend of the show but before we
wrap up Sure.
Vin Vashishta (26:44):
Takeaway zero, hire me.
Take away one would be that yourAI strategy must be actionable.
It must inform decision makingthroughout the firm.
It must be pushed down even tothe front and it must be value
centric.
It starts with the opportunitiesthat the business has identified
(27:04):
and moves forward from there.
And that sort of ties into thesecond point, which is
technology must never lead.
Strategy, business needs andvalue must always dictate what
technology should be used tocreate that value in the best
way possible.
And then really that the conceptof your center of excellence put
(27:27):
everything together.
As quickly as you can.
You're going to be takingresources away from other teams.
You are going to be takinginfrastructure and ownership of
infrastructure and software awayfrom other teams.
That's hard.
You have to have buy-in.
That's why strategy is socritical and connecting it to
core business strategy.
Cause that's the only way you'regonna justify it.
So that centralization, butwhoever runs your co oe be
(27:50):
ready.
At some point you will be out ofa job.
Because it's going todecentralize.
And that's the goal.
That's where we're going at ahigh level of maturity, but you
don't get there tomorrow.
Andreas Welsch (28:03):
And it's it's great to go in with eyes wide
open.
So great to have you call it outspecifically as well.
Hey it's, been a pleasure havingyou on.
Thank you so much for joining usand for sharing your expertise.
For those of you in the audiencefor learning with us.
Thank you for your time.
Vin Vashishta (28:18):
See ya.
Today we'll talk about making your AI strategy
actionable.
And who better to talk to aboutit than someone who helps
leaders do just that.
Hey, how are you?
Vin Vashishta (00:12):
I'm good.
Thanks for having me on.
How are you doing today?
Andreas Welsch (00:15):
Doing all right.
Hey, why don't you tell us alittle bit about yourself, who
you are and what you do.
Vin Vashishta (00:20):
Sure.
Vin Vashishta, I've been intechnology for over 25 years.
Been working in the data sciencemachine learning AI space for
the last 10, doing strategywork, AI strategy, data
strategy, data, product strategyfor just about eight years now
because as I started V squaredin 2012.
(00:40):
And almost immediately realizedI could not do any interesting
projects unless I got C-levelbuy-in unless the ROI was there.
And I began to see all of theproblems that were stacked up
behind asking for buy-in.
And that's where I got involvedin strategy because you can't do
this without holistic strategy,which is what we're talking
(01:02):
about today.
So I've helped clients, Fortune100 clients, SMEs, startups.
It's been interesting over thelast five years how interested
startups and SMEs are suddenlyin doing this in a more rigorous
way and doing this in a way thatreturns value to the business
instead of this exploratorypilot way.
So that's my story.
(01:23):
The only funny part about it wasI tried to get into AI in the
early nine or early midnineties.
When that first boom happenedand no one wanted to hire me, I
went to college for machinelearning.
I think I was gonna graduate andgo straight into Microsoft.
Yeah, I had to wait almost 15years for it to come around.
Andreas Welsch (01:44):
See, that's what I like about the industry.
There are lots of secondchances.
Awesome.
I've, been following you onLinkedIn for quite some time and
I'm sure many of of you in theaudience are familiar with you
and your content as well.
So I'm really, looking forwardto learning more about defining
an AI strategy from you.
And, to those of you in theaudience, if you're just joining
(02:06):
the stream, drop a comment inthe chat.
What do you think an AI strategyfor your business needs to
answer?
And if you have any otherquestions, please do put them in
the chat as well.
I want to make sure that we haveenough time to answer them and
keep this interactive.
Vin Vashishta (02:21):
Yeah, take advantage of the free time if
you got a question.
Definitely throw it into thechat because this is a good
opportunity.
Andreas Welsch (02:28):
Exactly.
Doesn't get better than.
At least not cheaper.
Vin Vashishta (02:33):
I promise it'll be worth every penny.
Andreas Welsch (02:37):
I'm, really curious.
I, posted a poll on, LinkedInearlier this week, and I would
love to get your perspective.
So I asked the community howshould organizations start their
AI journey, start with strategyor start with projects.
What do you say?
Vin Vashishta (02:53):
I think you should start with opportunity
discovery.
It's the same place that youshould start any sort of
strategy planning is, andinstead of saying the technology
enables these opportunities, sowe should do them I take a more
traditional strategic approach,which is, here are the
opportunities that are presentedto the business.
Here are the opportunities thebusiness is best suited to go
(03:16):
after.
Just from a basic core strategystandpoint.
Not even thinking about thetechnology at this.
And then from there we have yourbusiness strategy and then
business model, operating model.
And you should have a thirdpiece, which is the technology
model.
And that explains why thebusiness uses technology to go
after the opportunities thatit's decided on, that it's
(03:37):
discovered, that it's gonethrough the process of saying,
this is what we should be doing,not what we could be.
And when you do that, you havestrategy and value driving
technology versus technologysaying, this is possible, so why
don't we do it?
Yes.
That's the big mistake upfrontthat I think most, clients that
(03:59):
I get brought in to help, mostbusinesses that I see struggling
with data science and machinelearning, just because it's
possible, they think it'sprobably something that should
be done, and there's a lot ofFOMO.
There's a lot of what if thecompetition gets there first?
It's way more important to saythese are the best opportunities
for the business.
Technology agnostic.
(04:19):
How do we use data, analytics,AI to be better at capitalizing
on those opportunities.
And if you use the discoveryframework, you end up with a
better AI strategy because it'sthrough to the technology model.
It's tethered to your businessmodel, your operating model, and
(04:40):
core strategy.
It isn't trying to change corestrategy or drive core strategy,
it's just an extension.
Andreas Welsch (04:48):
The point that really resonates with me is what
you mentioned is it a solutionlooking for a problem, right?
We could be using AI, so let'sthrow it at a problem or let's
let's use AI and let's find aproblem that we can apply to it.
To your point, I feel we stillsee that quite a lot in
businesses today, and it's greatto hear from you really think
(05:09):
about the opportunity first,what do you really want to
accomplish?
And then think about your AI andtechnology strategies and
ancillary support.
Vin Vashishta (05:18):
Yeah.
And it forces us to justifyourselves.
I think that's one of thecritical pieces of it, is now
we're not assuming there's valuenow.
Now we're making the case.
And when you make the case, wetalk about buy-in, we talk about
stakeholders, keeping themengaged.
If they've already decided thisis what they need to do, this is
what they should be doing, and Icome along and say, I got great
(05:42):
news for you.
I can help you do this in a waythat generates more value, where
you'll be more productive or ina way that'll cost you less.
That's engagement right there.
I'm not pitching them my petproject, I'm pitching them their
pet project made better withthis technology.
Now I'm a partner.
Andreas Welsch (06:01):
That's, that's awesome.
And I think, also just the lastpiece you mentioned, partnership
is really important in this,too.
And seeing on eye level.
Now strategy can, be a prettybig term.
And I'm sure even if you do workwith C-level executives,
strategy obviously it's veryimportant.
(06:21):
But as you cascade it down inthe organization, it's still a
big term and it might be alittle fuzzier or nebulous.
So what would you say is thefirst thing that the leaders
should really consider whenworking on their AI strategy?
I know you already mentionedlook for opportunities and then
see what other strategies cansupport it or might need to be
adjusted.
But what's the first thingleaders should take a look at?
Vin Vashishta (06:44):
I would actually say three things.
I think one is the concept ofcontinuous transformation.
When you look at where thebusiness is today, they're not
going to go from level onematurity to level 30 in one
step.
So looking at it as continuous,I think that's really important
(07:04):
because you are going to startmaking decisions that amplify
and support your goals next yearand in three years.
And when you start looking attransformation as a continuous
process your decision makingchanges.
And I think that's a criticalcomponent of defining strategy.
(07:25):
Strategy must inform decisionmaking across the enterprise.
That's the implementation pieceof it, which often gets lost.
We have this great strategy,this great thesis of value
creation and justification forwhy we're using AI data
analytics.
That's awesome.
But do people downstream havethe capability to make better
(07:48):
decisions now than they couldbefore with the presence of this
strategy?
Did this do anything?
Did it actually improve?
And I think we overlook that AIdata strategy analytics strategy
isn't just for the dataorganization.
It lives throughout the entirebusiness, every business unit,
and all the way down tofrontline employees.
(08:09):
They're making decisions aboutdata, about what software to
buy, about what functionality toenable, how to spend their time.
What literacy really means.
These are all decisions that arehappening anywhere from
mid-level management to thefront line, and if strategy
doesn't inform decision making,it's useless.
(08:31):
It also has to create thisalignment.
If I'm in two different businessunits say a supply chain and
marketing, if I'm making adecision about buying a piece of
technology and that decision issomething that could impact
marketing, then my strategyneeds to be able to inform me
that, hey, I need to talk toother business units.
(08:53):
Because there could be someimpacts.
They might already havesomething.
And this happens so frequentlywhere there's a tool that can
just be repurposed.
Yes.
Where all you have to do is buya couple of seats and you're
done.
There's no adoption.
There's also the concept of, ohcould we be centralizing?
Oh, so now instead of havingfive places where data exists,
(09:13):
we can keep it to one place.
Oh, that'll help the data teamtoo.
It'll make it less expend and,so these decisions are what I
think is critical for strategyto inform.
Andreas Welsch (09:24):
That's fantastic insight.
Yeah.
Especially I think in largerorganizations like you
mentioned, there's a focus onwhat's right in front of your
nose or what's in your area ofinfluence where there could be
so many opportunities across thedifferent business units.
Fantastic.
So hey, why don't we take aquick look at the chat.
(09:44):
I see there's a question fromMike Nash.
So Mike is asking (09:47):
When using a discovery framework, how do you
weigh up one opportunity overanother?
Do you use a form of metric ormeasure?
Vin Vashishta (09:57):
I think that's just traditional strategy.
Whatever the business iscurrently using for KPIs should
be the beginning of where wetrace value back to.
So I don't want to introduce toplevel new metrics to begin with
and new KPIs to begin with.
I want to dovetail into whateverthe business is already using
(10:18):
because this is hard enough.
I don't want them to changeovernight.
I'm going to provide decisionsupport systems.
I'm going to begin to evolveKPIs and improve them.
Decisions with outcomes usingKPIs and get to the point where
KPIs are causal.
But I'm not going to do thattomorrow.
(10:39):
I'm not going to try to do thatas step one.
So no matter what the businessis using right now, I'm just
going to dovetail with that.
And my message is always, soyou're discovering opportunities
this way.
Is this how we should bediscovering opportunities?
Maybe it is.
Maybe you're right and I'm gonnahelp you prove.
Now we're gonna have data tosupport these KPIs that you're
(11:01):
already using.
And part of the discoveryframework and discovering
opportunities is discoveringopportunities to improve the
numbers, the data points that weuse to measure success and what
we're watching.
that helps us make betterdecisions.
And so opportunity discoveryisn't just products, how we're
(11:23):
gonna use data to generaterevenue.
It's also how are we gonna usedata to optimize your business
model and your operating model.
And so that becomes part of theDiscovery framework.
Andreas Welsch (11:34):
Fantastic.
Thanks, thanks for sharing that.
I'm, looking at the chat again.
It seems that also coversCynthia's question, who asked:
Do you have a standard way ofidentifying the value of a new
initiative?
So not just increased revenue orhard savings, but cost
avoidance, increasedproductivity, and so on.
So maybe if you can expand alittle bit on that or if you
(11:56):
have an example.
Vin Vashishta (11:58):
Opportunity scoping is number one, it's
critical.
Number two, it's badly done.
Product management is used totraditional software products.
Data products are a completelynew type of asset, and so we
have to measure the value of adataset differently.
We have to measure the value ofresearch artifacts differently
(12:19):
because they can be reusedmultiple times.
And so when you create software,you create it for an internal
use.
And it doesn't generalize.
You have to make significantmodifications to it in order for
it to be used again.
But a data set can be used totrain.
Five models, a hundred models,the value of that dataset.
(12:41):
When you try to capture that,you know the, what is the ROI of
capturing this particulardataset.
You have what's right in frontof you, but then you have
everything else that it opens uplater.
So this concept of assessing thecomplete and total picture of
ROI very, difficult.
So what I come into this doingis I'm trying to figure out what
(13:06):
are the most valuable processesthat we have right now?
When you look at the valuestream connected to the
workflow, what are the mostvaluable processes?
Those are going to create themost valuable data sets.
And so that's where I want tobegin.
Are we capturing data aboutthat?
And so KPIs that I'll capture,things like what percentage of
(13:27):
our workflow is producing data,and I want to start with the
highest value Generat.
Areas of the workflow.
And that's not just internally,but that's also customer
products.
How much of their workflow isbeing captured?
How much data do we have?
And those are the beginnings.
If you have nothing if you'restarting at zero and you don't
(13:48):
have any metrics to capturevalue, that's where I begin.
And then you can extend forwardfrom there.
Andreas Welsch (13:55):
Building on that when you say you look at,
processes is there a bias tolook more at the the processes
that impact your top line?
So from lead to cash if youwill, or is it more at the tail
end of this part-Finance.
(14:16):
How, can I optimize my business?
Where do you typically seeopportunities or where do you
see a bigger impact if you getyour strategy right?
Vin Vashishta (14:26):
Products.
I want the team to generaterevenue as fast as possible.
At the very earliest stage.
And yes, process mining, processdiscovery, process mapping,
process mining those are theearliest.
I go from transparent to opaque.
We want transparent.
Most of all the business doesright now is opaque and so we
(14:49):
need to start gathering dataabout that.
And so I think when you askwhere do we begin, where is
that?
If you're at zero, where's thebias for your first few
initiatives?
Yes.
And I like to sit down and findpeople's biggest.
because it's not so much ROIthat gains trust and that gets
those coalitions built and getsyou advocates for next
(15:12):
initiative.
Next initiative.
At the very early beginning, ifI can take pain away from
someone.
I have a friend for life.
That person will go to bat forme and every time I need a
recommendation, some sort ofsocial proof, I can go to them
and say, Hey, can I get a quotefrom you?
Can I send somebody over to talkto you about what this did for
you?
(15:33):
And so that's not the highestROI at that level.
But when you look at it from along-term perspective, the
allies are sometimes almost morevaluable because, now, I can
justify more initiatives.
I can get to those highest valueuse cases now.
Because I have someone who'sproven that I'm okay to touch
(15:56):
this really important thing.
It's okay.
I'm not gonna break it.
I'm not gonna mess with thegolden goose or I'm not gonna do
anything bad.
This will actually help.
And I have proof now and that'sreally the hesitation, is if I
try to go straight for thehighest value use case.
In some organizations, everyonesays yes, that's awesome.
(16:17):
We've scoped a very high valueuse case.
It's connected to eithercustomer value or some sort of
internal high value.
High need for automation usecase, high need for scaling, and
those are good criteria if Ineed.
If scale is a problem, that'susually a good high value use
case, but more times than not, Ican't touch that because no one
(16:39):
trusts that this data person orthis new data team can touch
this product because oh, what ifyou
Andreas Welsch (16:48):
And I, remember seeing that as well in my roles
or early projects, right?
It's building that clout, thatreputation.
And, to your point that's socialproof.
Just goes to show thattechnology is just this tiny
little piece in this.
Vin Vashishta (17:03):
It's culture whole thing.
So yeah, it really is.
I think what's unique with wherebusinesses are right now and
probably will be for the nextthree years is it's interesting.
The need requires people who aretechnical strategists that
understand data science andmachine learning at a
practitioner level, but alsounderstand strategy at a
(17:27):
practitioner and a planninglevel.
And when you don't have oneperson or a group that you have
people in that group who areboth, that have that hybrid
capability set, it doesn't.
I can't explain how valuabletechnical strategists are right
now in these early to mid phasesof data and AI maturity.
(17:53):
There are people who are in thedata science field right now who
are looking for a next step.
Some of'em are looking toproduct management, and that's a
great gateway into strategy.
Some of them are looking atgoing straight into the
strategist role.
These are career paths and datascientists don't have career
paths.
They just keep putting stuff infront of your data scientist
(18:13):
title, and eventually the datascientist leaves because there's
no opportunity for growth.
And if you open up roles likeproduct management, if you open
up roles like strategists, yourtechnical strategists, your AI
strategist, and datastrategists, if you open those
career paths to them, that'shuge when it comes to retention.
You don't lose your best people.
Andreas Welsch (18:34):
I think that's actually a good segue.
I'll pick another question fromthe audience.
Carly says (18:39):
Hey I'm, hoping to learn more about CoEs and
whether the recommendations areto centralize machine learning
teams or to embed them onproduct development teams to be
closer to stakeholders.
Vin Vashishta (18:50):
It's an arc.
And this is and I think we'veseen, I'm not gonna name the
company, but there was a bigtech company that recently
completed their arc.
Phase one is everything'sdecentralized, everything's
chaos.
You have platforms all over theplace.
You have people all over theplace.
You have stakeholders all overthe place.
There's nothing.
There's nothing that's undercontrol.
(19:10):
There's nothing that's reallysupporting or uniform, and so
you have to centralize.
These are essentials.
You have to go through thecenter of excellence model.
You have to create a singleprocess for data analytics, data
science, and eventuallyresearch.
You have to create a single setof tools so that you can.
(19:33):
Instead of having 18 differentcapability sets scattered
throughout the organization,really you have to build the
central, the center ofexcellence so that your business
learns how to do data science.
It becomes reproducible.
You have ways of getting thingsinto production, integrating
with existing product lines,developing and deploying new
product lines.
(19:54):
All of these things have tohappen.
And if you have eight groupsthat are separated, it's way
less efficient.
But once that happens, and thisis the transition that most
people aren't aware of, then youdecentralize, you take all of
your resources and you put themin either use internal user
(20:15):
facing, internal usersupporting, or product teams,
external product, customerfacing teams, and you have an
innovation team.
Because if you lose that,decentralization has that
massive threat.
When you centralize innovation,initiatives get prioritized
because you know the techiescontrol it.
(20:36):
Hey, we're gonna do all thiscool stuff, and so you end up
with a lot of innovationinitiatives.
Create the need to have aframework for monetizing
innovation, keeping innovationconnected to business value.
And so that's critical in theCoE so that when you distribute,
now, those innovationinitiatives are still included
(20:58):
because each team and eachproduct team knows how to
monetize innovation.
And so that gets continued.
Andreas Welsch (21:08):
That's an awesome answer.
I think that ties it nicelytogether.
And, also the, evolution.
I've been seeing a lot morequestions about that lately,
too.
When is a good time to evolveyour CoE or to transition it
into the business if, you can,to that point.
Thanks for, answering.
Vin Vashishta (21:25):
I think when you've got a track record of
success if it was gonna be onething that I would say is when
you have a history of success ofinternal initiatives that do
cost savings and improvedproductivity, and you have a
track record of successdelivering incremental new
features to existing productlines, and you figured out how
to deliver new product.
(21:45):
When you've got those running,the processes are documented,
repeatable, then you can startsaying, okay, now what can we
start decentralizing?
Andreas Welsch (21:55):
Good, point around that decentralization and
making sure that it also staysconnected, too.
Maybe to a nucleus that remainsso you know what is going on in
the different areas so that youcan support if needed.
I know early on in this episode,you mentioned there are
different kinds of strategiesthat are influenced or that you
(22:18):
should look at.
How many dimensions of strategyare there and, how do you make
your AI strategy actionable?
Or how do bring all of that totogether and get it to a point
where the rubber meets the road?
Vin Vashishta (22:32):
I think actionable, again, it's back to,
it informs decision making, andif you look at what decisions
must be made across theorganization, this is an
extension of process mapping.
and your process discovery valuestream mapping, you're also
mapping decision chains, andwhen you map decision chains,
you begin to understand how yourdecision support, your internal
(22:56):
platform is going to be created.
When you understand whatdecisions are being made about
data science, machine learning,just data in general across the
organization, you have a betterunderstanding of what your
strategy needs to cover.
What are the business needsaround.
What will they become in orderto support those opportunities
(23:17):
that you've decided the businessshould be pursuing?
What will be the implications ofthose changes?
And you can see how just massivethis thing gets, how fast this
gets to just mind blowingproportions.
And so I think the biggestcomponents of your business
model, your operating model, andthere is now this third
(23:39):
construct, which is a technologymodel.
Because technology is anintegral part of how the
business creates value and howit delivers value to customers.
And so it needs to be that.
Third foundational column incore strategy, and then
technology can branch out tocover continuous transformation.
(24:02):
You can cover your cloudstrategy, your AI strategy, your
data strategy, your analyticsstrategy, your 5G strategy.
It can cover your IoT strategy.
You can cover your quantummachine learning strategy.
Whatever comes this concept ofthe technology model connects
it.
And so that's where you begin tohave this umbrella for of the
(24:23):
complexity.
It is just as complicated asyour operating model and just as
complicated as your businessmodel.
And so it needs that new pillar.
Andreas Welsch (24:33):
Looking at these different kinds of strategies
and parts of the business thatare impacted who, should be
involved in developing acompany's AI strategy?
Is it just the data team or justthe AI team, or just the head of
AI, head of AI CoE?
Who do you work with most?
Who's, involved in, developingthat strategy?
Vin Vashishta (24:52):
Obviously I should own it.
You should bring me in and haveme take over and I'll help you.
I'll take care of this.
I think the best person to ownthe implementation is whoever's
in charge.
Your C level data leader, yourorganizational data leader, and
(25:13):
even in a startup down to whereyour, one data scientist
essentially is your CDO.
That person, even at that seedstage owns it.
They own implementation and sothey are a strategic leader, but
they also lead strategy.
This is an important conceptthat.
(25:34):
Data leaders and dataorganizational leaders don't,
bring into the process is thatyou are going to own the
implementation.
But when it comes to planningand creating the AI strategy,
that's going to happen incollaboration with everyone else
because they have expertknowledge about what they need,
(25:58):
what customers need.
They are your connection tovalue.
And so if the AI strategydoesn't have that, that
tethering to value creation, ifit doesn't support their
decision making, you failed.
If it lives in a silo and no oneknows what it is, are they
supposed to use it?
(26:19):
You can almost put a KPI aroundthis.
What percentage of the businesswould be able to tell you what
the AI strategy is?
What percentage of the businesscan answer the question?
Why do we use data?
If that's a low percentage,you're in trouble.
It's a bad strategy.
It's not effective.
Andreas Welsch (26:36):
Can you maybe summarize the, top three
takeaways for, our audiencetoday?
I see we're getting close to theend of the show but before we
wrap up Sure.
Vin Vashishta (26:44):
Takeaway zero, hire me.
Take away one would be that yourAI strategy must be actionable.
It must inform decision makingthroughout the firm.
It must be pushed down even tothe front and it must be value
centric.
It starts with the opportunitiesthat the business has identified
(27:04):
and moves forward from there.
And that sort of ties into thesecond point, which is
technology must never lead.
Strategy, business needs andvalue must always dictate what
technology should be used tocreate that value in the best
way possible.
And then really that the conceptof your center of excellence put
(27:27):
everything together.
As quickly as you can.
You're going to be takingresources away from other teams.
You are going to be takinginfrastructure and ownership of
infrastructure and software awayfrom other teams.
That's hard.
You have to have buy-in.
That's why strategy is socritical and connecting it to
core business strategy.
Cause that's the only way you'regonna justify it.
So that centralization, butwhoever runs your co oe be
(27:50):
ready.
At some point you will be out ofa job.
Because it's going todecentralize.
And that's the goal.
That's where we're going at ahigh level of maturity, but you
don't get there tomorrow.
Andreas Welsch (28:03):
And it's it's great to go in with eyes wide
open.
So great to have you call it outspecifically as well.
Hey it's, been a pleasure havingyou on.
Thank you so much for joining usand for sharing your expertise.
For those of you in the audiencefor learning with us.
Thank you for your time.
Vin Vashishta (28:18):
See ya.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.