Automate Finance End-to-End

In this episode of "Full Tech Ahead," host Amanda Razani speaks with Prashanth Saradesai, Head of AI for Wiss, about how accounting and finance organizations can successfully implement artificial intelligence. 

Saradesai shares insights from evaluating over 200 AI vendors, noting that many finance teams hesitate to adopt AI because they struggle to prove a clear Return on Investment (ROI). He explains that treating AI as an isolated "point solution" (e.g., merely extracting invoice data) is ineffective; instead, organizations must automate end-to-end workflows to see real value. 

Furthermore, he emphasizes that in the strict finance sector, "good enough" is unacceptable, making human-in-the-loop processes essential to mitigate AI hallucinations. 

Finally, he advises companies to refine their internal processes, build strong data foundations, and prioritize change management to ensure a successful AI adoption.

Key Quotes: 

"In finance, 'good enough' is not an answer. You can't say, 'hey, my finances are good enough.' Even one number doesn't work; it flows through your financial reporting."
"If leadership has the vision of bringing AI to the organization, but if you are not bringing everybody in the organization... there won't be any value."
"Start investing in AI from the perspective of using it for your end-to-end workflow... being AI native, thinking from a standpoint of making all your employees AI fluent, are the North Stars."

Takeaways:

Focus on End-to-End Workflows: Using AI for a single task like invoice extraction won't drive significant ROI. AI should be implemented to handle the entire workflow—from extraction and matching to approvals, posting entries, and reconciliation.

Fix Processes and Data First: The biggest mistake companies make is starting with the technology. Organizations must first clearly define their internal processes, clear bottlenecks, and build a solid, well-organized data foundation before deploying AI.

Prioritize Change Management: An AI initiative will fail if the vision stays only at the executive level. Training employees on both the advantages and the limitations (like hallucinations) of AI is crucial for successful, company-wide adoption.

Capture Decision "Context": As the industry moves toward autonomous AI agents capable of "computer use," the organizations that will succeed are those building "context graphs"—documenting not just their raw data, but the specific reasons and context behind their past business decisions.

Find Amanda Razani on LinkedIn.  https://www.linkedin.com/in/amanda-razani-990a7233/

Follow the FTA LinkedIn Page: https://www.linkedin.com/company/full-tech-ahead/

Visit the FTA website: https://fulltechahead.com/

Check out the Substack Channel: https://fulltechahead.substack.com/

Episode Transcript

SPEAKER_01 (00:19):
Hello and welcome to Full Tech Ahead.
I'm excited to be here todaywith Prashant Sarah Design.
He is director of AI for WIS.
How are you doing today?

SPEAKER_00 (00:31):
Hi, Amanda.

SPEAKER_01 (00:34):
I am doing well.
Can you share a little bit aboutWIS?
What services are providedthere?

SPEAKER_00 (00:41):
Yeah, absolutely.
So WIS is one of the top hundredforms.
We focus on mid-market.
We provide accounting tax aswell as client advisory services
within New York, New Jersey, andeven nationally as well.

SPEAKER_01 (01:00):
Okay, great.
Well, we're here today to talkabout accounting firms and
finance organizations and howthey can select the right AI
implementation partner.
So you, I think you spent thelast year doing a study of over
200 vendors.
Can you talk a little bit aboutthat?

SPEAKER_00 (01:17):
Basically, our approach is as technology was
evolving.
We see how AI is taking over allaspects of the business.
So we started evaluating.
So WIS as a firm uh decided togo in that direction very early.
So we we took a stance that weare gonna be AI native and we

(01:39):
are gonna transform our ownbusiness internally through AI
and and the evolving technologythat it is, right?
So so we started um evaluating,uh, and then that's how there
were a lot of startups coming upin the market, and we started
evaluating them.
So we have our own internalframework through which we

(02:00):
evaluate these uh differentvendors because as you know, the
the venture capital field islike so so uh bustling for last
few years from from the adventof ChatGPT, right?
So uh basically we startedreviewing a lot of these finance

(02:22):
uh vendors or the AI platformsuh which claim to be having
solutions which will solve ourproblems.
So we started looking at uh someof these, and what we what we
learned is only around you know60% of the vendors that we
looked at mainly have are theyare trying to use AI in some way

(02:47):
or the other, right?
But rest of the market is is notreally using AI.
They are still in the initialstages of adoption, or some of
them are just sitting on the onthe beach thinking about it and
they don't know how to go aboutit.
I think that's where somebodylike WIS comes in and helps uh
those organizations.

(03:07):
So a lot of our clients areasking us, hey, there is Chat
GPT, Claude, there are so manysolutions today.
How can we implement it, right?
That's where we are helping themimplement those solutions.

SPEAKER_01 (03:20):
Okay.
So why do you think it is thatso many finance teams are not
yet exploring AI options?
Is it that overwhelm of thenumber of options?
Or from your experience, why isthat?

SPEAKER_00 (03:34):
Yeah, we like I feel and and what we are seeing at at
WIS is when you look at AIadoption, you can have some of
the point solutions.
What I mean is like, hey, I wantto process invoices.
You can use AI to just extractinvoices into a readable format,
right?
But that's not gonna drive ROIfor these companies.

(03:57):
So when we look at the wholeworkflow, finance workflow, uh,
for example, then if we canbring in AI, that's when it's
gonna make a larger impact,right?
So so for the same invoice uhexample, if we can have a
workflow which can not justextract invoices, it can do
three-way match, it can code theGLs and then help us to go

(04:22):
through the approval processthese organizations have, then
you post an entry, then youperform reconciliation.
So if you can, if AI can executean end-to-end workflow, that's
where it's meaningful, andthat's where organizations can
show the return on investmentsthat they are making, right?
So a lot of them are not ableto.

(04:44):
And some of the challenges thatthey have today is they don't
have their processes welldefined, right?
We come across that.
There are two major aspects Iwant to mention.
One is their process, becausethey haven't defined their
process really well.
There are a lot of bottlenecks,there are a lot of manual
processes, right?

(05:05):
So they need to fix thoseaspects first before they say,
like, hey, I want to transformmy entire workflow.
And the second thing we see isthe data foundations, which is
very critical for AI, right?
So if you AI is as good as whatdata you provide to it, right?
So these LLM models are verygood these days, but again, they

(05:26):
are trained on the internetdata, right?
So if you want to have morespecific solutions for your
organization, then you need tohave a good data that you feed
into LLMs, right?
So so we see there are a lot ofgaps in the data uh foundations
that these organizations have.
So these are two main concerns.
And then the third one I justwant to mention, which I have

(05:48):
lived through my experience, isalso the chain management
process.
So if you look at it, ifleadership has the vision of
bringing AI to the organization,but if you are not bringing
everybody in the organization,have value, there won't be
enough value that you can definewithout doing that.

(06:10):
Because if if everyone in theorganization understands what
value AI brings, then I thinkyou will be successful as you
roll these out.
Because like that's where a lotof these AI initiatives fail
with with the change managementaspect for the organizations.

SPEAKER_01 (06:26):
Yeah, imagine communication and training are
critical.

SPEAKER_00 (06:31):
Absolutely.
Absolutely.
I think defining that vision,training your employees on
advantages of AI, and more thanadvantages, I think what are the
limitations of AI?
I think if you can talk throughthe limitations, that's when
everybody will know, okay, theseAI models can hallucinate,

(06:53):
right?
So when they hallucinate, how doyou reduce that hallucination,
right?
So you know, and then the otheraspect of it is especially in
finance, you cannot rely on anLLM model that it is generating
results, right?
Because what happens is inmarketing and some other field,

(07:14):
maybe if you're developing acampaign, it's okay.
You have text and reviews, butin finance, good enough is not
an answer, right?
So you can't say, like, hey, myfinances are good enough.
No, even one number doesn'twork, it flows through your
financial reporting.
And there are other challengesto it, right?
From a from a financialreporting standpoint, from an

(07:36):
audit standpoint.
So so from that angle, it isvery, very important for your
employees to know thelimitations, have the
human-in-the-loop process foryour finance workflows that that
is critical for the success.

SPEAKER_01 (07:50):
Yeah, especially in the finance sector, how
important or critical is it tohave no errors?
Obviously, I'm sure many leadersare worried about AI and LLM
errors.

SPEAKER_00 (08:05):
It's massive.
It's massive, right?
Because as I mentioned, AIerrors in finance uh are not
cosmetic, right?
You have you have you have a lotof repercussions uh if you get
get any numbers wrong.
Uh these general AI LLMs giveyou plausible, right?
Like if you look at these LLMmodels, they are

(08:26):
non-deterministic, right?
So they are gonna give youprobabilistic answers to the
questions that you put in.
So you need to be you need to becareful, right?
Otherwise, repercussions aredangerous.
And and that ties back to whyCFOs, finance advisors, are a
little hesitant on implementingAI at a scale that we think it's

(08:47):
gonna work.
And the good news is this fieldis evolving so fast.
What we spoke last month, likewe can be conflicting them if we
speak today, because they aresolving.
These AI models are evolvingexponentially.
Whatever challenges we areseeing last month, they are
getting solved, right?
So if you look at the pace, thethe the positive aspect of it is

(09:10):
for FCFO, you need to constantlylook at it.
And and and what they need to dois be on the train, right?
Like don't don't just don't sitand watch.
They need to get on the train,build the processes, test it
out, figure out the depth andthe limitations, and then move
ahead, right?
If you if you don't do this, theyou will be left behind.

(09:31):
There'll be somebody who's gonnawork through this and they're
gonna disrupt the market foryou.

SPEAKER_01 (09:37):
I imagine one thing they're really focused on is
that return on investment.
So how do they track that?

SPEAKER_00 (09:45):
Yeah, that that's that's a very, very good
question.
I think I think in the market wesee that ROI has been a big, big
uh question mark because thesetechnologies are evolving.
Uh, there is huge investmentsmade by organizations, but they
are not able to show the ROI forthat, right?
So, so I just want to give thisbackground because like last

(10:07):
year, right, we we we spoke somuch about agents, right?
And then those agents were doingone step, two step, three step
at a time, right?
So I think like recently, lasttwo, three months, with some of
these more powerful reasoningmodels, which are able to solve
for a lot of it.
They're not just able, they'renot just solving one, two, three

(10:31):
steps.
There is agentic orchestration,multi-agent uh orchestrations
that are happening, and they areable to reason very, very well
and execute end-to-end workflow.
So I think that's a big shiftthat has happened with the
technology, right?
Now, when you think of ROI, youare able to look at it from that

(10:53):
lens.
And how do organizations track?
I mean, you need to define avery robust metrics for all of
this, right?
So now if you're starting, ifyou are starting and automating
an end-to-end workflow, you needto set your metrics and be able
to track them, right?
So so what's happening is if youdon't define metrics before you

(11:16):
start investing, then you'rejust like flowing.
You don't have a way to look atit from and and evaluate it for
return on investment, right?
So what is the biggest thing isto come up with metrics, track
them regularly, right?
So you know how you'reprogressing, figure out where
there are gaps, and thenre-evaluate and if you need to

(11:39):
invest in something else or not,right?
Because sometimes organizationsget committed to a vendor or a
vendor solution, and then theythink, hey, ROI is gonna come.
They were sold on a on a on somevision, but if you're not
tracking it, you spend thatmoney for six months, eight
months, and then you realize,like, no, this is not the right

(12:00):
solution for me.
And you couldn't, you couldn'tshow anything at the end.
And and we have seen a lot ofcases like that, right?
So that's why I think definingthat metrics becomes very, very
critical.

SPEAKER_01 (12:12):
What from your experience, what are some of the
common mistakes or roadblocksthat finance teams make when
they're implementing these AItools?

SPEAKER_00 (12:22):
I think I I want to call out three three mistakes
here.
So the first one is I say a lotof them start with technology,
right?
So I mean, if you want to solvea problem, hey, I want to use
this technology to solveproblems, but instead, process
is where you need to look at,right?
So we we we do uh advise a lotof our clients where first thing

(12:46):
we do is looking at theirprocess, understanding their
process, what are their data,how does it look, how accessible
the data is, right?
So once we were able to mapthat, then technology
conversations can come in,right?
So I think that's the firstmistakes a lot of them do.
Um, and then the second one isuh treating AI as a point

(13:08):
solution, right?
Like as I mentioned earlier withan invoice example, like, hey,
you want to solve just one.
No, you you need to look at itholistically and see how you can
draw out.
And in some cases, we have comeacross where you define the
workflow, you think a technologycan solve the problem, but you
may not be able to solve thewhole workflow at once, right?

(13:30):
So sometimes you need to havethat plan to solve the whole
workflow, but you have to takesmaller bytes and be able to uh
get there in two months, threemonths, right?
So if you try to solve all ofit, there will be challenges.
So I think that's the secondsecond mistake a lot of them do.
And then the third is the chainmanagement, right?

(13:51):
I think the the chain managementis is very crucial for
organization.
Starts from the leadership andthen drawing the vision and then
bringing the team along is iscrucial.
And sometimes what happens isthere is a vision from the
leadership.
It doesn't percolate down to theuh people who's who are
executing day-to-day work,right?

(14:13):
And and there is that big gap uhwhich a lot of organizations
don't close up front.

SPEAKER_01 (14:18):
So that that's another challenge or a mistake
that well, this, as youmentioned, this technology is
rapidly advancing.
What do you see on the horizonin the future in the finance
sector?

SPEAKER_00 (14:34):
It's it's evolving so much, right?
As you mentioned.
Uh so on the on the horizon, Ithink uh I think there are a lot
of startups, a lot of uhsolutions being developed in the
market, right?
So organizations have to look atit where they invest in those
kind of solutions, you buy thosesolutions, or where they they

(14:59):
need to invest in buildingsomething internally, right?
So so especially just quickly,wherever there are data, your
proprietary data that you don'twant to give out to your vendors
or or any of the LLMs, and youneed to make sure that you
protect your IP, right?
Because so you need to look atit from that perspective.

(15:19):
And the second aspect of what'shappening in in the market and
with these technologies is uh wewere talking about agents doing
a workflow, right?
Where it is heading with thingslike open claw and what we see
computer use aspects coming intoplay.
So the the vision are the thingsthat are going the direction

(15:43):
it's going is uh like thesecomputer use, if you give enough
context, if you give all thedata that it needs and you give
access to these models, they areable to execute everything 24-7,
right?
I think I think the horizon uhis is basically you will have

(16:06):
systems at some point which willbe uh edge-based, or like what I
mean is like you can have oneserver with cunt your data is
contained within that server,and you give access to an agent,
uh, something like OpenClaw, uhkind of a solution.
Now, a lot of the biggerorganizations are coming up with

(16:26):
similar solutions, right?
Like Claude came up with uh uh asolution dispatch, and then
Nvidia has uh a solution thatthey announced recently.
So all of this as they matureand as security aspects come
into play, as you're able to putyour data securely into these uh
these servers uh and then giveaccess to these agents, they're

(16:49):
they're probably gonna do a lotof the things uh 24-7.
I think I think we are headingin that direction.
I think people who are are theorganizations who are investing
in capturing their context,right?
Context is beyond data, what aresome of the decisions that you
make, right?
So so for we run a lot of clientuh that we advise and and a lot

(17:12):
of clients within theirorganization, there are a lot of
decisions made, right?
So so are you capturing thosedecisions, right?
Because LLMs, you can give data,but the decisions are not
captured.
Like organizations who cancapture those decisions, who can
capture, hey, why did you take adecision?

(17:33):
What were the reasons?
And if you capture that kind ofan information, which is more of
what we call as context, ornowadays it's also referred to
as context graphs.
If you are building that contextgraph, that's gonna give you a
lot of leverage when tools likethese computer use and open claw
things you want to adopt.

(17:54):
So if you have that built out,people who are doing that today
will be successful, able to getto ROIs much faster than others.

SPEAKER_01 (18:03):
It will be interesting to watch the future
unfold for sure.

SPEAKER_00 (18:07):
Absolutely.
Absolutely.
I mean it's it's exciting forfor me having been in this
field, the the pace and and theand and the possibilities are
just unbelievable.
It it it amazes me every daywhen I wake up and and see the
news, see the things coming up.
It's it's so exciting.

SPEAKER_01 (18:29):
Well, if there was one key takeaway you could leave
our audience today with, whatwould that be?

SPEAKER_00 (18:34):
Start investing in AI from a perspective of using
it for your end-to-end workflow.
And when you look at it fromthat lens, you start defining
your process, you start clearingyour bottlenecks.
And then once you get there, youmake sure your data gets.
So I think that question willlead to multiple uh other things

(18:57):
that actions that you can taketo get there.
I think if if somebody is justlike trying and not really
investing in AI is is gonna bereally dangerous uh in a way.
So I think a lot oforganizations are doing in doing
it in in phases or in in smallerbytes in a way.

(19:18):
But I think being AI native,thinking from a standpoint of
making all your employees AIfluent are the North stars that
that organizations have to setup.

SPEAKER_01 (19:32):
All right.
Well, thank you so much forcoming on the show and sharing
your insights with us today.

SPEAKER_00 (19:38):
Thank you, Amanda.
Thank thank you for uh giving methis opportunity to uh share
share our thoughts and thankthanks for uh bringing me and
and Viz into this conversation.

SPEAKER_01 (19:48):
And thank you to our audience.
If you have any questions orcomments, leave those below and
I'll be sure to respond to it.
Have a great day.