April 15, 2025 • 9 mins
I spoke with Thomas Bueler, the Co-founder and Chief Product Officer at Vecflow, a legal AI platform designed to help lawyers review and draft documents for use cases in M&A and class action litigation. We discussed the role of AI in expanding legal workflows, the importance of minimizing hallucinations in large language models, and the best practices for evaluating different AI products.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Welcome to Reinventing Professionals, a podcast hosted by industry analyst Ari Kaplan,
(00:07):
which shares ideas, guidance and perspectives from market leaders shaping the next generation
of legal and professional services.
This is Ari Kaplan and I'm speaking today with Thomas Bueller, the co-founder and chief
product officer at Vecflow.
A legal AI platform designed to help lawyers review and draft documents with use cases
(00:30):
in mergers and acquisitions and class action litigation.
Hi Thomas, how are you?
Great, thanks for having me.
Tell us about your background and the genesis of Vecflow.
I'm co-founder and CPO of Vecflow.
Prior to founding Vecflow, I worked in a very small litigation from in Alaska.
I've also done some ML research at Columbia.
So it was in some sense, no brain or starting Vecflow in around 2022 as large language
(00:54):
models were developing and as law firms were right for disruption.
I think our initial thesis at Vecflow was around needing law firms to adopt AI and to adopt
load language models.
We very quickly recognized that we needed to build end workflow style solutions to automate
more robust task at law firms like a document review and drafting.
(01:17):
And we currently have top use cases as you mentioned in M&A in class action.
I think what we're really focused on is we're writing both a general purpose AI
solution for every line in being able to automate and one click workflows around the law firm
as well.
How do you see the role of AI expanding in legal workflows?
In 2024, you saw a sort of expansion of the role of AI coming from chat based applications
(01:42):
to more fully workflow style solutions for law firms.
So you're going to continue to see if the beginning of 2025 is any indication more
for reasoning based models working at law firms as well as more AI integration within
the particular applications in many different practice areas.
For instance, if you're in M&A firm, I think that to use Oliver in 2025 means to be able
(02:07):
to draft fully forstrafts of like share purchase agreements, LPAs, whatever other type
of document that you're working with just from a term sheet.
So we'll be able to get your first draft much more quickly there.
Where law firms implementing Oliver?
The firms we're working with really think about this as less as a product and more as a
vast set of their firm here in terms of in terms of where on the firm we are used were
(02:31):
used particularly in the drafting and review stage.
One of their classic examples is like due diligence.
You can review with Oliver over a thousand different documents and you can easily analyze
for like change of control provisions and then you can use that information when doing
later steps down the line as well.
I think in terms of actual implementation details, we have many different kinds of deployments
(02:54):
at firms so you can either deploy in our cloud in our private cloud, you can also deploy
in a VPC deployment at a firm and you can as well do it on-prem.
So with our partnership with meta, we can deploy our AI always on-premise for law firms
as well.
And this allows firms to use confidential information within their product and satisfy all firewalls
(03:17):
and content security measures there with their clients and so they can better provide value
for their clients at the end of the day.
How do you approach the onboarding and training process?
We've spent a lot of time over the past few months really perfecting our onboarding process.
So I think like we know that it can be a steep learning curve when it comes to using
(03:38):
AI in law and so we do both kind of an onboarding session at the beginning.
We also do many different workshops, weekly workshops.
We have the first like 60 minutes there focusing on a tutorial about how lawyers can use our
product for particular tasks, whether that is in like due diligence, drafting, writing
of complaints or just basically legal research with SEC Edgar.
(04:02):
We then have a tutorial there.
We switch the tutorial to be a sort of side by side workflow, right?
So we work side by side as in person as we can be with the power users of our tool to get
them to be more power users.
And then of course the power users start to increase their own productivity and talk to
other lawyers down the line about our product as well.
(04:24):
So I think like it's a real comprehensive process of onboarding.
We also try to track as many metrics usage and accuracy as we can and we try to get that
to be as visible as possible to as many people at the firm as we can.
So it's really easy to measure the value both of a highlight of the Oliver platform, but
also of a sort of like holistic overview of how the firm is spinning their own time because
(04:49):
oftentimes firms don't even have a good understanding of where they're spinning the time initially.
How does Vectlo minimize the risk of hallucinations?
This is the top concern of a lot of lawyers and a lot of firms.
I think that the first thing I mentioned everyone is hallucination is a making up of fake information.
Accuracy is something different.
(05:10):
That's like how many questions can you accurately or how many tasks you can accurately respond
to.
We at Vectlo have spent a lot of time perfecting a solution to this and we have very low
risks of hallucination in general.
Every time you ask a question, we make sure it's validated with source information and
we give that source information precisely to a user.
(05:31):
We also have very strict guard rails.
So for instance, every response that you ask a model, we put it through over a hundred
different other models to verify that it's the right answer.
And finally, we rely upon asking the lawyer to clarify any ambiguous questions.
So rather than going through a particular task and then at the end having your associate
(05:54):
draft you the wrong document, we try to have our system Oliver asks the lawyer, is this
the right pathway?
Do you want to do X or Y?
I think this really helps keep the lawyer in the loop, but it also helps educate Oliver
on what type of task it's doing going down the line.
And finally, we recently had an email from a new client of ours that mentioned, look,
(06:14):
rethink Oliver hallucinated on this particular aspect.
And in fact, what was going on is that he was reviewing a 6,000 page document and on one
page there was a cursive handwriting on a colored piece of paper.
And our lawyers couldn't read it, but our system was able to extract the text and get the
detail that their client actually did something bad there.
So I think that definitely in combination with a lawyer and a tool, you can really see a lot
(06:39):
of value coming out of this.
What are some best practices for evaluating different AI products?
Evaluation is extremely hard.
It comes up every time when I talk to a firm.
It's almost associated with the market mapping problem.
It's been really hard to map how products compare to others in the past few years.
(06:59):
And I think the law firm is definitely don't have the number of people to accurately evaluate
every product on the market.
And so you're going to need to look outside of just the product and the pitches that you
are being seen.
There's a number of third party studies.
Valse is in the right direction.
As features are developing very fast, you also have to look at the team that's behind the
product and the platform.
(07:20):
There are some teams and products that we know are very good, maybe in the past, but probably
don't have the expertise or the technical background in order to build state of the art
products going into 2025 and going into 2026.
There.
How do you see the use of legal technology in law firms changing in 2025?
(07:41):
Thirty years ago, you had a shift towards digitization and having legal research beyond
the line.
We know that transformed firms in many different ways that kind of equalized a lot of firms
that were able to do legal research across different practice areas.
But it did not really change the underlying business model of the firm in 2025.
(08:01):
And we're going to see and we have already seen a shift from chat as a model to workflow
as a model.
You're going to see ever more indications that this workflow based approach really promises
to bring about underlying business change to a firm.
We have some of our customers who were previously dealing with 200 of one type of case a month
(08:24):
and now can easily deal with 500 of one case a month with the same amount of people.
When you do that, you can both charge more and change away from available model because
I think in the day, if you're doing 200 or 500 cases a year, it can be a push button workflow
that you can leverage your own internal information.
So if 30 years ago the change was about how public information was dealt with public case
(08:47):
law, I think you're going to see a change coming up about how private information is dealt
with at a law firm.
This is Ari Kaplan speaking with Thomas Bueller, the co-founder and chief product officer
at VEC Flow, a legal AI platform designed to help lawyers review and draft documents
with use cases in M&A and class action litigation.
(09:09):
Thomas, thank you so very much.
Thank you Ari, wonderful talking.
Thank you for listening to the reinventing professionals podcast.
Visit reinventingprofessionals.com or Ari Kaplan advisors.com to learn more.
Welcome to Reinventing Professionals, a podcast hosted by industry analyst Ari Kaplan,
(00:07):
which shares ideas, guidance and perspectives from market leaders shaping the next generation
of legal and professional services.
This is Ari Kaplan and I'm speaking today with Thomas Bueller, the co-founder and chief
product officer at Vecflow.
A legal AI platform designed to help lawyers review and draft documents with use cases
(00:30):
in mergers and acquisitions and class action litigation.
Hi Thomas, how are you?
Great, thanks for having me.
Tell us about your background and the genesis of Vecflow.
I'm co-founder and CPO of Vecflow.
Prior to founding Vecflow, I worked in a very small litigation from in Alaska.
I've also done some ML research at Columbia.
So it was in some sense, no brain or starting Vecflow in around 2022 as large language
(00:54):
models were developing and as law firms were right for disruption.
I think our initial thesis at Vecflow was around needing law firms to adopt AI and to adopt
load language models.
We very quickly recognized that we needed to build end workflow style solutions to automate
more robust task at law firms like a document review and drafting.
(01:17):
And we currently have top use cases as you mentioned in M&A in class action.
I think what we're really focused on is we're writing both a general purpose AI
solution for every line in being able to automate and one click workflows around the law firm
as well.
How do you see the role of AI expanding in legal workflows?
In 2024, you saw a sort of expansion of the role of AI coming from chat based applications
(01:42):
to more fully workflow style solutions for law firms.
So you're going to continue to see if the beginning of 2025 is any indication more
for reasoning based models working at law firms as well as more AI integration within
the particular applications in many different practice areas.
For instance, if you're in M&A firm, I think that to use Oliver in 2025 means to be able
(02:07):
to draft fully forstrafts of like share purchase agreements, LPAs, whatever other type
of document that you're working with just from a term sheet.
So we'll be able to get your first draft much more quickly there.
Where law firms implementing Oliver?
The firms we're working with really think about this as less as a product and more as a
vast set of their firm here in terms of in terms of where on the firm we are used were
(02:31):
used particularly in the drafting and review stage.
One of their classic examples is like due diligence.
You can review with Oliver over a thousand different documents and you can easily analyze
for like change of control provisions and then you can use that information when doing
later steps down the line as well.
I think in terms of actual implementation details, we have many different kinds of deployments
(02:54):
at firms so you can either deploy in our cloud in our private cloud, you can also deploy
in a VPC deployment at a firm and you can as well do it on-prem.
So with our partnership with meta, we can deploy our AI always on-premise for law firms
as well.
And this allows firms to use confidential information within their product and satisfy all firewalls
(03:17):
and content security measures there with their clients and so they can better provide value
for their clients at the end of the day.
How do you approach the onboarding and training process?
We've spent a lot of time over the past few months really perfecting our onboarding process.
So I think like we know that it can be a steep learning curve when it comes to using
(03:38):
AI in law and so we do both kind of an onboarding session at the beginning.
We also do many different workshops, weekly workshops.
We have the first like 60 minutes there focusing on a tutorial about how lawyers can use our
product for particular tasks, whether that is in like due diligence, drafting, writing
of complaints or just basically legal research with SEC Edgar.
(04:02):
We then have a tutorial there.
We switch the tutorial to be a sort of side by side workflow, right?
So we work side by side as in person as we can be with the power users of our tool to get
them to be more power users.
And then of course the power users start to increase their own productivity and talk to
other lawyers down the line about our product as well.
(04:24):
So I think like it's a real comprehensive process of onboarding.
We also try to track as many metrics usage and accuracy as we can and we try to get that
to be as visible as possible to as many people at the firm as we can.
So it's really easy to measure the value both of a highlight of the Oliver platform, but
also of a sort of like holistic overview of how the firm is spinning their own time because
(04:49):
oftentimes firms don't even have a good understanding of where they're spinning the time initially.
How does Vectlo minimize the risk of hallucinations?
This is the top concern of a lot of lawyers and a lot of firms.
I think that the first thing I mentioned everyone is hallucination is a making up of fake information.
Accuracy is something different.
(05:10):
That's like how many questions can you accurately or how many tasks you can accurately respond
to.
We at Vectlo have spent a lot of time perfecting a solution to this and we have very low
risks of hallucination in general.
Every time you ask a question, we make sure it's validated with source information and
we give that source information precisely to a user.
(05:31):
We also have very strict guard rails.
So for instance, every response that you ask a model, we put it through over a hundred
different other models to verify that it's the right answer.
And finally, we rely upon asking the lawyer to clarify any ambiguous questions.
So rather than going through a particular task and then at the end having your associate
(05:54):
draft you the wrong document, we try to have our system Oliver asks the lawyer, is this
the right pathway?
Do you want to do X or Y?
I think this really helps keep the lawyer in the loop, but it also helps educate Oliver
on what type of task it's doing going down the line.
And finally, we recently had an email from a new client of ours that mentioned, look,
(06:14):
rethink Oliver hallucinated on this particular aspect.
And in fact, what was going on is that he was reviewing a 6,000 page document and on one
page there was a cursive handwriting on a colored piece of paper.
And our lawyers couldn't read it, but our system was able to extract the text and get the
detail that their client actually did something bad there.
So I think that definitely in combination with a lawyer and a tool, you can really see a lot
(06:39):
of value coming out of this.
What are some best practices for evaluating different AI products?
Evaluation is extremely hard.
It comes up every time when I talk to a firm.
It's almost associated with the market mapping problem.
It's been really hard to map how products compare to others in the past few years.
(06:59):
And I think the law firm is definitely don't have the number of people to accurately evaluate
every product on the market.
And so you're going to need to look outside of just the product and the pitches that you
are being seen.
There's a number of third party studies.
Valse is in the right direction.
As features are developing very fast, you also have to look at the team that's behind the
product and the platform.
(07:20):
There are some teams and products that we know are very good, maybe in the past, but probably
don't have the expertise or the technical background in order to build state of the art
products going into 2025 and going into 2026.
There.
How do you see the use of legal technology in law firms changing in 2025?
(07:41):
Thirty years ago, you had a shift towards digitization and having legal research beyond
the line.
We know that transformed firms in many different ways that kind of equalized a lot of firms
that were able to do legal research across different practice areas.
But it did not really change the underlying business model of the firm in 2025.
(08:01):
And we're going to see and we have already seen a shift from chat as a model to workflow
as a model.
You're going to see ever more indications that this workflow based approach really promises
to bring about underlying business change to a firm.
We have some of our customers who were previously dealing with 200 of one type of case a month
(08:24):
and now can easily deal with 500 of one case a month with the same amount of people.
When you do that, you can both charge more and change away from available model because
I think in the day, if you're doing 200 or 500 cases a year, it can be a push button workflow
that you can leverage your own internal information.
So if 30 years ago the change was about how public information was dealt with public case
(08:47):
law, I think you're going to see a change coming up about how private information is dealt
with at a law firm.
This is Ari Kaplan speaking with Thomas Bueller, the co-founder and chief product officer
at VEC Flow, a legal AI platform designed to help lawyers review and draft documents
with use cases in M&A and class action litigation.
(09:09):
Thomas, thank you so very much.
Thank you Ari, wonderful talking.
Thank you for listening to the reinventing professionals podcast.
Visit reinventingprofessionals.com or Ari Kaplan advisors.com to learn more.
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.