E01 – Transcript

Alright. Welcome, everyone to the launch of Keymark's new podcast. I am Ed Mcquiston, Keymark's new CEO started back in June.

And I've got with me Colin Toomey.

Yeah, Thanks, man.

Thanks for having me.

Absolutely.

So Colin leads our our government sector.

We're actually recording this at the Riverbanks Zoo in Columbia, SC, where we just held our government summit today and had some great conversations about different technologies.

But what I'm here to talk to you about today with Colin is IDP.

And IDP is stands for Intelligent Document Processing.

And it's a huge part of the AI conversation and ways to surface unstructured data and turn it into something usable for your AI enablement that didn't exist previously.

And Colin, I've been in the spirit of the zoo.

I've been referring to IDP as a keystone technology.

I I went on a safari once and I learned that a rhino, which there's some right back there, is considered a keystone animal.

They're an animal that a whole ecosystem exists because of that animal.

The oxpecker you see on the back, the bugs, the oxpecker eats, all those things are because the rhino exists.

If the rhino are extinct, so, so go all the others.

Well, I look at IDP as a keystone technology to this whole AI enablement piece because of its ability to to pull data and surface data.

I wonder, you know, what are your, what are your thoughts on IDP and how it's evolving right now?

I definitely never thought about it in the context of a rhino.

Yep, step one.

But I appreciate that there's definitely keystone.

I mean, when I think about AI enablement and where IDP plays, I just think about you have to, you have to identify what what work is coming in and, and the data associated with that.

So you can drive it down, whether it's to an AI powered search capability or an AI assisted workflow or decision making capability or even a downstream system, you know, whatever it may be.

So I think, I think it is the front door.

It's a rhino, it's a keystone, right?

Whatever, you know, foundational piece.

So yeah, I think it's, it's, it's how unstructured data is coming in the door.

And if if you're not getting it there, then you're missing the biggest opportunity you have.

And, and, and as you know, and you know, we both been in this space for a very long time.

IDP is, you know, a, a rebranded name for capture technologies that have been around for a long time.

KeyMark of course has specialized in that capture for a long time.

But what IDP means today is something different in the sense of speed to value and and that's probably the thing that's changed the most and really just the last couple of years.

And you and I saw some interesting demonstrations even today of IDP.

So can you talk about that speed to value piece?

I agree it it's, it's the biggest shift I've seen in like 20 years of our kind of technology.

Like you said it, it was document capture of old and it's it's nothing new is under the sun, but it can get cooler and faster.

You know what I mean?

Like and it really has.

And so, you know, what took months of terms of configuration and set up for a capture process is now demonstratable in minutes production buildable in weeks, you know, and that is monetary savings on the front end, which is makes the project easier to cost justify, which brings ROI like it.

It has so many kind of rolling downstream benefits so that that speed to value.

I I agree is unique.

I mean, back in the day we were building templates and yeah, and trying to find zones and, and using regular expressions and custom code.

And it was a lot of work.

It was all doable.

Yeah.

But you needed a big enough project and a big enough ROI.

And to me, the big shift is that you can now do quick wins, smaller capture projects that you couldn't do or cost justify before because of the speed, the configuration, because of the that that's unique.

And that is kind of the an LLM powered capture capability is unique.

It's not hype.

And to me, it's one of the most proven applications of AI in the document space because capture workflow is fairly structured, right?

Like you're, you're trying to do three things.

And that's why you can apply AI so well to it because you're trying to figure out what is the document type.

Well, and, and where is the end of the beginning of the document.

You know, that's the, that's the separation piece.

The classification is what is the document type.

And the third is what is the data?

And that, you know, when you get into a workflow process and, and using LLMS, that is way broader.

Yeah.

And there's a lot more domain expertise.

But when you get into capture, like it doesn't really care the name of the document.

It doesn't, you know, so I think that's kind of the unique part why I think IDP has taken off and right now, because it's such a perfect use case.

Well, and, and, and I think, you know, to your point, you mentioned kind of this idea of LLM powered.

And you know, one of the things I think in addition to speed, the value that's changed is IDP, you know, previously intelligent capture, whatever it was called, there were, there were financial barriers to entry, but there were also practical barriers to entry.

You know, there were wheelhouse applications.

You think about AP, you think about mortgage, we heard about those all the time, variety of different things.

But there were a lot of things where either to your point, there wasn't the volume to create the ROI associated or it was just deemed too complex.

Maybe it had handprint or, and you know what we're seeing now is that barrier to entry being removed in terms of the types of things, the volume that's needed to justify what you can do.

I mean, to your point, I don't have to go pull thousands of samples to, to train something.

That barrier to entry being removed to me just opens up a, a, a completely different profile.

What you can think about from an IDP standpoint. 100%. A good use case is working with a state agency customer of mine and they've, what they're trying to do is review thousands, which is not a big number when you, when you've been in this space, the longer you don't think thousands is a, there's not enough ROI, but thousands of annual reports.

So they do loans, the kind of government funded loans for certain types of projects, and the borrowers.

They need to annually assess the risk of each borrower and their ability to repay fundamental to their business.

And every year they request every borrower to send annual reports, could be a 10 pager, could be 100 pager, but it's a many pager.

And they have to manually review each one of those thousands and determine risk kind of profile.

And they had to look at 100% of them every year to determine that 2% of them were risky.

And what they, what IDP opens up the door is it can handle that lightweight project.

And in I mean, you can not only demonstrate it in minutes, but like you can deploy that in weeks and get the value.

And it's game changer.

I mean, they're talking about reducing, you know, four external consultants that are doing this kind of risk, financial risk evaluation.

And now their internal people can do that work and they're just getting fed, hey, here's the risky looking, you know, kind of borrower scenarios.

And so, yeah, that that that buried entry is now you can kind of hit the kind of use cases that you could never touch.

Yeah, no, absolutely.

And, and I think so the other thing that is evolving is IDP has traditionally been this very transactional, transactionally oriented, daily transaction flow oriented technology.

So you know, in I've got N number of mortgage documents coming in a day or I've got this number of invoices coming in a day.

Now we're starting to talk about applying IDP to content at rest.

Sure, right.

So I've got information sitting in a repository.

Maybe it's in SharePoint, maybe it's in OnBase, maybe it's in some other content repository and we're talking about being able to unearth information from there.

Can you talk about maybe how so, you know, there's IDP element, but how is that enabling my agentic model downstream or my automation or analytics models downstream?

Sure.

I mean, whether or not we all know this, we all got to say this out loud.

Like AI powered applications are thirsty.

Yeah, and they're thirsty for one thing.

Yep, right.

And it's data.

And, and so when we talk about unstructured content and IDP solutions, those AI-powered applications, they do not want, you know, 3 keywords on a document like they want all the data because that gives them context that gives them, you know, semantics.

It gives them all what they need to help make a decision or a determination for whatever it is.

So that that to me is kind of the the the key concept on why IDP is so important for kind of AI powered applications is because it's enriching those documents with all the additional data.

That was too costly, right.

That was really the rub, because we always knew that we wanted all the data off of that.

Of course, that was not the concern, right?

It just it was too costly. Yes.

So we settled for five.

That's right value, that's right right.

We settled for seven. Yep. You know, and because if you had to key it from image or configure the software, it was just too and now that is just not true.

And and that opens up.

So it's kind of like a perfect storm, right, For sure. You know, on the front end it it it's less costly to acquire the data.

And now we have applications that are AI powered that really want it and are thirsty for it.

Well, and I and I and I like that, that thirsty analogy because I mean that's, you know, we're we're certainly seeing that.

And I think that, you know, as people are looking at the, you know, the criticality of data to enable AI, IDP, a space that you and I know was we operated off in this niche that nobody really kind of understood.

Now, you know, they're tracking something like 450 IDP vendors and, and counting.

And I think one of the things, you know, I, I know I talked about my keynote today, this idea of buyer beware, right?

So the great thing about IDP to your point is it is, it can be, you know, we can demo it in minutes.

And you, you made that clarification.

Production is a different animal, right?

And so one of the things that so, you know, for, for those being indoctrinated, we're going to, you know, a good IDP tool is going to go through a classification process, right?

What kind of document is this?

What is this a mortgage note?

Is it a HUD?

Is it an application?

Is it, and then it's going to separate, is it 5 pages, 12 pages, 13 pages, you know, whatever.

And then and only then once we know what it is and how long it is, then we're going to start, you know, pulling data out of it.

Unfortunately, I think the way it's being demoed out there often is like, you know, you said, hey, I've never seen this before.

It pulls it in and recognize it in a demo.

And that's great.

But when we start to talk about real production use cases, there is this human in the loop concept.

Because as good as the tech is, it's not always right.

And the, the, the downstream, you know, downside of it not being right is hallucinations and our AI bad data and our analytics.

Can you talk about, you know, kind of the human in the loop piece of it and why it's so important?

Yeah, you hit on a lot of good stuff there.

But yes, so the explosion of IDP capabilities is showing up everywhere.

And so now it's not just pureplay vendors, which is where we're traditionally used to.

And, and so you're getting ERP vendors, court case management vendors, you know, medical record systems, right?

Everyone claims management, everyone's getting a lightweight bolt on IDP capability.

And so as, as buyers, as customers out there, the, the challenge is like, well, what's the difference, right?

And you mentioned a key difference.

And I think a key difference is those IDP's maturity and tool set to not just tell me the classification results, grab the data, but to help me perfect it when it ain't right.

Because that's really the rub.

And, and that's what we see as the big separation, the big divide between kind of production mature kind of IDP tools that, that really were kind of backing as a, as an experienced mature partner in this space compared to the, you know, more bolt-on and, and I don't think bolt-on is always bad.

I think invoice processing is an interesting use case because you have a lot of financial core systems and they're bolting it on in a very specific way.

And I think they're going to get that.

I think they're going to get that because it's such a big market and that historically used to be like that was our very niche.

Yeah.

And and so I think, you know, there are going to be some categories with a bolt on might not be bad and it could be in, you know, but you don't want to get out over your skis.

That's right type of thing.

So I agree with that.

So land the plane here.

So or some of the rhino down on the Savannah.

I'm going to go ride the rhino in a minute.

But in, you know, somebody listens to this.

I got to get me some of that IDP.

Sure, right.

What are some steps they should be looking at taking #1 is you've got to, you got to evaluate your current state situation.

And, and so our recommendation to all of our our customers or potential customers is, and what we do is an evaluation, right?

It's it's an assessment, right, of where are you right now?

What tools do you have, what technology you have?

Maybe you have nothing, maybe you have something.

What are your processes?

And that is what we do.

And so I, I would start with that, which is the evaluation, the assessment piece that that is really kind of Ground Zero rather than jumping to, hey, I saw some, you know, some cool AWS textract kind of demonstration that kind of auto categorize.

You're like, great, that might be the right fit.

And we've had times where we recommend that kind of more surgical, tactical, but assessment I think is the first piece.

And then I think it's start small, right?

Yep, keep the scope tight and I think that's gonna help with kind of quick wins and ROI.

That's the and work with, I would say work with partners and not vendors, meaning look for partnership, you know, because this is probably gonna be something critical to powering other downstream exactly.

AI applications. Having a step plan.

And so don't think of this as very siloed because this kind of enabling capability is gonna be useful across multi department across the enterprise too.

And I think the only thing I would add to that is there's a lot of legacy IDP applications that are out there and, and there's an opportunity to do something that wasn't necessarily doable before because of some of the things that we talked about.

You know, I don't want to convert to a newer one because it's going to take so long.

It's going to, you know, cost so much to implement.

It's, you know, it's, I'm paying a lot of money to just process the stuff I've already processing.

And I think what we're seeing is it, it's not going to take a significant amount of time like it used to it, it doesn't require the same level of effort.

It doesn't require the same level of service cost to get it up and running.

And now I can look at expanding the, the use case that I'm using my legacy one to other document types across the organization.

So thank you, Sir, for, for, for joining.

Thanks everyone for watching or listening wherever you are.

And we'll look forward to doing another podcast with you soon.

Take care.