Caroline Ramackers (17s): Hello, everybody, and welcome back to the Orange Chair Podcast. Today’s episode is part 2 of our AI and the B2B space discussion. Joining us once again to discuss these topics is Jim Wanner, CEO of Keymark, and special guest Anthony Macciola. An expert in the automation industry.
Now, without further adieu, let’s jump back in.
Jim Wanner (39s): So what do you think the next generation of AI-driven IDP services look like? You seem like you have a good vision for this right now.
Intelligent Document Processing
Anthony Macciola (48s): Yeah. Yeah, it’s gonna be interesting to watch how it evolves. When I talk about it, I like to talk more about kind of the art of possible as opposed to what the specific technology is, right? So if you think about the art of the possible and you think about the problems you’re trying to solve today in the IDP market, most people have figured out that processing of fixed forms should be pretty simple to do today. Right.
Jim Wanner (1m 23s): Pretty straightforward at this point.
Anthony Macciola (1m 24s): It should be. Right? I think it’s gonna get even easier whether or not that makes a big impact in the market. Who knows? Right? You don’t have to spend a lot of time and effort today setting up a fixed form.
I think in the next two, three quarters, there’ll be no setup for a fixed form. It will just be something that happens dynamically. I do know for sure. For semi-structured documents, like an invoice, or a bill of lading, or an explanation of benefits or utility bill — the most common kind of semi-structured documents that you come across today that probably represent 80 to 90 percent of the types of semi-structured documents — models will emerge.
That makes the need for setting up, configuring, training, labeling, that’s just gonna go away. And you’re gonna be able to get out of the box F1 scores, for people who know what F1 scores mean, in the low to mid-nineties, and then if you want to extend the models, fine-tune the models, you’ll have tools that will allow you to quickly and easily do that. But they will
Jim Wanner (2m 40s): And if you didn’t want to, you could put a human in the loop right there, and you’re fine. You get a return on investment, from day one.
Anthony Macciola (2m 48s): Yes. So the time to value on processing structured forms and semi-structured forms is just gonna collapse. It’s gonna be much, much shorter.
Jim Wanner (2m 59s): So then the hype really about this is translating this ability to deploy quickly and accurately. And if you need a human in the loop, you put a human in the loop. If you don’t, if you want to get a higher accuracy, you can do that. But that’s true tangible value you can get from AI today. Would you agree?
Anthony Macciola (3m 17): Absolutely. I think that will be one of the biggest impacts when we fast forward 18 months and look back, and people ask or they start to realize what the biggest impact AI had in the IDP market, I believe the biggest impact will be on the total cost of deploying a solution and the time to value.
I think it will change how vendors sell IDP solutions and how buyers expect to be sold. If you think about looking over the last 18 months, you engaged an opportunity, you qualified on an opportunity, you defined the proof of concept, you set up a proof of concept, and that was a big long drawn out sales process, right? I’m seeing engagement scenarios change where you know what? On my first sales call or my second sales call, give me your documents. I’m gonna process them for you in real-time as part of my sales engagement.
Jim Wanner (4m 22s): Boy, is that exciting?
Anthony Macciola (4m 23s): Yeah. And what does it do from the standpoint of minimizing the angst of the customer, right? I don’t have to wait four months to see what you’re gonna be able to do before I even sign a contract or even talk about pricing, I’m gonna be able to see what you’re gonna do, and it’s gonna be so far down the road from a confidence level factor that would probably be good enough for me to do the deal, right? But I know you’re gonna make it even better than that. So it just changes the whole dynamics of the sales engagement process. Which for me, is really exciting.
Jim Wanner (5m 4s): So if we’re rolling out a system, obviously, it has to produce results. It has to be a B2B to handle mission-critical applications, possibly life and death scenarios in the medical field, there’s going to be significant risks. There’s going to be some challenges associated with those things. What should the end user be looking for to make sure that they know how to deploy this and apply these technologies in this mission-critical scale that we’re talking about today?
Anthony Macciola (5m 33s): Well, all of you, we haven’t referenced it, maybe this is the first time you referenced, right, all this as soon as you have the right plumbing, right, you’ve got the infrastructure and the backbone in where you can employ or insert these new AI services, right? So system interconnectivity, the workflow, whether it’s human in the loop workflow or automated workflow. Right? The process analysis or data analysis to assume or to be able to measure the productivity enhancements that you’re actually realizing, right? All of that infrastructure, I’ve been assuming that it’s in place. If it’s not, that actually becomes the precursor to getting all these advancements on AI. Right? Because what AI does is it just makes that infrastructure more effective and more efficient. So, and then understanding from an industry domain perspective. Right?
Jim Wanner (6m 43s): You have to you have to know the industry.
Anthony Macciola (6m 44s): Yes. And then the ability, I’m kind of jumping threads here, but the ability to measure your productivity gains, and to measure your profit or process gains, right, from an efficiency standpoint, it kind of assumes that you understand what your baseline true metrics are. So there’s an aspect of process intelligence, data intelligence, that is assumed in all of this. So, you know what your baseline is. You know how to measure your progress, and you know how to tell whether or not you’re getting the gains that you expected.
I think a lot of that basic capability is so important here. And whether it’s your BPM platform, your ECM platform, your RPA platform, whichever system of record you’re deploying and basing this on is really critical and making sure those backbones are ready to accept these kinds of plugins and services. I think Gartner’s referred to it as a composable application framework, right? Being able to bring best-of-breed components together to solve the problem at hand.
Jim Wanner (8m 19s): Very important to have the infrastructure in place. Most people should have it with their larger organization. Would you agree with that? Or…
Anthony Macciola (8m 27s): I would have thought at an enterprise level, but most people have made those bets today and have their preferred vendor list, right? What will be interesting as this stuff gets easier to deploy, it wants to go down market, mid-market, SMB, right? I would think that SMB penetration for IDP at scale has been a little hampered because of those upfront entry costs, it’ll be interesting to see what happens as this becomes easier and more cost-effective to see if it goes down market and opportunities that may not have been worth pursuing before now become viable, it’ll be interesting to see what kind of infrastructure and environments exist in those environments or whether it’s a completely different infrastructure ecosystem than the ones we’re used to in the enterprise space.
Identifying opportunities for AI
Jim Wanner (9m 28s): So I always think about this, when we’re ever trying to solve a problem, you’re thinking of vertical solutions, you’re thinking of applications, you’re really thinking of low-hanging fruit, and you’re trying to find those that have the best return on investment for the customers. And I always joke that somebody’s buying something because they’re getting paid, they’re paying somebody, or there’s a compliance issue associated with it. So when you’re looking out there, what are the real practical steps to look at whatever you’re doing today, identifying those areas that are right for opportunity and making sure you have the steps in place to implement these systems.
Anthony Macciola (10m 4s): Yeah. I think a lot of it comes back to driving efficiency within your organization, right? The environment’s just getting more and more competitive out there. And I think more and more it’s something that I was focused on at both ABBYY and Kofax, and it’s been the backbone of the digital transformation market, right? People have been focused on redesigning their critical business processes, but back when they started, how those processes function was a byproduct of the infrastructure they deployed. So customers had to deal with organizations based on the way the infrastructure of the organizations wanted.
Well, the whole digital transformation phase was just turned upside down on its head, right? You wanna redesign your business processes based on creating the most optimal customer experience because today, customer experience is what either allows you to retain your customers or what allows you to acquire new customers, and companies that don’t have good customer experiences aren’t gonna be around for very long.
Jim Wanner (11m 21s): True.
Anthony Macciola (11m 22s): So everything that’s being done back-office and even front-office through customer experience automation and things like that, I think today is being driven by creating optimal experiences. So when you’re deploying these new AI capabilities in IDP. How does that benefit you? Right? It allows you to create better customer experiences. What’s a typical example? Someone’s applying for a mortgage, right, and I needed your w2 from last year the year before.
Well, maybe 18 months ago, you’d email me a PDF of your W2s and someone would get around to looking at it a day or two later and then they’d realize, oh, no, one of them is the wrong year. And they send you a text or an email, and you’d get it and get around to it.
So something that was simple to fix took like four or five days of elapsed time, right? If you’re using this new stuff, and I can pretty much understand any document you’re sending me, and I can tell you that the W2 that you sent me wasn’t from 2022, it was from 2020, and I can do that quickly and turn it around and enable you with the right structure to, you know, within 10-seconds, the guy sending it to you, go back to him and say, “one of those is wrong.” Right? It’s a lot easier for him to rectify that right now. And with the system help, rectify that, it just creates an overall better efficient, more efficient experience, right? Everything kinda wants to become real-time, whether or not it’s at setup or whether it’s at run-time.
I think a lot of these new advancements and the performance around a lot of these new AI-driven capabilities will allow customers or partners and customers to create those kinds of environments.
Jim Wanner (13m 33s): So for dealing with data like you just described, that needs to be very secure. So do we have any ethical issues associated with using AI in an IDP type of marketplace?
Ethical Considerations
Anthony Macciola (13m 44s): Yeah. I think we I think we get to dodge a bullet. The applications of AI in the IDP market are very narrow, right? It’s to help me better understand my content. As opposed to GPT and generative AI, there’s all sorts of ethical issues around that, right? We’re using it in such a narrow scope, and the results of what we use it for can be validated, verified, confirmed, and secured, right? My prediction is that the use of AI, regardless of the type of technology, within the IDP market, assuming you’re not popping this stuff out into the cloud somewhere, I think the ethics and the concerns around the use of that technology will be a fraction of what we’re hearing in the general market. I don’t think Congress is gonna be trying to implement AI governance and criteria for the use of multimodal transformers around processing an invoice.
Jim Wanner (15m): I would love to sit down on that subcommittee here.
Anthony Macciola (15m 4s): It would be good entertainment.
Jim Wanner (15m 6s): It would be ex… yeah.
Anthony Macciola (15m 7s): Yeah. So I think we’re somewhat removed or insulated from a lot of the ethical issues that are likely to arise from AI, again, assuming that you’re not popping your data into the general-purpose cloud.
Choosing vendors
Jim Wanner (15m 27s): And then how do, I guess, if we’re going through all this process, we’re trying to figure out who we’re gonna work with, how do the technology partners play a role in this process? And if I’m an organization, how do I look at the landscape and make sure that I’m picking the right vendor? How do I bring them in-house? What what would you recommend to somebody out there?
Anthony Macciola (15s 48m):
You know, I think if I was a buyer right now, a customer, I’d be a little apprehensive because it’s hard to separate hype from real tangible value. Right? And there are people doing demos out there with GPT that look really cool. Right. And I want to make sure I don’t come across like a GPT hater. It is fantastic, and it is going to solve some massive problems out there and enable a whole new set of solutions out there that it could, I’m just saying specifically as it relates to documents and IDP, some of the things I mentioned before, like predictability, like training ability, like security and privacy, right? I didn’t mention it about performance.
You know, we get asked to run enormous volumes of content through our systems on a daily basis, or we get asked for very, very low latency, transaction processing times, and it’s kinda counterproductive with going to the cloud. So I think not letting the hype get the most out of you and realizing what your fundamental challenges are and the things that are important to deploying solutions today in a real-world environment, those pillars don’t get to go away. They have to stay in place. And as you’re applying AI and looking at different approaches to AI, do that sanity check.
If using this approach doesn’t allow me to measure my results, and maintain predictability, and determine when I need to pop a human in the loop. If I pop a human in the loop, does it allow me to take the results of their review and better my models? Does this AI approach allow me to maintain the integrity and the privacy of the data that I’m processing? If the answers to any of those questions are no, then I question the approach you’re pursuing.
Jim Wanner (18m 23s): Got it. Anthony, that is a great way to end this. We very much appreciate your time. It’s always a pleasure to speak with you, my friend. And I think that, you know, this is one of these types of podcasts that I hope that more and more people listen to because if you are looking to deploy mission-critical applications.
Anthony is a wealth of knowledge and the information he shared with you today you should not take lightly. So with that, we’ll sign off and we appreciate you, Anthony, very, very much. Thanks, everybody.
Caroline Ramackers (18m 55s): Thank you for joining us for this episode of the Orange Chair Podcast. This podcast is brought to you by Keymark Inc. Experts in automation software for over a quarter of a century. Never miss an episode by subscribing to our channels wherever you listen to podcasts.
You can also find us on Instagram at the Orange Chair Podcast. For more information, please visit our website at keymarkinc.com. That’s keymarkinc.com. Until next time.
