Intelligent document processing (IDP) provides the foundation for intelligent automation. While IDP isn’t a new technology; the capabilities have spiked dramatically in the past few years.
I sat down with Doculabs’ IDP hero, Richard Medina, Co-Founder and Principal Consultant, for our second Two for Tuesday.
Here are today’s two questions:
How do you become an IDP hero?
There are a lot of new vendors and technologies – it's amazing but what questions should I ask vendors to ensure they’re a good fit?
You can watch/listen to the conversation below or scroll down for a lightly edited transcript.
3 Key Takeaways:
Getting buy-in requires showing concrete benefits like productivity, cost reduction, and business KPI improvements
IT, operations, and business units all have different motivations for adopting IDP
Ask vendors key questions like ability to innovate, productization, ease of use, and precision of outputs
Rich explains how to become an "IDP hero" by successfully introducing intelligent document processing (IDP) into your organization.
He notes advanced AI/ML allows classifying, extracting, and validating unstructured data like documents with far less human involvement.
You should become familiar with IDP, then build a business case focused on three constituencies - IT, operations, and business units. Each has different motivations around accuracy, automation, costs, and key performance indicators.
Concrete before-and-after data is persuasive. Vendors should be vetted on innovation, productization, ease of use, and output precision. Handling disruption and false positives are adoption challenges. But IDP's benefits make it a compelling technology.
Marty: How do you become an IDP hero at your company?
Richard: So introduce IDP – intelligent document processing – into your organization. You may have heard about intelligent document processing. What it is, is the introduction and use of artificial intelligence and machine learning for handling documents that might start out as paper or might be images or might be emails. The idea is we use artificial intelligence and machine learning to
help classify what documents they are,
all the relevant information that you need from them,
what you got for completeness and correctness,
and then hand it over to the downstream lines of business and systems that use that information.
What's new about IDP as opposed to the 1990s or even as far back as 2022 are the advances that have occurred in artificial intelligence and machine learning which make formerly intractable documents identifiable, reduce unidentifiable documents that you would have to use humans for, and go straight through. There are some disadvantages to it, and it's still a very young technology, but we can talk about that later. Right now we're talking about the good stuff.
So job number one is to get familiar with it.
Number two is to identify your organization and the business case for IDP – there are typically three groups you are looking at for achieving value:
IT – it might be a center of excellence for capture or might just be an IT group.
Capture operations – if you have one like a scanning or document capture organization.
The actual lines of business.
Take an example of a bank – it might be loan origination. If you look at IT, the primary benefits you should be delving into for this new technology are: first, to really attain high levels of accuracy and automation that were not achievable before with old–style OCR and so on for classification, extraction, and validation. The second is so you can push through more documents per hour and that means less labor – fewer people can do more.
The second is to address changed and new documents – you're always getting new document types and even with AI you can use it to know those new documents, but it takes what's called training. But the newest technologies let you automate that so you don't have to know the new document types coming in – it can learn those and identify them. And the third benefit is to shorten the development cycle and reduce development costs – if you want to bring out new workstreams or make modifications, it's low–code, no–code.
You might be asking, hasn't it always been low–code, no–code for the last five years? And the answer is no, this is pretty new stuff.
The second group for the business case is capture operations. You might have a departmental or enterprise capture center doing hundreds of thousands or millions of pages per year. They are very cost–margin conscious and using this technology you can dramatically reduce your technology costs by having fewer vendors. Secondly, reduce your labor costs so you don't have to do document prep and data entry. Third you can take on more workstreams and volume with the same staff. And last, you can meet your customer SLAs and make them happy – both your internal lines of business and external customers.
Marty: So when you go to these groups of people, you know what you're saying – how do you get them to buy/see the value?
Rich: Good way to show it is what you can show – compare the current way, business as usual, and what you're doing, how you blow through reams of documents, how many documents per hour you can process, how long it takes you to process each document. Then you can extrapolate to show how slow it is, how much it costs, and so on. That's the easiest way to make the business case.
But it gets more interesting for the lines of business. Rather than IT where you're talking about developing software and solutions, or capture which is using the solutions, you've got the lines of business where you're taking advantage of the solutions. They are the customer of IDP.
The best way we've found, though it is more complex, is to go to different business users and understand their objectives, business objectives, and key performance indicators. For loan origination for example, it's not just labor and technology costs, it's KPIs like improve pull–through rate, abandoned loan rate, or application approval rate – how many loans per time do they get, how many do they lose when customers leave, how much time does the bank waste on false positives?
If you can address those, you'll make lines of business happy. You understand at a superficial level their processes, then reduce cycle time for each. In loans it's the application, then loan approval, closing, funding – shorten each stage and the overall cycle time, improve KPIs. How? Digitize and automate with IDP to reduce errors and rework, data entry, QA/QC – go straight through highly accurately. Reduce status queries and latencies. You can do all this with IDP.
Marty: Alright Richard, I want to be an IDP hero but there are a lot of new vendors and technologies – it's amazing but what questions should I ask vendors to ensure they’re a good fit?
Rich: It's not just IDP, it's part of broader AI and now generative AI. Folks think of it as magic or dangerous magic. But there are four key questions to ask that aren't obvious.
First, a bit obvious – ask your vendor if they can innovate. Are you competitive in IDP? Can you develop these advanced AI/ML capabilities? It's not dumb – IDP AI advances so quickly most vendors use older algorithms from pre–2022. They can't keep up. They need to partner with cloud vendors or others. If they don't innovate now, is it on their roadmap? Are they including large language models?
Second – productization. Are you a real vendor or just two folks in a garage? Can you successfully productize, market, sell, deliver, and support these capabilities? Like Tesla – you need the workflow, software, services around the AI engine. Many vendors can't do this end–to–end.
Third – can we successfully use IDP? How disruptive is it moving from legacy capture to this? Is it sustainable, can we get ROI? That leads to the fourth issue...
The problem with large language models is precision. They aren't precise, they confidently give wrong answers. Like a menu for a policy number. These false positives must be reduced when dealing with dollars and regulated data. It takes so much verification and QC that it may not make business sense.
So those are four key questions to ask vendors to become an IDP hero for your organization.
Marty: Alright Richard, as you can tell, you are an IDP hero here at Doculabs! I'd like to thank everyone for tuning in this week and thanks a lot for your time as well.
Rich: What? Thanks!