Rude? Perhaps. True? Definitely.
But let’s face it, there are a lot of B2B technology solutions that are hard to grok let alone implement successfully. That CRM system you’re exploring? Functionally complex, and plenty of ways to get it wrong. Automated, multi-channel customer service application? Watch out, or you’ll be under a deluge of angry Twitter missives. But intelligent document processing? That’s a completely different animal, and the reality is that organizations and the vendors themselves have been getting it wrong for years.
“How can that be?” you ask. The reason is that it’s complex stuff when you’re after something that is supposed to deliver significant amounts of highly-accurate data from documents. And the reality is that most document automation systems simply, improved the “process” word without actually delivering on the “intelligent” word because process improvement is always easier than optimization of a complex system.
The result was still some pretty good cost reduction by just automating data entry. The catch is that a lot of that data entry was and is still incorrect. If you’re thinking to yourself, “yeah, but we’ve been successfully using document capture software for over a decade.” I’m talking to you too. Would you buy a self-driving car expecting to have to review and approve every action it makes? No way. But that’s what you’re doing with your current software, whether its an IDP solution or you’ve created your own solution using various OCR toolkits.
“Yeah, but we sample our data to ensure accuracy,” you say. Really? How do you go about that? How many samples and what confidence interval do you use? Yeah, I thought so. If you don’t have experts on staff that know how to apply data science to deliver a highly-optimized system, you’ve got a problem even if you don’t know it. Most systems we’ve had the pleasure of analyzing deliver on average, 10% more inaccurate data than the organization believes it is getting, and provide about 20% less automation overall. Yikes!
“We had our vendor configure the system, so we didn’t need to become experts. Take that!” you say. Pssst. I have something you need to know – your vendor didn’t deliver what you think they delivered, and the reality is that the process was probably so painful you were happy to get out of it with some of your budget left intact. How do I know? Because we have talked with IDP vendors (we license our software to many of them) and, when it comes to delivering highly-optimized systems, they just don’t know how to do it. We’ve spoken with their implementation specialists, and they don’t know the difference between a confidence score and a conscientiousness score.
“Our vendor is using deep learning, so that has to deliver, right?” some may ask. Wrong again. Don’t get me wrong, deep learning in the right hands has powerful impact. Look at GPT3 and the offshoots. Look at Alpha Go. Deep learning has the chance to do some pretty magical things. When Parascript started using a variety of deep learning neural networks for handwriting recognition, the results were astounding. But there’s a catch – you have to know what you’re doing (aka “are you a data scientist that knows how these systems work?”) and, you have to have copious amounts of data. When we talk with organizations, they can barely scrape together a few thousand high quality tagged samples, let alone 100,000 or 1,000,000. Did your vendor tell you that?
Ok, so I’m having a bit of fun poking my finger in the eye of the IDP industry, the analysts, and all the frothy marketing claims that surround it. But the reality is still the same: you need to really know your stuff to implement IDP successfully, and few organizations are prepared. There is a lot of hype out there; so much so that it’s hard to tell a fact from fiction. Trust with a vendor has to be earned by delivering real results, not just fancy marketing claims.
But help is here. We’ve spend a lot of time working on materials aimed to get you up-to-speed, to be more conversant on IDP topics, and we even provide practical guides on how to approach an automation project that involves documents. We are confident that, after you educate yourself, you’ll know more than many analysts offer advice on the topic. We hope you find it useful.
Go here to take your first step towards #Insightful IDP.