What is due diligence for IDP and why is it important?
Alex Lipinski
Due diligence is the investigative process of vetting an investment or agreement to verify facts and make informed decisions. Good due diligence reduces risk and protects decision-makers from signing off on costly mistakes.
With new intelligent document processing vendors emerging monthly, technology iterating quarterly, and orgs cycling through solutions like flavors of the month, your ability to analyze a market full of showy IDP software and make a determination on whether it’s a right fit for your enterprise is becoming an insanely valuable skill.
Takeaways
- Analysts now track over 450 IDP entrants, a 15% increase year-over-year, marking an essential need for tight decision-making and software selection skills.
- IDP is no longer a primarily back-office thing with 62% of IDP systems now involving external users.
- 66% of new IDP projects are initiated to replace a previous IDP system.
- Proof‑of‑concept evaluations are essential to verify AI accuracy, integration, and security before signing a contract.
Think like an analyst.
A good IDP solution should...
- Read many file types well (scans, PDFs, emails, images).
- Sort documents and pull out fields you care about.
- Understand both tables and long complex documents.
- Let someone review uncertain items before moving on.
- Format data into standard and necessary formats.
- Let you isolate, track, and protect data. Avoid data pooling for training.
Why are there so many IDP solutions all of a sudden?
Traditionally, intelligent capture has been the champion of back-office workflows like mailroom automation, AP/AR invoice processing, and audit prep — activities involving predominantly structured and semi-structured documents. As advancements in machine learning and natural language processing have given way to richer capabilities for mastering unstructured documents, there’s been a shift from back-office use-cases to industry-specific front-office functions.
Today, over 60% of IDP use-cases are in support of processes where external users create, access, and share unstructured documents/data, including customer service, employee onboarding, contract and agreement analysis, claims intake, licenses and permits processing, and beyond. [2]
Combine that with the understanding that 90% of enterprise data is unstructured — and that data quality and data quantity have a dramatic impact on enterprise GenAI results — and the demand for IDP to capture unstructured data climbs higher than ever before.
As such, the IDP market is growing fast, with an increasing rate of 15% year-over-year [1] as new entrants in document intelligence capture capabilities spring out of the woodwork in response to the rising number of use cases, and the growing demands for data to support better GenAI results and data analysis.
Solution red flags to avoid
Deep Analysis, a market analysis and due diligence firm, reports that they’re now tracking 456 companies globally that sell IDP as a standalone product or a feature [1].
The problem for those 456 vendors and for you is that differentiation between IDP products is extremely difficult with product messaging that’s more or less the same. To filter at least a few of the not-so-great solutions out of the decision process, beware of these common red flags.
-
Claims of 99% or “near-perfect accuracy” without proof.
Analysts call out this claim as one of the most misleading in the market because the claim doesn’t tell the whole story. What were the sample documents? Did anyone check if the results were actually correct? -
Unclear Data Policies
Data privacy still ranks among the biggest concerns with AI and IDP. Training a model requires sample documents, but who is providing those? You? If so, how is that data being handled? -
Consumption or token pricing
Uncapped consumption models mean you can go way over budget during document surges. And token pricing is unpredictable and may vary depending on document complexity. Roughly 88% of surveyed IDP purchasers have indicated that they prefer the predictability and stability of fixed pricing models. [2] -
Human-in-the-loop as an upsell
The work is only done if it’s accurate. Classification with the risk of errors, even 1% errors, is dangerous. Human-in-the-loop verification is still a necessity to reach high‑quality outcomes and retrain models safely. To sell it as an add-on is to sell an incomplete solution. Experts advise always placing HITL where accuracy must be guaranteed. -
Really neat demos and UI... and that’s it
Demos are meticulously curated for specific use cases that look great in the demo but fall apart in the real world. PoCs are a must.
The last claim — GenAI as the problem-solver
The effect of GenAI hype and advertising on IDP purchase decisions is apparent. Today, over 66% of new IDP projects are started just to replace old ones that either don’t work as promised or don’t deliver the same GenAI capabilities as promised in the new one. [2]
These replacements coincide with the dramatic rise of GenAI baked into solutions. As of early 2025, over 80% of IDP vendors have advertised GenAI capabilities somewhere in their IDP solutions, with some touting it as the predominant feature [3] even though data quality is the cause for successful GenAI, and not the other way around.
Putting the cart before the horse is a problem. Confusing the cart for the horse is Don Quixote-level insane.
That’s not to paint GenAI a villain. LLMs are fantastic at zero-shot/few-shot learning and summarization. But for raw data extraction at scale and at relatively lower costs, discriminative machine learning is superior.
Basic due diligence questions
Here’s a loose framework for due diligence that combines categories used by analysts to rate vendors with a criteria that aims to pinpoint performance while avoiding red flags.
- Is the solution purpose-built to align with our priority use cases?
- Is there evidence that the solution can handle a wide variety of document types utilizing modern machine learning and natural language processing?
- Can we verify data encryption, access controls, and a clear no‑training‑on‑my‑data policy?
- Is the platform easy to deploy and maintain? Is it predictable and transparent with pricing, with clear visibility into costs over time and usage safeguards? Are we being sold this one solution or an entire platform?
- Does the vendor seem like they know what’s going on? Do they have a history of innovation and a clear roadmap?
- Can we see confidence scores? Track model versioning? Can the vendor show us every step a document takes from upload to storage and what happens to data and processing records afterwards?
Do your homework, and IDP will work for you
Intelligent Document Processing (IDP) has matured but the market is crowded and the 450+ vendors are not all created equal. Case studies are better than demos, and success depends on measurable ROI not GenAI claims. IDP works with data quality, model transparency, and measurable ROI. Ask vendors to prove not just what their models can do, but how they do it, and how they’ll perform over time. In the age of AI everywhere all the time, skepticism is a virtue.
