Kunal Chopra is the CEO of Certivo, an AI-powered compliance system of record for global manufacturers.
A compliance manager at a manufacturer asks a supplier for evidence that a component meets a regulation. The reply arrives as a scanned PDF, a test report photographed on a phone, a bill of materials in a spreadsheet nobody else uses and an email that says the material hasn’t changed since last year.
The information is all there, but the manufacturer’s software can’t use any of it until a person reads each document, decides what matters and types values into the right fields.
I’ve spent the last several years working with manufacturers on product compliance, and this scene repeats at nearly every company I’ve visited. Capable people spend most of their time retyping information that already exists elsewhere.
AI-native software can alleviate these challenges, specifically through the growing trend of contextual intelligence, but this will also challenge the way work gets done.
Traditional software needs structure first.
Every system of record built over the last 50 years shares one assumption: Data must be structured before the software can make sense of it.
Businesses don’t produce information that way. MIT Sloan cites estimates that 80% to 90% of enterprise data is unstructured. Suppliers, labs and engineers send what their own systems produce, and much of what matters lives in email.
So, organizations pay people to translate and to build supplier portals and data-cleansing projects to push the structuring work upstream. However, this leads to repeated challenges: records lagging the documents they came from, two people reading the same certificate produce different entries and unclear answers when an auditor asks why a value was recorded.
AI-native software takes data as it comes.
AI-native systems can now accept information in any format from any source, whether a PDF, a spreadsheet or an email thread, and make sense of it without requiring anyone to reformat it first. (Disclosure: My company, Certivo, is one of several developing AI-native software to manage unstructured data.)
Earlier extraction tools depended on templates and broke when a layout changed. AI-native systems handle variation the way a knowledgeable person does, by understanding what the document is, what it’s for and how it relates to everything else the organization knows.
This is referred to as contextual intelligence. A contextually intelligent system extracts what the task requires rather than everything on the page.
In compliance, that means recognizing that a product diagram, a supplier email and a bill of materials each carry specific information, and knowing which regulations it must be matched against. It can also mean linking a product number in a drawing, a part number on a supplier declaration and a supplier ID in purchasing into a single view of the product, where traditional software sees three unrelated strings.
The same principle is reshaping contract review, claims adjudication and customer due diligence, wherever the source material is documents and the output has to be a defensible determination.
Considerations For Contextual Intelligence
Reading unstructured input doesn’t lower the standard for the output. A concentration figure pulled from a test report still has to be completely accurate, traceable to the page it came from and repeatable, so matching it against regulatory thresholds should run on deterministic pattern-matching rather than probabilistic guesswork.
The same document should produce the same determination every time, with its evidence attached, which keeps unstructured input auditable and defensible.
This is also where the trade-offs live.
A system that reads anything can’t recover a declaration a supplier never sent. Extraction quality varies with how unusual a document is, so early deployments need people checking the system’s work until it has earned trust. And an AI-native layer feeds the ERP or PLM underneath it rather than replacing them, so the integration work remains.
While investigating AI-native software, companies should ask four questions of any system: Can it show where every extracted value came from? Does the same document produce the same answer twice? What does it do with a document type it has never seen? When it’s wrong, who finds out and how quickly?
And if the inputs are already structured and stable, a traditional system may be the better fit.
Workflows, workloads and speed all change.
This is a dramatic shift in how people work with systems, how much work there is and how quickly results arrive. Teams no longer need clean rows and columns before they can start. They load what they have and the system builds the picture with context far faster than manual structuring.
Once the system has assembled what it can, it reports what is missing. A component with no supporting declaration or a test report covering the wrong revision surfaces as a list item rather than as a discovery made during an audit. The human’s job becomes closing gaps the system has identified, which requires judgment, rather than keying in values.
People whose expertise was knowing where to find things need new roles, and the gap list only helps once the team has validated it enough to trust it. When results arrive in days instead of months, approval processes built for the old pace become the bottleneck, and leadership has to decide which determinations the system can act on alone.
That is what makes automatic compliance a realistic operating posture. A compliance system of record that reads evidence in whatever form it arrives, maps it to obligations and flags what’s absent can maintain defensible due diligence continuously.
Because it knows exactly what is missing and from whom, it can also request the declaration, chase the supplier and route the exception, so the system of record becomes a system of action, within whatever boundaries the organization sets.
Contextual intelligence is the near-term payoff.
Whether artificial general intelligence is months or years away is a debate I’ll leave to others. Contextual intelligence, however, is one way that AI can make life easier and faster for the people who used to be the bridge between messy information and structured systems.
Messy data isn’t going away. What changes is where the structuring happens. It moves from the human’s desk into the system, and the human moves to the questions that need judgment.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?