AI’s Next Raw Material May Be Something We Do Not Yet Call Data
Updated: 1 day ago
Something I came across recently made me look at the AI race from a different angle.
We have long discussed model power, chips, and computing capacity. But another area is becoming increasingly valuable:
Things people know that are not explicitly written anywhere.
Some studies call this ‘tacit data.’
A technician sensing a problem from the sound of a machine. The detail a doctor looks at first in a case. Why an engineer rejects one of two theoretically possible options.
Or why a finance professional looks at a technically correct number and says, ‘Something does not add up here.’
Much of this knowledge was absent from databases until now. Because we did not see most of it as data.
Gradient Ventures’ July 2026 analysis, Everyone’s Watching the Wrong Benchmark, describes this shift through striking examples.
A former Goldman Sachs analyst is paid not to build an LBO model, but to explain how they think through an LBO. Doctors are asked to explain their clinical reasoning, lawyers their approach to evaluation, and senior engineers their decision processes.
AI companies are no longer interested only in the result.
They want to know how it was reached.
You can find the DCF formula online.
But it is much harder to find why an experienced finance professional questions terminal growth for one company, considers cash-generating capacity more critical for another, or examines contractual and financing structures together in project finance.
The formula is written down.
Much of the judgment is not.
I think this is where the development becomes interesting for business.
Until now, digitization has mainly meant moving what we already have into digital form:
transactions, documents, reports, spreadsheets, and data.
AI raises a harder question:
How much expertise that is not yet structured can be made to work digitally?
This is exactly where I see the connection with Ainsideme.
We are not trying to copy a finance professional’s mind or treat personal intuition as correct and transfer it into software.
In cash flow, company valuation, trade and project finance, foreign trade, tax, and risk, we turn established financial theory, professional methodologies, and real-world practice into defined financial model structures.
The distinction matters.
You can tell an AI:
‘Value my company.’
That is possible.
But conducting the valuation in a framework suited to the business model, maintaining consistent assumptions, testing risks through scenarios, and making sensitivities visible is something else.
In cash flow too, merely producing numbers for the next twelve months is not enough.
Collections, payments, working capital, exchange rates, debt service, and financing needs are parts of the same economic structure.
Financing is not only about which source is cheapest. The company or project structure, debt capacity, use of equity, and commercial conditions must be assessed together.
In foreign trade, payment method is not independent of financing, financing is not independent of currency risk, and currency risk is not independent of cash outcomes.
So the part that is difficult to digitize is not merely information.
It is the context in which information is used, the other information it is combined with, and the decision it serves.
That is why I think know-how is gaining renewed value in the AI era.
For a long time, know-how lived in people’s experience.
Some entered procedures.
Some spread across spreadsheets, reports, and ERP systems.
But much of it is still hidden in this sentence:
‘I look at this here, because…’
The AI world is now pursuing that ‘because.’
Gradient’s subsequent analyses point in a similar direction: models are converging on procedural tasks, while differences persist in work requiring substantial context and domain knowledge.
This matters not only to AI companies.
It also raises a strategic question for businesses themselves:
Where does institutional knowledge actually live?
In the ERP?
In reports?
In spreadsheets?
In procedures?
Or still in the working methods a few people developed over many years?
If it is the last, some of the company’s valuable assets are not yet truly digital assets.
That is also what this development means for Ainsideme.
Our goal is not to build a general-purpose AI that thinks in place of finance professionals.
It is to turn the intersections of financial theory, professional methodology, and practical experience into transparent, testable, repeatable financial models.
AI can increase the speed and reach of that structure.
But value may not come only from a more powerful model.
The financial framework within which the model operates, which information it assesses together, and where it needs human judgment may become increasingly important too.
Perhaps one of AI’s next most valuable raw materials is not being newly discovered.
It has been inside companies for years.
We simply did not call it data.
Source: Zach Bratun-Glennon, Gradient Ventures, Everyone’s Watching the Wrong Benchmark, July 9, 2026; Gradient Ventures, Hidden Capabilities Revealed, July 29, 2026.
Comments