Proprietary data: the asset AI cannot copy

How it makes money

AI lowers the cost of copying a product's features. A competent team can now build a passable chatbot or summarizer in weeks. What it cannot build is the record underneath: decades of insurance claims, a library of licensed images, a history of transactions. A model trained on everyone's data gives everyone the same answer. A model fed data only you hold gives a different answer.

So the business model choice that matters here is the moat: why a rival cannot copy you next year. When software is cheap to copy, the answer moves to what the software runs on.

Data at the core

Verisk, which supplies data and analytics to insurers, heads the description of its strengths "Proprietary Data Assets – Data is at the core of what we do." It collects and analyzes "billions of records" and builds its products into clients' workflows. Now it puts AI on top: it has solutions that "leverage AI and generative AI to automatically extract unstructured data from and summarize medical records" during claim processing. Revenues were $3,072.7 million in 2025, up 6.6%.

Trained on its own

Getty Images turned its library into an AI product. "In the fall of 2023, we launched Generative AI by Getty Images," the company writes, a service "designed to be a commercially safe AI image generation service that is trained with Getty Images creative content." The selling point is the data: images generated from a library the company owns, rather than from pictures gathered from the open web.

How it fails

Proprietary data stops being a moat when it stops being proprietary: licensed to the model makers for a fee, leaked, or matched by a public source that becomes good enough. It also fails when the data is valuable but the company cannot turn it into a product customers prefer to the alternatives.

Questions

What is proprietary data?

Information a company holds that no rival can buy, license or gather for itself, usually collected over years in the course of its business, such as Verisk's insurance records or Getty Images' photo library.

Why does proprietary data matter more with AI?

AI makes features and software cheap to copy, so the data a product runs on becomes the part a rival cannot reproduce. Verisk and Getty Images both now build generative AI products on data they own.

Should a company license its data to AI companies?

Licensing brings revenue now but can help a rival's model later; keeping the data means building AI products on it yourself. The answer turns on how hard the data would be for a competitor to gather or replace.