If AI collapses the cost of making your product, the question is no longer how cheaply you can make it. It is what stops a competitor making the same thing. That is a trade secret and patent question, and most AI strategies never get to it.

A Bain & Company analysis puts a number on what is at stake. It models US$4.7 trillion of profit moving across the economy between 2025 and 2035. Only about a quarter of that is productivity. The rest is innovation and market share changing hands. The authors’ framing is blunt: “The Internet was a distribution technology. AI is a production technology.” The internet changed how products reached customers. AI changes how the product itself is made – the analysis, the diagnosis, the code, the physical work.

Their prescription is to move early, build data no one else has, and rewire workflows so that “the data you build, the signals you watch, and the recursive improvement that compounds with every interaction become proprietary intelligence.”

Proprietary only if you keep it that way

“Proprietary” is doing a lot of work in that sentence. Data and workflows are not owned in the way a building is owned. They are protected, if at all, as trade secrets – and a trade secret only exists while it is actually secret and you can show the steps you took to keep it so. The moment the profit shift starts, three things test that:

– Vendors and platforms. Your AI workflow runs on someone else’s model, cloud and tooling. Whether the data and the improvements it generates stay yours is decided by the contract terms, not by the fact that you generated them.
– People. The engineers who rewired the workflow can leave. What they carry in their heads is not a trade secret unless the documented process, and the confidentiality obligations around it, say otherwise.
– Copying. Bain’s “Rewired” cluster – pharma, healthcare, aerospace, enterprise software – is where incumbents and new entrants fight over US$1.5 trillion. In those sectors, a genuinely new AI-enabled method of producing a result is often patentable. A patent is the only right that stops a rival who reaches the same method independently.

The strategic move is to decide, for each piece of the new production layer, which protection fits: secrecy for the data and the operating know-how that cannot be reverse-engineered, patents for the methods that can. I set out that choice in Patent or trade secret? Match your IP protection to how long the value lasts. The productivity quarter of Bain’s number is the part every competitor gets too, which is the trap described in Is Your AI Strategy Just a Faster Way to Stand Still?. The other three quarters go to whoever owns something the others cannot copy – the argument in AI Isn’t Your Advantage – Your IP Strategy Is.

For a board, the question to add to the AI paper is simple. Of the data, workflows and methods this investment will create, which will we be able to stop others using, and what have we done to make that true.

Read the article: AI Will Shift $4.7 Trillion in Profits. What’s Your Stake?, Bain & Company.