The Data Dilemma

Your moat goes away whether you license your data or not. What’s still up to you is whether you get paid for it.

Every executive knows their data is extremely valuable. It’s been the smart thing to believe, and to say to your peers, for a decade now. I’ve spent seven years building data programs inside a large enterprise, and from where I sit, most of that value has never been pulled out. Subpar governance, path-dependent legacy processes, and ROI that can’t compete with lower-lift but still high-impact data projects have kept it locked up. There are decades of value left in that data even if nothing new came along. But progress doesn’t wait for us to catch up. In the world of data, that’s good news, because data is one of the not-so-secret key ingredients in AI. Your data now has buyers it didn’t have before.

AI companies need nonpublic data to train, validate, and fine-tune models so they can get better at complex, economically valuable tasks. What throws off vast quantities of this exact data? Existing businesses. Years of bids and estimates, project schedules, field reports, support tickets, maintenance logs, claims decisions. A company builds up those records just by operating, and that’s what models are short on right now. It’s the scarcest and most necessary ingredient for models to become what investors in the frontier labs need them to be to justify their valuations.

More effective models only enable your business. Your existing agreements, relationships, equipment, strategy, etc can be executed more effectively and cheaper using models trained on the appropriate data.

That puts you in a strong position. AI companies want to pay you for data you already have and keep creating, so their models get better at the work you need done in your own business. You’d be licensing the data, so you still own it. You can keep using it to train your own models and run your operations, and unless you sign an exclusive deal, you can license the same dataset to more than one buyer. Customer and confidential information can be anonymized or removed before anything changes hands.

There’s a catch, though. If you’re worried about giving up a competitive advantage by sharing your data, remember that your competitors are making the same decision you are. It’s a prisoner’s dilemma. Here’s what each choice actually pays:

You license / Competitor licenses

You: $100K–$2M + more useful models
Them: $100K–$2M + more useful models

You license / Competitor protects

You: $100K–$2M + more useful models
Them: $0 + more useful models (built on your data)

You protect / Competitor licenses

You: $0 + more useful models (built on their data)
Them: $100K–$2M + more useful models

You protect / Competitor protects

You: $0 + less useful models
Them: $0 + less useful models

Whatever your competitor does, licensing pays you and protecting doesn’t. And once anyone licenses, the models get better for everyone, including the people who held out.

Your competitive advantage only survives in the bottom-right box, and that requires every one of your competitors to hold out too. If even one of them licenses, the capability you’re protecting ends up built into future models, where anyone can use it, including your competitors. That outcome is overdetermined.

So here’s the dilemma. If your edge is worth more than the payout, your whole industry might be better off if nobody licensed. But nobody can enforce that, and rational competitors will license anyway. Your moat goes away either way, so you might as well get paid for it.

There’s also a time component. AI companies have more money right now than they’ve ever had, and they know it won’t last. This data is what’s holding back the work they need their models to do. Because most companies haven’t licensed their data yet, it’s scarce, and that pushes the price up. So does the deadline these companies are up against. They’re willing to overpay for data today to get ahead of each other, win market share, and prove the technology works before their burn rate dries up investment. They’ll keep buying this kind of data after that, but every deal that closes makes the next one a little less valuable. As more data comes to market and things settle into a more sustainable equilibrium, prices will come down.

The best-informed and savviest owners will get paid for a dataset they already have. They’ve been paying to create it for years without ever counting it as an asset. If you wait because this feels new, you’ll probably miss the window. The instincts that got you here were hard-won, but you built them in a much slower market. In this AI moment, the caution that has served you well can cost you a lot. Most of your competitors are still being cautious, which works in your favor if you aren’t.

If you want to know what your data is worth and who’s buying it, that’s the conversation I’m having with owners right now. Contact Harvest Intelligence.