Own the data layer, not the check-in

English · Nederlands

Diagram: your documents, files and databases feed retrieved context into an AI model, producing a unique SaaS application.

Most AI features are only as good as data that someone else owns. The model is rented, and so are many of the places, prices, reviews and documents it reasons over, on terms the owner can change on a Monday morning. In 2011 I wrote about five lessons Foursquare and Facebook could learn from Google+. Reading it again in 2026, the lesson I ranked fifth turns out to be the one that decided the game.

What I saw then

I wrote the post the week Google+ launched, convinced its location features would shake up the check-in apps. I gave Foursquare and Facebook Places five lessons. Use a curated database of places instead of hoping the crowd fills it in. Give merchants a positive reason to claim their listing, not just a defensive one. Make privacy granular and even fun, as Circles did. Weave location into every form of sharing. And be radically open, with data you can take with you and APIs that connect to other networks.

Google+ is gone, and the check-in as a mass habit mostly went with it. What survived was the dull lesson at the bottom of my list: whoever has clean, current places data wins, because every map, search box and app needs it in the three seconds before the user gives up.

What is the same now

Swap “check-in app” for “AI feature” and the structure is identical. The visible product sits on a data layer that is expensive to build and easy to underestimate. In 2011 I quoted Google’s 50 million places. In October 2025 Google made grounding on Google Maps data, covering more than 250 million places, generally available to developers building on its Gemini API (Google). The places database became a product you rent by the call, sold to the very AI apps that compete for local questions.

The same happened with text. In 2023 Reddit started charging for its API, and the popular third-party app Apollo shut down as a result (TechCrunch). In February 2024 Reddit signed a data licensing deal with Google, reportedly worth about $60 million a year (Engadget). Data that had been free to read became the asset on the negotiating table.

For a B2B tech company the question is uncomfortable and practical. Which data do your AI features depend on? Who owns it? And what happens to your margin and your roadmap if the owner reprices it or switches it off?

What is different

Two things break the 2011 analogy.

First, openness now comes from the incumbents. In 2011 I praised “data liberation” as if openness were the challenger’s weapon. Today the Overture Maps Foundation, whose members include Meta, Microsoft, Esri and TomTom, publishes open map data, including nearly 54 million places at its general release in July 2024 (Linux Foundation). Foursquare released more than 100 million places under the Apache 2.0 license (Simon Willison). When large players give a layer away, they are telling you where they think the money is not. The base layer is becoming a commodity. The value moves to freshness, rich attributes, behavioural signals and the application on top.

Second, regulation now assigns data rights instead of leaving them to contracts. Under the EU Data Act, users of connected products have had the right since 12 September 2025 to access the data their products generate and to have it shared with a third party in the EU. Design obligations for new products follow from 12 September 2026 (William Fry). If you sell hardware in Europe, “our data” is partly your customer’s data by law, and a competitor can ask for it on their behalf.

And to be honest about my own post: Google+ was open and still failed. Openness does not win distribution. Owning data that users and partners cannot easily get elsewhere does.

What to do about it

  1. Map your data dependencies. List every external dataset, API and model your AI features call. For each one, note the owner, the contract term, the price-change clause and a named fallback. No fallback means a board-level risk, not an engineering detail.
  2. Separate commodity from moat. Build on open layers such as Overture or Foursquare’s open places where the data is generic, and stop paying premium prices for it. Put your own money into data nobody else has: product usage, outcomes at your customers, service and failure history.
  3. Write data rights into your contracts. Make sure customer agreements let you use aggregated, anonymised usage data to improve the product, and that supplier agreements protect you against sudden repricing.
  4. Turn the Data Act into an offer. If you make connected products, design data access as a feature, with a clean interface, and charge for the services you build on top. Do not wait for the first request from a competitor to discover what your obligations are.
  5. Give contributors a positive reason. My 2011 merchant lesson still holds. Customers and partners feed your data layer when it helps them directly, through better results, benchmarks or visibility, not because your terms say so.

The toy and the business

In 2011 the check-in was the toy and the places data underneath was the business. In 2026 models get cheaper and more alike every quarter, so the same question comes back: which part of the data layer do you own, and who holds the keys to the rest?

Sources