In most verticals, you can see two distinct layers of AI tooling emerging. They’re often described as if they’re competing, but they serve different workflows and different moments in the professional’s day.

The first is the data layer. These are tools that give AI agents access to the underlying data of a domain — property records, MLS listings, financial filings, case law databases. They’re high-volume, relatively low-margin per query, and valuable primarily as infrastructure. They solve the “I need the AI to know what’s in this system” problem. The AI can ask questions and get answers about records that exist in a database somewhere.

The second is the workflow layer. These are tools that help professionals do something with documents they already have in hand — a lease that just arrived, a rent roll that needs reviewing, a set of offering memoranda that need comparison. The AI isn’t fetching from a database; it’s processing a document that came from outside the system, extracting the right things, surfacing the right concerns, structuring the output for the next step in the professional’s process.

These two layers are complementary, not competing. The data layer answers “what does the market look like?” The workflow layer answers “what does this specific document say, and what do I need to do about it?”

The mistake is trying to build both at once. Each has different technical requirements, different pricing logic, different buyers, and different distribution channels. The data layer needs breadth — coverage across as many records as possible. The workflow layer needs depth — enough domain knowledge to process one type of document correctly, every time, in the format the professional actually needs.

The professionals who use both want them to be separate tools that work together. They don’t want an all-in-one platform that does both passably. They want the data tool to be best-in-class for data, and the workflow tool to be best-in-class for workflows.

Build for your layer. Let the other layer be someone else’s best-in-class. +++