When a contractor asks whether their software is “AI-ready,” they usually mean “does it have an API.” We used to think so too. Then we spent a couple of years pulling real data out of the systems contractors actually run, to build a clean financial source of truth on top of them. The lesson was humbling: the presence of an API tells you almost nothing about whether you can get usable data out.
Take one large construction ERP. Its REST API is documented and modern-looking. In practice, under a real customer’s credentials, only one read endpoint responded at all; the rest returned 401s from a different auth realm or 404s because they were never deployed. Even the one endpoint that worked returned 15 fields where the underlying database held 39, missing exactly the fields you need for collections and due-date math. The honest recommendation was to keep reading the database directly and ignore the API.
The opposite failure is just as common. A widely used small-business accounting system has a perfectly good API, but it exposes only the first three transaction custom fields. If a contractor has already used those three slots, the field that bridges their jobs to their invoices is simply invisible to the API, and the whole integration quietly falls back to manual CSV exports. Nothing is “broken.” The API is just blind to the one field that matters.
And then there are the systems with no pull API at all. One popular landscaping platform offers only a Zapier hook; to get reports out programmatically you drive its web UI and export per-widget, capped at 100 days per pull. That is not a knock on the product. It is a fact a practitioner has to plan around, and it is invisible from a feature list.
So the real ingestion picture, across the systems we work with, is a patchwork: API where it is genuinely good, direct database reads where the API is hollow, scheduled report or CSV delivery where the API hides fields or costs too much, and browser automation where there is no API at all. A serious data layer uses all four and reconciles between them. That is the unglamorous work nobody demos.
How common is this? By our count, roughly 80% of the software in this industry has no usable API or clean export at all. We have had to find on the order of ten different ways to get data out, browser automation included. That is not a fringe case; it is the median case.
This is why we score a system’s AI-readiness by a different question than most buyers ask. Not “does it have an API,” and definitely not “does it have an AI feature.” The question is: how cleanly can a practitioner get complete, trustworthy data out of it, by any method? Answer that, and you know whether AI can ever do anything useful on top.
Drawn from our hands-on work getting data out of contractor systems and from operator interviews. Product observations are our own experience, framed analytically. See the full software index.