Entity resolution in legal data.

When a homeowner sues a large builder, the name on the lawsuit is rarely the whole story. The defendant is often one small company inside a much bigger enterprise — and the record that matters belongs to the whole enterprise. Finding that bigger picture is not a name search. It is a machine-learning problem.

One builder, many names

A large homebuilder may operate under one legal name in one county and a different name in the next. Each of those companies is legally separate. Together, they act as one business.

There is nothing sinister about this. Big builders routinely set up separate companies for each project or region, and that structure serves real business purposes. But it has a side effect: the enterprise's full history — its lawsuits, its repeat disputes, how it responds to similar claims — stays hidden unless someone connects all those names across all those records. For attorneys who represent homeowners, no one had done that work.

The information gap

Builders track their own record. They know how their claims and cases have gone, across every market. The attorney on the other side of the table has historically worked from personal experience alone. That gap is not bad lawyering. It is a structural imbalance, and closing it is what DAIS was built to do.

The stakes are concrete. The named defendant is often a small company created for a single subdivision, with a thin record — maybe a few cases in one county. Judge the case by that record and you are reading a fraction of the picture. A defendant with no visible history may sit inside an enterprise that has handled hundreds of similar claims. That is negotiating with the wrong map.

Why a name search isn't enough

The obvious fix is a thorough search for the builder's name and its variations. The trouble is that related companies often carry no trace of the parent's brand. A project company named after an oak tree on the land shares no words with the national builder that owns it. A search returns nothing — not because the company is absent, but because the names simply are not related.

So name matching catches the easy cases and misses the rest. Worse, it misses exactly the companies most likely to be obscured.

Why machine learning solves it

No public record has a column that says "parent company." The connections are real, but they show up as scattered clues across many record types and many places. So the right question is not "do these names match?" It is "how many independent clues point to a link between these two companies?" One clue alone proves little — a shared address might just be a common law office. Several clues lining up makes a real link far more likely. Weighing many clues at once, across thousands of company pairs, and turning them into a scored probability is what machine learning does well and a simple database search cannot.

For the specific data inputs that feed this resolution — and what the resulting map of connected companies looks like — see How machine learning tracks the same builder across dozens of legal entities.

What the full picture shows

Once the companies are connected, records can be added up at the enterprise level. Instead of one project company's few cases, you see the whole enterprise's history across every market where it operates — the view the builder has always had from the inside.

That view answers questions no single record can. How does this enterprise respond to a given type of claim? Does its behavior differ by region, or shift over time? It also allows comparison: how one builder's aggregate historical pattern stacks up against the market as a whole. That is context for a matter, not a prediction about any case. But an attorney who walks in with it is working from a very different information position.