Across service-centric industries, the AI conversation has moved past experimentation. Firms are no longer asking whether to adopt it, but why the returns are so uneven. . Studies trace it back to the data – fragmented, inconsistent, and spread across systems that were never designed to work together. The story is the same in the legal industry as well. Deloitte’s 2026 Chief Legal Officer report found that 79% of legal departments increased AI investment over the past year, with 67% ranking data as one of the highest investment areas. The question is no longer whether to adopt AI, but how to make it part of the way the firm operates.
What’s changing isn’t simply the adoption of AI. It’s the role AI is expected to play. Rather than AI being a productivity tool, it is evolving into an operational capability that connects information across systems, coordinates work, and supports decisions throughout the matter lifecycle.
That shift makes connected operations a competitive advantage. Trusted data, integrated business applications, and connected processes are no longer modernization projects. They are the foundation that allows AI to move beyond isolated use cases and become an orchestration layer across the firm.
Our research done with independent research house Briefing, shows many law firms are still building that foundation. Across 30 UK law firms with more than 250 employees, systems integration scored just 6 out of 10, the lowest-rated capability in the study. When matter or project management, finance, and document management remain disconnected, AI can improve individual tasks, but it cannot orchestrate the business.
AI cannot reason over data that was never captured
Start with the clearest case. The same research found that 3% to 5% of billable value disappears in the pre-matter window which is before a matter or project formally exists in the system. Lawyers start work, the matter or project record is not open yet, and that time is never recorded against anything.
No model fixes that. There is no missing entry to find, no pattern to surface, no anomaly to flag. The data was never created. With connected data across your matter or project lifecycle and an AI layer sitting on top of it, the revenue leakage is flagged before it happens. A
Capture comes before intelligence. You cannot analyze what you never wrote down.
Partial integration produces confident, unreliable output
The 6 out of 10 is the deeper issue. When onboarding, matter or project management, and finance run on systems that are only partly connected, the same fact lives in three places in three slightly different forms. A person handles that by knowing which version to trust. A model does not. It treats all three as signal, averages them or picks one, and returns an answer that looks authoritative and is quietly wrong.
The research shows how ordinary this is. Between 55% and 60% of the finance leaders said they could not reliably trace a write-off back to where it started. If people with full context and a direct financial stake cannot follow that thread, a model reading the same disconnected records will not find it either. It surfaces correlations that are artifacts of the data structure, not real signals about the business.
Feed a capable model a fragmented system of record and it does not warn you the data is fragmented. It answers anyway. That is the failure mode. Not a wrong answer you catch, but a plausible one you act on.
Connect the record first, then layer the intelligence
The firms getting value from AI are not the ones with a connected system of record, where the matter or project is opened cleanly at intake, time is captured from the first hour, and finance reads from the same source that delivery writes to.
This is the practical case for AI-powered legal matter management that runs on a connected platform, rather than a model bolted onto a stack of disconnected tools. When the underlying record is whole, the intelligence layer has something real to work with. Forecasts reflect actual work in progress. Risk flags point at genuine exposure instead of data-entry noise. The model stops guessing across gaps because the gaps are closed.
The sequence is the whole point, and it is the part most firms get backwards. They buy the point solutions first because it demos well, then wonder why it underperforms on their own data. The integration score should come before the model selection. A 6 out of 10 foundation caps what any AI on top of it can do, however good the model is.
For legal specifically, and for any enterprise weighing an AI program, the diagnostic question is not “which model.” It is “what is our systems integration score, and would we trust a decision made on top of it.” If the honest answer is a 6, the first step is the connected record that makes one worth running.
None of this requires a transformation program. It requires a matter or project record that finance, delivery, and billing all can trust at the same time, captured cleanly from intake, tracked through delivery, and readable by any model built on top of it. That’s what AI-powered legal matter management or project management is built to do: give every stage of the project or matter lifecycle one connected source of truth, so the AI conversation can start with clean data instead of working around the gaps in it. See how it works →”
