Timesheets: legal's most underused dataset

Over the past eighteen months, almost every serious law firm has started building an AI knowledge layer. Vector databases, internal chat tools, precedent search agents, repositories of model documents and standard form clauses. Some are built internally, some bought from Harvey or Legora, some routed straight through Anthropic or OpenAI. The arms race is real, and the budgets behind it are serious.

What is striking is what most of these projects ignore. Almost without exception, they are built on documents: precedents, model agreements, memos, opinions, training materials. These are valuable, but they are a curated view of what a firm thinks its work looks like. They are not a record of what its lawyers actually did end-to-end. That record exists, in granular form, in the firm's timesheets. And almost no one is treating it as a strategic asset.

The dataset hiding in plain sight

A modern law firm's practice management system contains, by some margin, the most detailed operational dataset the firm will ever own. Every fee earner, on every matter, has been recording, in increments of six minutes, what they did, when, in what sequence, for whom, and at what cost. For most firms this stretches back a decade or more. It covers transactional and contentious work, international and domestic, matters led by partners as well as by associates. It captures both the deliberate and the accidental: planned strategy and unplanned scope changes.

The reason it is underused is not that the data is missing. It is that the data has been treated, for forty years, as a billing artefact. The narrative line on a time entry exists because someone needs to put it in front of a client. It is almost never analysed for what it actually contains.

Why the narrative matters more than the code

The phase and task codes on a time entry, where they exist at all, are inconsistent, added as an afterthought, and often wrong. They tell you very little about what work was actually performed.

The narrative is different. It is dense, imperfect, sometimes cryptic, and full of signal. It says which documents got fought over, which counterparties dragged their feet, which due diligence streams blew out, which arguments survived to trial. Once you can parse those narratives programmatically, the question changes from "how much did this matter bill" to "what did we actually do, how did we do it, and how did the shape of the work compare to what we priced for".

That single shift is what moves pricing, business development and matter monitoring from anecdote to evidence.

What it unlocks

Pricing teams gain real comparables. Instead of a pricing manager asking three partners what they remember about the last similar matter, the firm can pull every comparable, normalised, with scope items completed, deviations, cost to serve and realised margins. Fixed fees and alternative arrangements stop being a guess.

Business development gains a view of where the firm actually has experience, not just where it would like to claim it. Signals for cross selling become real. So do unused client relationships.

Matter monitoring gains a live frame of reference. A budget overrun stops being a finance question discovered at month end and becomes an operational alert in week three.

Governance gains an audit trail. How AFAs are priced, how they land, and where the gap sits, all become something a firm can manage, not something it discovers after the deal.

Why this has not happened until now

There were good reasons. Narrative variability is high. Confidentiality concerns are real. Hallucination risk on legal text was, until recently, prohibitive. Practice management vendors built their products around billing workflows, not analytics.

Most of these constraints are now soluble. In recent client work, Ayora's enrichment at the level of the narrative line has been producing matter profiles at roughly 95 percent accuracy, with processing in a secure environment and anonymisation built in from the start. The technical question has largely been answered. The cultural one has not.

The barrier now is who owns the dataset internally. In most firms, timesheets belong to billing and finance. The people thinking about pricing strategy, AFAs, BD and matter management rarely have a proper seat at that table. Until they do, the most valuable dataset the firm owns will keep being used for invoices and nothing else.

A posture shift, not a tech project

Building a smarter precedent bank is a tech project. Treating timesheets as a strategic dataset is a posture shift. It changes who in the firm gets to ask questions of the operational record, and what kind of questions they get to ask.

The firms that take this seriously over the next two years will be making pricing, BD and resourcing decisions on a different evidence base from everyone else. That is a bigger competitive gap than any precedent search tool will close.