In-house legal department, technology company

The challenge

Requests arrived by email, Slack direct message, a ticketing queue nobody maintained, meeting invitations, and corridor conversations. No two lawyers received work the same way and none of it was measured. The general counsel could not answer the CFO's question - what would two more headcount actually buy - because the department had no description of its own workload beyond the shared sense that it was drowning.

Primitives most affected

The transformation, phase by phase

1. 01

   Name

   Two weeks logging every inbound request across all five channels. The result was uncomfortable and clarifying: 41% of requests were one of six recurring question types the department had answered many times before, and a further 22% were standard NDAs. Only a minority of inbound volume required legal judgment at all.

2. 02

   Systematise

   One front door, with a structured intake form that classified by request type and captured the information a lawyer would otherwise chase. A written risk tier defined what was low, medium, and high, with routing rules per tier. Prior answers to the six recurring questions moved into a searchable stock rather than living in individual inboxes.

3. 03

   Automate

   The intake layer now surfaces the existing answer at the front door for recognised routine questions, resolving them without opening a request. Standard NDAs run a first-pass review against the department's own playbook, with a named lawyer reviewing before anything is returned. Everything sits inside the company's existing enterprise tooling, because that is what security review had already approved.

The shift to AI-native

This is what AI Adoption looks like when it is done deliberately rather than as a consolation prize. The department could not re-architect its own technology stack, could not select tools outside enterprise procurement, and could not change the sign-off chain above it - so a full AI-native rebuild was never honestly available. What was available was every one of the six primitives, named and systematised, with automation installed inside the constraints that already existed. The critical decision was refusing to start with contract drafting, which is where the visible volume sat. The bottleneck was upstream, at intake, and attacking it recovered more capacity than a drafting tool would have while touching no legal judgment at all. The department did not get bigger. It stopped absorbing work that never needed a lawyer.

Outcome

Median cycle time

6 days → 1.5 days

Routine requests deflected

34% of volume

Headcount added

0