Government legal function
A public legal directorate with a fixed budget, unbounded demand, and a nine-month procurement cycle. We ran the naming work in parallel with the approval track and installed automation inside the statutory sign-off chain rather than around it.
Practice
India · Public-sector legal directorate
Transformation
10 weeks
Track
AI Adoption
Primitives moved
Research, Evidence, Actions
The challenge
The directorate answered advice requests from across several departments with no way to reuse its own prior positions - the same question would be researched from scratch by different advisers months apart, occasionally producing inconsistent advice on the same statutory point. Backlog was measured in months. Procurement, data residency rules, and delegated decision-making authority ruled out most of the commercial market and any redesign of the approval chain.
The transformation, phase by phase
Phase 01
Name
Week one began two tracks simultaneously: the documentation work, which required nobody's approval, and the security and data-protection assessment, which required a great deal of it. We mapped request types, the action sequence for each, and the statutory points on which advice recurred. Eleven questions accounted for a large share of all advice given.
Phase 02
Systematise
The eleven recurring positions became a dated, anchored advice stock inside systems already accredited for official records. A standing evidence rule - no assertion without a pin-cited source - was adopted for human-drafted advice first, four weeks before any automation was discussed. Provenance capture was designed in from the start: sources relied on, named reviewer, and version of the material reviewed.
Phase 03
Automate
Retrieval over the advice stock surfaces the directorate's existing position and flags what has changed since it was written. First-pass drafting runs only on high-volume, low-discretion advice categories, inside the existing sign-off chain, with a named human reviewer of record documented before deployment. Nothing discretionary or rights-affecting was automated, and that boundary was written down before anything shipped.
The shift to AI-native
The strongest case for recovered capacity in legal work sits in the public sector, where unmet demand becomes delay and backlog rather than foregone revenue - and the hardest constraints on realising it sit there too. Both facts are structural. The decision that made this engagement work was sequencing: starting the naming work and the procurement track in the same week rather than in series, which was worth an entire quarter. By the time the approval arrived, the directorate had a specified requirement derived from documented process instead of a general desire for AI. The anchoring discipline turned out to strengthen the records and audit position rather than threaten it, which is the argument that ultimately carried the internal case.
AI-native markers this firm now meets
- AI runs first, systematically, on defined categories
- The organisation gets measurably better over time
- Turnaround is fast on routine categories
Outcome
Advice backlog
−58%
Repeat questions
answered from stock
Discretionary decisions automated
0, by design
We had been told for two years that our constraints made this impossible. What we actually needed was someone willing to design around them instead of arguing with them.