How TVPi measures legal AI capability

The Index is built entirely from the public record: tracked AI moves, vendor deployments, investment and build activity, talent signals, and verified case studies. Every input is observable and sourced, which is what makes it publishable and comparable across the whole market without any firm having to participate.

Recency is weighted — a deployment announced last quarter counts for more than one from three years ago — and significance is weighted, so a firm-wide rollout outweighs a pilot. Nothing self-reported enters the Index unless it has been published or independently verified.

The limitation is inherent and worth stating plainly: the Index cannot see what a firm has not made public. A firm doing serious work quietly will sit lower than its capability warrants.

The Maturity Model closes that gap. It is a confidential 24-question diagnostic completed by the firm, scored across six weighted dimensions:

• 20%Deployment reachWhich tools are live, how many people hold seats, and how many actually use them.
• 20%Workflow embeddingWhere AI is part of the standard way work gets done, rather than a pilot.
• 15%Governance & riskPolicy, approval routes, client disclosure and model-risk controls.
• 15%Talent & operating modelDedicated roles, an innovation function, and how far training has reached.
• 15%Value measurementWhether realised time, margin and quality impact is actually measured.
• 15%Client-facing deliveryAI in pitches, pricing, matter delivery and client-facing products.

Each answer carries a band score from 0 to 100. Dimension scores are the weighted mean of their answered questions; the composite is the weighted mean across dimensions. Unanswered questions are excluded rather than scored zero, so a partial diagnostic gives an honest provisional placement rather than a punitive one.

1. 0+1. Exploring

   Interest and experiments, but no deployed capability the firm depends on.

2. 25+2. Piloting

   Tools are live with a defined group; value is anecdotal and governance is forming.

3. 45+3. Operationalising

   AI is standard in some workflows, with policy, owners and budget behind it.

4. 65+4. Scaling

   Broad adoption, measured value, and AI appearing in how work is priced and sold.

5. 82+5. Leading

   AI is part of the operating model and a commercial differentiator with clients.

Diagnostic answers are visible only to the firm's own seat holders and to TVPi. They are never attributed, never shown to another firm, and never published.

Peer benchmarks are aggregate only, and no cut is released until at least five firms sit behind it. Firms see peer data only once they have contributed their own — the model is a mutual exchange, not a one-way read.

A self-assessed score is labelled as such. TVPi Verification is an evidence-backed review: deployment and seat data, governance documentation, usage reporting, and value measurement artefacts are examined against the answers given.

A verified assessment carries the Verified mark and is the only form of Maturity Score TVPi will reference externally. Verification is reviewed annually and can be revoked.

The model is versioned. Every assessment records the version it was scored under, so a future re-weighting never silently rewrites a firm's history. This page describes v1.