Client
Harbour & Croft
Sector
Professional services, legal
Headline
8 hrs · Saved per fee earner each week
Time to ship
Six weeks
The problem
Harbour & Croft is a mid-sized Liverpool law firm with strong commercial and employment practices. Their fee earners were spending a significant proportion of every week on first-pass contract review — reading commercial agreements, identifying the standard clauses, flagging deviations from the firm's playbook, summarising the document for the partner. The work was clerical for an experienced lawyer; the cost of doing it manually was real both in fees written off and in associate retention. Two associates had given fee earner exit interviews citing "too much document grunt work" in the previous year.
What we built
A contract review pipeline scoped to commercial contracts — Harbour & Croft's highest-volume document type. The system reads an inbound contract, identifies the clauses against the firm's clause library, extracts the key terms (indemnity, liability cap, change of control, exit, IP assignment, jurisdiction, governing law), compares to the firm's playbook, and produces a structured first-pass review marked up against the original document.
Every flagged deviation cites the relevant clause in both the contract and the playbook. The fee earner reviews the marked-up output rather than re-reading the entire document. Anything the system is uncertain about defaults to "needs review" rather than guessing.
Built in six weeks: two weeks scoping the playbook and clause taxonomy with the partners; two weeks building the extraction and comparison pipeline against an eval set of 80 real contracts; two weeks integrating into the firm's document management system and rolling out to a pilot cohort of three fee earners.
What changed
- 8 hours of fee earner time saved per week across the pilot cohort — the headline number the partners had asked us to move.
- First-pass review time down from 90 minutes to under 25 on a typical commercial contract.
- Quality consistency up: every contract is now reviewed against the same playbook in the same way, with the deviations cited rather than missed.
- Associate retention conversation has shifted — the document grunt work that was driving exits is now a fraction of the week.
What we did not do
The system does not finalise legal advice. Every flagged deviation goes to a fee earner; every contract that gets signed off has a named lawyer who has reviewed it. The AI does the first-pass and the structural work; the lawyer does the judgement and the client communication. The SRA expectation that a person is accountable for legal advice is satisfied at every step.
The system also has explicit refusal behaviour: contracts outside the trained scope (jurisdiction outside England & Wales, contract types outside the commercial scope) get routed to a human first reviewer rather than being processed. We were conservative on scope by design.
Audit and confidentiality
Model hosting is in a UK region. Client documents do not leave UK infrastructure. The firm's existing document management system remains the system of record; our pipeline reads and writes through its API but does not duplicate documents or store them externally. Full audit log per contract — model version, retrieved playbook entries, fee earner sign-off, timestamps.
What is next
Harbour & Croft's commercial partner is the named internal owner. We are on retainer for the next two quarters; the firm is scoping a phase two extension into employment contracts (a different clause library, but the same pipeline shape). The firm has also asked us to scope an internal knowledge assistant against their precedent library for next year.
Related reading
For the broader pattern: AI for legal firms in Liverpool: what is actually working in 2026.