Documents in minutes, not days: 15 hours a week back for every fee earner

The problem

Every new piece of client work arrived as a bundle of PDFs. Before any billable work could start, a fee earner had to read them, extract the names, dates, figures and references, look up the supporting information that wasn't in the bundle, and type all of it into a set of templates to produce the documents the engagement required. Four to six hours per set was normal, the same data was keyed three or four times, and each re-keying was a chance for a typo to reach a client. The firm's most expensive people were spending more than half their week on work that generated no fees.

Why an off-the-shelf tool didn't fit

The firm had tried template tools and had looked at general-purpose AI assistants. Neither solved the actual problem. Template tools still needed someone to type the data in. A chatbot could draft one document from a pasted PDF, but it couldn't read a whole bundle, look up the supporting data, fill several different templates consistently from one set of facts, or keep a record of who produced what. And nobody wanted client documents passing through a consumer tool. The job needed a system that lived inside the firm's own environment and did the whole sequence, not one step of it.

What we built

A bespoke document-generation platform, built around the way the firm already worked:

• A secure, multi-user dashboard with role-based access, so fee earners, support staff and managers each see what they need and nothing more.
• Automated ingestion of the incoming PDFs, with AI extraction of the key data at high accuracy — including scanned documents, not just clean digital ones.
• Intelligent enrichment: the supporting information that used to need a manual lookup is retrieved automatically and attached to the record.
• Generation of every required document from that single set of facts, in parallel, in the firm's existing formats — so one intake produces the full set rather than one document at a time.
• A review step built into the workflow. The system drafts; a fee earner checks and approves every document before it leaves the firm. The AI removes the typing, not the professional judgement.

The platform is model-agnostic: the AI model doing the extraction can be swapped as better or cheaper models appear without touching the workflow, the templates or the permissions. The firm owns the code, the accounts and the data, and hosting was set to their requirements.

How it was delivered

It started with an Automation Audit — a workshop with the fee earners and support team, and a written map of where their hours went. Document production was the obvious first target. The platform was built as a one-month Automation Sprint: the first two weeks on extraction and enrichment, tested against real historical bundles until the outputs matched what the team would have produced by hand; the second two weeks on the dashboard, the templates and the review flow. The team ran it alongside the manual process for the final week, then switched.

The impact

A document set that took four to six hours of a fee earner's time now takes under twenty minutes, most of it the review. Turnaround per engagement is 85% faster. Each fee earner has around 15 hours a week back, and because that time is billable, the firm estimates roughly £120,000 a year of additional capacity without hiring. Re-keying errors have gone with the re-keying. The firm is now extending the same platform to the next document-heavy process.

Is this your problem too?

If your fee earners or specialists spend their week turning incoming documents into outgoing ones — extracting, re-keying, formatting — the same pattern applies, whatever the sector. The audit is how we find out whether it will pay back for you, and you keep the map either way.