A legal-aid clinic screens intake questions around the clock without lawyer hours

Representative examplesEvery case study in this library is an illustrative composite of the kind of engagement we deliver — written to show our method and standards, not to name clients.

Where they started

Funded by a mix of grants and council contracts, the clinic provides free legal help across a defined set of matter types, income bands, and geography. Its intake phone line opens two hours a day because that is what staffing allows, staffed by caseworkers who also carry full caseloads. The people calling are often in distress, frequently unsure whether their problem is even legal, and occasionally in crisis — the clinic's duty of care starts at the first ring, whoever answers it.

What it was costing

The clinic's intake phone line opened for two hours a day and overflowed constantly. Prospective clients with urgent matters waited days to learn the clinic couldn't help with their issue type, and staff spent mornings triaging voicemails.

What they could see

• Outside the two-hour window, urgent callers reached an answering machine that promised a call back with no promise of when.
• Caseworkers spent the first hour of each day triaging voicemails for people the clinic could never have helped.
• Some callers learned after days of waiting that their matter type fell outside eligibility — a hard no delivered late feels like a cruel one.
• When the line was busy, nobody knew how many people had given up without leaving a message.

The constraints we worked inside

• Legal advice cannot come from a chatbot — the assistant screens and informs, never advises.
• Eligibility rules (matter types, income bands, geography) were precise and had to be applied exactly.
• Vulnerable callers sometimes disclose crises — escalation to a human is instant and obvious.

What had been tried before

An answering machine script was rewritten to list eligible matter types.

People in crisis don't self-assess against a menu; most callers still left messages saying only 'I need help' and the triage started from zero.

A volunteer-staffed callback rota extended the line's effective hours.

Volunteer turnover made screening inconsistent, and one over-generous callback created expectations the clinic then had to refuse in person.

What we proposed

We proposed an assistant that runs the clinic's actual screening questions in order, applies the published eligibility rules exactly, and states outcomes plainly — no soft rejections, no false hope — while treating crisis phrasing as an interrupt that goes straight to a human callback queue. The rules live in deterministic logic the assistant narrates rather than decides, because eligibility is exact and advice is not ours to give. Overnight screenings become a prioritized morning report with full transcripts, so lawyer hours start where they matter. Everything about the crisis path was tested with the clinic's own people in the room.

Just as important is what we ruled out, and why:

• Hiring dedicated intake staff — Grant funding is fixed and contested; the constraint was hours in the day, and payroll would consume grant money faster than it created capacity.
• A phone-tree IVR system — Menus punish exactly the callers the clinic exists for — distressed people who don't yet know what their problem is called.
• An assistant that offers general legal information — The boundary between information and advice is the whole risk; the clinic ruled out anything that could be read as counsel.

How the work ran

Delivered by the systems pod — automation specialist over 6 weeks, with working increments reviewed with the client every week.

The stack, and the reasoning

LLM assistant platform

Selected for strict conversation control — the screening must follow the clinic's questions in order, not improvise empathetic detours around them.

Rule-based screening flow

Eligibility rules are exact — matter types, income bands, geography — so the rules live in deterministic logic the assistant narrates rather than decides.

Callback escalation

Crisis phrasing and certain matter types interrupt the flow to a human queue with urgency framing; escalation had to be louder than every other feature.

Transcript reporting

Lawyers start from a prioritized morning summary with full transcripts, so the overnight volume becomes a triaged queue instead of a mystery.

What went wrong

How we worked together

Cadence

Twice-weekly half-hour reviews with the clinic coordinator during the build, then a weekly transcript review once screening went live.

Client side

The coordinator owned the screening script and rules; a senior caseworker tested the crisis path and the social worker reviewed every escalation phrasing.

Decisions

Anything touching eligibility wording was decided by the clinic's supervising solicitor in a standing Thursday slot — no exceptions, including ours.

They provided

The written eligibility rules, anonymized intake call notes for realistic testing, and caseworker time to screen live calls alongside the pilot.

What changed

The headline: screening availability, with 63% of eligible matters booked to callback without staff contact in the first month — 2 hrs → 24/7, read from Assistant transcripts plus callback log. A second check: crisis-flagged conversations reaching a human within the hour at 100%.

The morning triage hour turned into a working session on real cases — the queue arrives prioritized, with transcripts, and caseworkers start where the need is sharpest. Callers outside eligibility now learn it in minutes instead of days, and several have written back to thank the clinic for the referral routes the assistant offers. The crisis path is the part staff check first when they audit transcripts; it is also the part they trust most, because they wrote its wording.

The result was read from Assistant transcripts plus callback log against the pre-engagement baseline over the stated window, with a guardrail check on crisis-flagged conversations reaching a human within the hour. Where platform-reported numbers and business outcomes differ, this record says which layer it is quoting.

What they own now

• The assistant running the clinic's screening questions end to end, with the versioned screening flow definition.
• The eligibility rules encoded as a readable rules file the clinic edits, alongside the workspace credentials.
• The crisis-trigger list and its eval cases, maintained with the social worker's annotations.
• A scheduled morning summary delivering overnight transcripts to the triage inbox.
• A training session for new caseworkers on reading and auditing transcripts.