Technology and AI

Designing an AI intake assistant for a law firm

How to design an AI intake assistant for solicitors: question sets by practice area, urgency rules, booking rules, boundaries, notes and testing.

Ava research teamPublished 6 min read

On this page
  1. Start with the job, not the script
  2. What callers value differs by matter
  3. Design principles
  4. Question sets by practice area
  5. The design process
  6. Urgency and escalation
  7. Boundaries the assistant must keep
  8. Booking rules
  9. Testing before launch
  10. Example: a design workshop
  11. Next steps
  12. Frequently asked questions
  13. References

Key takeaways

  • Design the assistant around one job: getting a new enquiry to a booked consultation with a note the solicitor can act on.
  • Write short question sets per practice area, with clear urgent triggers and a list of things never to ask or say.
  • Capture other parties' names carefully for the conflict search, and leave the decision to act with a person.
  • Test with at least twenty realistic scenarios, including attempts to get advice and distressed callers, before going live.

Start with the job, not the script

An intake assistant is only as good as the procedure it follows. Before writing any questions, agree what the assistant is for. For most firms, the answer is narrow: answer every new enquiry, collect what a solicitor needs for a first conversation, spot urgent matters and get them to a person, book or request the consultation, and hand over a clear note. Everything else, including advice, merits and the decision to act, stays with people.

A narrow job makes the assistant easier to design, test and supervise. It also makes it easier to explain to callers, who need to know in the first few seconds what the assistant can and cannot do for them.

What callers value differs by matter

Callers do not all want the same thing from first contact. The Legal Services Consumer Panel's 2026 tracker found speed mattered most in accident and injury claims, where 37% compared providers on it against 25% overall, while an online service mattered more for employment disputes, at 39% against 24% overall[1].

What consumers compared when choosing a provider, by type of matterLegal Services Consumer Panel tracker, 2026
What consumers compared when choosing a provider, by type of matterBar chart. Speed: accident and injury claims: 37%; Speed: all matters: 25%; Online service: employment disputes: 39%; Online service: all matters: 24%.Speed: accident and injury claimsSpeed: accident and injury claims: 37%37%Speed: all mattersSpeed: all matters: 25%25%Online service: employment disputesOnline service: employment disputes: 39%39%Online service: all mattersOnline service: all matters: 24%24%What consumers compared when choosing a provider, by type of matterBar chart. Speed: accident and injury claims: 37%; Speed: all matters: 25%; Online service: employment disputes: 39%; Online service: all matters: 24%.Speed: accident and injury claimsSpeed: accident and injury claims: 37%37%Speed: all mattersSpeed: all matters: 25%25%Online service: employment disputesOnline service: employment disputes: 39%39%Online service: all mattersOnline service: all matters: 24%24%

Source: Legal Services Consumer Panel, 2026[1]

Show the data
ItemValue
Speed: accident and injury claims37%
Speed: all matters25%
Online service: employment disputes39%
Online service: all matters24%

That should shape the design. For personal injury, the priority is to book a call quickly and capture the accident date. For employment, callers may prefer to complete details online or receive a link to upload documents. For family matters, tone and safety matter more than speed. One script for every practice area will serve none of them well.

Design principles

  1. Say what it is: automated, here to take details and book a consultation, unable to give advice.
  2. Ask only what the next step needs. Every extra question adds time, data risk and drop-off.
  3. One question at a time, in plain English, with the caller's own words recorded.
  4. Spell back names and repeat back dates and numbers.
  5. Urgency beats completeness: when an urgent trigger fires, stop the script and get a person.
  6. Always offer a person, and never trap a caller in a loop.
  7. End every call with a clear next step and when it will happen.

Question sets by practice area

Starting points; each practice lead should approve their own set.
Practice areaMust askUrgent ifNever ask or say
Residential conveyancingBuying, selling or both; property location; price range; mortgage; target datesExchange or completion within days; lender deadlineWhether the deal is a good one
FamilyType of issue; other party's name; children involved; court papers receivedHearing date soon; safety concerns; child at riskDetailed questions about abuse; any view on likely outcome
EmploymentEmployee or employer; employer's name; key dates such as dismissal; current statusTime limit mentioned or approaching; hearing listedWhether they have a good claim
Wills and probateWill, lasting power of attorney or estate; who has died and when, if relevant; executor roleFuneral costs or urgent property issue; capacity concernsViews on who should inherit
CriminalWho is involved; whether someone is at a police station or in custody; court dateAnyone at a police station now; hearing today or tomorrowAnything about what happened or guilt
Personal injuryType of accident; date; where; injuries in brief; other party or insurerAccident date approaching a time limitEstimates of compensation

Give each practice lead ownership of their question set. They know which details save time on the first call and which questions upset people. Review the sets every quarter against real call notes.

The design process

  1. Set the scope

    Decide which calls the assistant takes: all new enquiries, overflow only, or out of hours.

  2. Write the opening and closing

    Approve the words the assistant uses to introduce itself and to end each call with a next step.

  3. Draft question sets

    Practice leads write the must-ask questions, urgent triggers and never-say list for their area.

  4. Define conflict data

    Decide which names and details the assistant captures for the conflict search, and how it confirms spellings.

  5. Set urgency routes

    For each trigger, name who is contacted, how, and what happens if nobody answers.

  6. Set booking rules

    Which fee earners take which consultations, slot lengths, channels and how far ahead to book.

  7. Design the note

    Agree one note layout the whole firm uses.

  8. Test

    Run at least twenty scripted scenarios and fix every failure before launch.

  9. Launch and review

    Go live for one practice area or for overflow first, and review every call in the first two weeks.

Urgency and escalation

Some matters cannot wait for a booking. Police station duty solicitors in England and Wales are available 24 hours a day[2], so if your firm does criminal work, a call about someone currently at a police station needs an immediate route to a person or to the duty arrangements. Similar urgency applies to hearings in the next day or two, injunction situations, imminent completions and any risk to someone's safety.

Write each trigger as plain examples of what callers actually say, not legal categories. Callers rarely say they need an urgent non-molestation order; they say they are frightened to go home. Test the triggers with varied wording and with callers who are upset and speaking quickly.

Boundaries the assistant must keep

The SRA has recognised that legal chatbots can respond quickly to potential clients, potentially around the clock, gathering information and triaging cases for urgency[3]. The same report makes clear that firms remain accountable for what AI tools say, including errors in advice from a third-party chatbot[3]. So the boundaries must be designed in, not left to chance: polite refusal wording for advice requests, no comment on merits or outcomes, no acceptance of instructions and no handling of payment details.

Vulnerable callers need particular care. In the Legal Services Board's 2026 public panel, 81% of legal service users agreed that AI legal tools should never act automatically for a vulnerable person in high-stakes situations[4]. If a caller seems confused, distressed or at risk, the assistant should take only essential details and bring in a person.

Booking rules

Booking is where intake turns into work. Decide which fee earners take first consultations in each area, the slot length, whether consultations are by phone, video or in person, and the earliest and latest times the assistant may offer. Hold back a few slots each day for urgent cases. If your firm charges for initial consultations, the assistant should state the published position and that a member of the team will confirm the arrangements; if it does not, say so clearly. Confirmation and reminder messages should go only to callers who have booked.

If you use Ava, the same design work applies: it follows the intake questions, urgency routes and booking rules your firm approves, so the quality of those decisions determines the quality of every call.

Testing before launch

  • A routine enquiry in each practice area, from first question to booked consultation.
  • A caller who asks directly whether they have a case.
  • A caller who gives the other party's name in an unusual spelling.
  • A caller at a police station, and one with a hearing tomorrow.
  • A distressed caller who mentions feeling unsafe.
  • A caller who wants a person straight away.
  • An existing client asking for an update on their matter.
  • A caller who becomes angry or makes a complaint.
  • A caller who speaks quickly, with background noise, or with limited English.

Score each test on three questions: did it stay within its boundaries, did it get urgent calls to a person fast enough, and could a solicitor act on the note without calling back to repeat everything?

Example: a design workshop

Example (fictional): a mixed practice holds a two-hour workshop with its family, employment and conveyancing leads, its receptionist and its COLP. They agree a scope of overflow and after-hours calls, draft three question sets and list eleven urgent triggers. The family lead removes two questions she thinks are too intrusive for a first call. The receptionist points out that many callers ring about existing matters, so a clear message-taking route is added. Testing finds that the assistant misses one way callers describe a hearing date; the trigger wording is widened before launch.

Next steps

Pair the design with the compliance work in AI phone answering and the SRA and automated call handling and UK GDPR. The escalation rules are covered in more depth in when an AI assistant should hand a call to a person, and if you plan to use the same approach on your website, read AI chat on law firm websites: guardrails.

Frequently asked questions

What questions should an AI intake assistant ask?

Only what the next step needs: contact details, practice area, a short description in the caller's words, other parties' names for the conflict search, key dates and anything that makes the matter urgent.

Should an intake assistant run the conflict check?

It can capture the names needed and pass them on, but a person should review the conflict search and decide whether the firm can act.

How do we stop an intake assistant giving legal advice?

Write clear refusal wording, test it with many phrasings of advice requests, review real calls regularly and correct the script quickly when it drifts.

How many scenarios should we test before launch?

At least twenty, covering each practice area, urgent and distressed callers, advice requests, existing clients, complaints and difficult audio conditions.

References

  1. Legal Services Consumer Panel (2026). How consumers are choosing legal services (Tracker Survey 2026) (opens in a new tab) ↑

  2. GOV.UK (2026). Being arrested: your rights: legal advice at the police station (opens in a new tab) ↑

  3. Solicitors Regulation Authority (2023). Risk Outlook report: The use of artificial intelligence in the legal market (opens in a new tab) ↑

  4. Legal Services Board (2026). AI in Legal Services: Research Report (LSB Public Panel) (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo