How to Choose an AI Voice Assistant for Legal Intake: A Buyer's Checklist for PI Firms

Buying an AI voice assistant for intake is like hiring a paralegal after a ten-minute interview. The demo is polished, the references are curated, and the real test starts after you sign. A checklist moves that test ahead of the contract.

PI firms carry extra risk. Callers are injured, scared, and one phone call away from the next firm on the list. Firms miss an estimated 35–50% of inbound calls during business hours, so this tool sits directly on your revenue line.

This guide gives you ten checks, a pilot scorecard, and a short set of answers to common questions. Use it on any vendor, including ours.

How to run this checklist

Work through the ten checks in order, and score each vendor from 0 to 2. A 0 means it fails, a 1 means partial, and a 2 means it passes a live test. Use the same script and the same test calls with every vendor.

Insist on live tests. A slide deck proves what a vendor wants you to see. A call from your own phone proves what the product does.

Bring three people to every demo. Your intake lead judges call quality, a case manager judges the case file, and someone from operations judges the integration. Their scores will differ, and that disagreement is useful.

Check 1: Define the calls the assistant must handle

"Intake" means different things at different firms. For some, it is the first conversation with a new lead. For others, it includes every call that arrives after a client signs.

Pull 30 days of call logs before you talk to any vendor. Sort calls into new leads, existing clients, medical providers, insurers, lien holders, vendors, and spam. A typical PI case generates about 150 calls over its lifetime, and many are status updates.

Suppose 600 calls arrive in a month. If 220 are status checks from existing clients and 90 are provider calls, fewer than half are new-lead intake. Your shortlist should reflect that split.

That mix decides the product category. An intake-only tool solves the new-lead problem and leaves the rest with case managers. Read up on AI intake versus AI operations before you build a shortlist.

Ask the vendor: Which call types does it resolve without a human, and which does it only route? Request a live example of each.

Red flag: Answers that list features instead of call types.

Check 2: Test answer speed and coverage windows

Speed decides who gets the case. A caller who reaches voicemail may dial the next firm on the list. Test the assistant at 7 p.m., on a Sunday, and on a holiday.

Check overflow too. Ask what happens when five callers phone at once, and whether anyone waits or meets a phone tree. A strong benchmark is an answer in under two rings.

Spam filtering matters as well. Junk calls burn minutes on metered plans and clutter your call log.

Ask the vendor: What is the answer time at peak volume, and how are spam and robocalls filtered?

Red flag: Coverage described as "extended hours" instead of 24/7 including holidays.

Check 3: Audit the qualification logic

The assistant should screen callers against your criteria, not a generic script. Write those criteria down first. Include injury type, incident date, liability facts, treatment status, prior representation, and venue.

Then run edge cases. Call as someone injured 11 months ago in a state you do not serve. Call as a person already represented by another firm, and as a parent describing a child's injury.

A good assistant captures the facts, flags urgent details, and never promises representation. It leaves the accept-or-decline decision to your team. Listen for any sentence that sounds like legal advice.

Ask the vendor: Can I edit the screening criteria myself, and how fast do changes go live?

Red flag: Criteria locked into a template that only the vendor can change.

Check 4: Verify handoff rules and live transfers

Some callers need a person right away. A caller describing a fresh crash, a safety threat, or a court deadline should reach staff quickly. Define those triggers before the pilot starts.

Then ask how handoff works in practice. A warm transfer passes the caller's name, the facts, and the case file match to the staffer. A cold transfer makes the caller repeat everything.

Check handoff windows as well. Some platforms let you set hours when staff take qualified calls and the assistant covers the rest. That setup protects your team's focus without leaving gaps.

Ask the vendor: What exactly does the staffer see on screen when a transfer arrives?

Red flag: A transfer that drops the call history.

Check 5: Confirm what lands in the case file

The call is not finished when the line goes quiet. Someone must record what happened, and that person should not be a case manager. Ask where the lead and the call notes land.

Native write-back puts structured data into the right matter fields. A summary sent by email or webhook still needs a person to file it. That gap shows up later as lost notes and wrong matter IDs.

Specify the fields you expect. A useful intake record holds the caller's details, incident date, injury summary, liability facts, referral source, and transcript link. If the vendor can only post a free-text note, expect cleanup work.

Ask to see a finished call inside your own system during the demo, whether you run Filevine, Litify, or Clio. Check the matter, the fields, the transcript, and the timestamp. Compare the result with your current process.

Ask the vendor: Show me one completed call in a live record. Which fields did it fill, and what still needs a human?

Red flag: A "Zapier integration" offered as the answer to every question about your case management system.

Check 6: Review confidentiality and data use

Callers share injuries, medical history, and fault details. That information is confidential even before they sign. ABA Model Rule 1.18 treats people who consult a lawyer about representation as prospective clients, and lawyers owe them confidentiality.

Put four questions to every vendor in writing:

• Model training: Does the vendor train any model on your calls or transcripts? The answer should be a clear no, written into the contract.
• Retention: How long are recordings and transcripts kept? Ask for a retention period you can set and a deletion process you can trigger.
• Access: Which vendor employees can open your calls? Look for role-based access and audit logs.
• Subprocessors: Which third parties touch the audio or text? Request the list and a notice period for changes.

Ask for a SOC 2 Type II report, not just a badge. The AICPA SOC reporting overview explains how CPAs assess service organizations. A Type II report covers how controls operated over a period, not a single day.

On HIPAA, HHS cloud computing guidance says a provider storing health data for a covered entity is a business associate. That provider must sign a BAA. Many PI firms are not covered entities themselves, so ask your counsel whether a BAA applies to you.

Intake callers describe medical details either way, so many firms still request one. A vendor that offers a BAA only on its top tier is telling you something about its priorities.

Red flag: "HIPAA compliant" with no signed agreement behind it.

Check 7: Confirm consent, recording, and disclosure rules

Recording laws differ by state, and some states require every party to consent. The Reporters Committee publishes a state-by-state recording law guide for phone calls. Check it against every state where you take calls.

Ask how the assistant announces itself and the recording. One short opening line can cover both. Build it into the greeting, then test it with callers in each state you serve.

Outbound follow-up adds another layer. The FCC's 2024 AI voice ruling held that AI-generated voices count as artificial voices under the TCPA. That generally means prior express consent before calling, and prior express written consent for telemarketing.

Ask whether consent capture and opt-out handling are built in. Have your counsel review the outbound workflow before launch. This is general information, not legal advice.

Ask the vendor: What does the assistant say in its first ten seconds, and how does it record consent?

Red flag: Outbound calling switched on with no consent workflow.

Check 8: Stress-test language, emotion, and audio

PI callers do not phone in on their best day. They cry, ramble, interrupt, and call from noisy roads and hospital hallways. Test for all of it.

Include a Spanish-speaking caller if your market has one. Add a heavy accent, a bad connection, and a caller who talks over the assistant. Score how often it recovers without asking the caller to start over.

Listen for tone as well. A caller who just lost a family member needs calm pacing and plain words, not a cheerful script. Have your intake lead rate ten recordings for empathy.

Ask the vendor: Which languages work in production today, not just on the website?

Red flag: Language support that disappears during the pilot.

Check 9: Price it at your real volume

Pricing models differ more than headline rates suggest. Per-minute plans reward short calls and punish long intakes. Per-call and per-customer plans cap the cost of long calls, while per-task plans charge for completed work.

Build three scenarios from your call logs: a normal month, a busy ad-push month, and a slow month. Ask every vendor to price all three in writing. Include spam, hold time, overage, and setup fees.

Here is an illustrative case. A firm taking 600 calls a month at four minutes each uses 2,400 minutes. At $0.25 a minute that is $600, and a 900-call month reaches $900.

Then price the work around the assistant. If staff still log every call by hand, the savings shrink fast. Case manager time recovered is the number that matters most.

Ask the vendor: What is my total bill in each of the three scenarios?

Red flag: A quote that leaves out minutes, setup, or integrations.

Check 10: Run a scored two-week pilot

A pilot turns opinions into numbers. Run it on live calls for two weeks. Shorter pilots mostly test setup, not steady-state behavior.

Start with after-hours and overflow calls, where the risk is lowest. Expand to business-hours calls once the numbers hold. Review ten recordings a week yourself.

Score every vendor on the same grid. The list below turns the earlier checks into pilot measurements.

• Answer rate
  • What to measure: Calls answered, including after hours
  • Suggested pass mark: Every call answered, none sent to voicemail
• Qualification accuracy
  • What to measure: Leads screened correctly against your criteria
  • Suggested pass mark: 95% of sampled calls
• Handoff quality
  • What to measure: Transfers that carry full context
  • Suggested pass mark: No caller asked to repeat facts
• Write-back accuracy
  • What to measure: Calls logged to the correct matter and fields
  • Suggested pass mark: 100% of test calls, no manual fixes
• Compliance
  • What to measure: Disclosure, recording notice, consent capture
  • Suggested pass mark: Present on every sampled call
• Caller experience
  • What to measure: Complaints, hang-ups, and drop-offs
  • Suggested pass mark: No rise versus your current process
• Staff time recovered
  • What to measure: Manual logging and callback hours saved
  • Suggested pass mark: Measurable drop each week

Pass marks are suggestions. Set your own with your intake lead before the pilot begins.

Red flags that should end an evaluation

Some findings are serious enough to stop the process. Treat any of the following as a reason to walk away.

• No live test. A vendor that only shows recordings or slides may be hiding edge cases. Insist on calls from your own phone.
• No written pricing. If a vendor will not price your three scenarios in writing, the real bill will surprise you later.
• No data commitments. Refusing to put training limits, retention periods, and deletion rights into the contract is disqualifying.
• No proof of write-back. If the vendor cannot show a finished call inside a real case record, assume your staff will do the filing.
• No PI references. Ask for a live PI deployment and a reference call. General legal experience is not the same thing.

Where HelloCounsel fits

We build HelloCounsel, so weigh this section accordingly. It is designed for PI, mass tort, workers' comp, and SSDI firms. It is not a fit for general practice firms or very small PI practices.

The Intake agent qualifies each caller on your criteria. It hands qualified callers to your intake team during windows you set and runs the full conversation outside them. It writes each lead into your CRM when the call ends.

Other agents cover what follows signing: case opening, medical records, and client updates. HelloCounsel's position is that 70% of calls don't require a human. It resolves them instead of routing them back to your team.

Agents work inside Filevine, Litify, SmartAdvocate, MyCase, CASEpeer, and other systems. Most firms go live in under two weeks from the first call. Pricing is custom, based on call and case volume, with customer data never used to train agents.

What to do in the first 90 days

Launch is the start of the evaluation, not the end. Assign one owner on the intake side and one on the case management side. Give both authority to change scripts and escalation rules.

Review a sample of recordings every week for the first month. Look for new edge cases, such as a caller type you did not test. Fix scripts quickly, and log every change.

At day 30, compare results with your pre-launch call logs. At day 90, rerun the scorecard and present it to the partners. Renewal talks go better when you hold the data.

Frequently asked questions

How many vendors should I pilot?

Two is enough, and three is the ceiling. More than that spreads your team thin. Run the vendors on different lines or time windows, then compare scorecards.

How long does go-live take?

Vendors range from same-day to several weeks. Self-serve tools launch quickly but need your configuration and testing. HelloCounsel reports that most firms go live in under two weeks.

Should the assistant tell callers it is an AI?

We recommend it. Disclosure builds trust and reduces complaints later. Check your state's rules and your bar's guidance, since requirements vary.

Does a PI firm need a BAA for intake calls?

It depends on whether your firm is a covered entity or a business associate. Many PI firms are neither, but they still handle sensitive medical facts. Ask your counsel, and favor vendors that sign a BAA or equivalent confidentiality terms.

Can an AI voice assistant replace my intake team?

For routine screening and after-hours calls, it can carry most of the load. Staff should still own accept-or-decline decisions and sensitive conversations. The assistant handles volume while people handle judgment.

What if a vendor refuses a pilot?

Treat it as a red flag. A pilot exposes edge cases that demos hide. If a free pilot is unavailable, ask for a paid one with a clear exit.

What should we measure after go-live?

Track missed calls, qualified leads, signed cases, and case manager time recovered. Compare each number with your pre-launch call logs. Review the results monthly.

Conclusion: Buy the test results, not the demo

The right assistant passes your tests, not the vendor's script. Define the calls, test the edges, inspect the case file, and read the data terms. Then let a two-week pilot settle the decision.

Keep the scorecard from the pilot. It gives you leverage at contract time and a baseline for every quarterly review after launch. Firms that measure from day one catch drift before callers do.

To see these checks play out on a real PI call flow, book a demo with HelloCounsel. Bring your call logs and your toughest test scenarios. We will show you what resolved calls and case file write-back look like in your own system.