Legal Tech and AI

Measuring the ROI of AI Intake Tools

A before-and-after plan to measure the ROI of AI intake tools at a law firm: baseline metrics, cost and value formulas, a worked example and common traps.

AssistCall research teamPublished Reviewed by Faruque Amin, VP of Sales and Marketing6 min read

On this page
  1. Why most firms cannot say whether it worked
  2. Define the question first
  3. Capture a baseline before switch-on
  4. The metrics to track
  5. The cost side
  6. Value staff time by role
  7. Value additional signed matters
  8. A worked example
  9. Traps that distort the result
  10. Measures that are not money
  11. When several things changed at once
  12. Questions partners will ask
  13. Review at 30, 60 and 90 days
  14. Report it in one page
  15. Frequently asked questions
  16. References
Legal Tech and AI: Measuring the ROI of AI Intake ToolsArticle coverLEGAL TECH AND AIMeasuring the ROI of AIIntake ToolsAva for LawyersInsightsLegal Tech and AI: Measuring the ROI of AI Intake ToolsArticle coverLEGAL TECH AND AIMeasuring the ROI ofAI Intake ToolsAva for Lawyers

Key takeaways

  • Only 20% of legal professionals surveyed by Thomson Reuters measure the ROI of generative AI[1].
  • Capture a 30-day baseline before switching anything on, or you will have nothing to compare.
  • Count value in two places: staff time saved, valued by role, and additional signed matters.
  • Control for seasonality and marketing changes before crediting the tool.

Why most firms cannot say whether it worked

Firms buy intake tools to answer more calls, book more consultations and free staff time. A few months later, most cannot say whether any of that happened. Thomson Reuters found only 20% of legal professionals are measuring the return on investment of generative AI[1]. The usual reason is not lack of interest. It is that nobody wrote down the starting numbers before the tool went live.

Legal professionals measuring the ROI of generative AI, 2025
Legal professionals measuring the ROI of generative AI, 2025Donut chart. Measuring ROI: 20%.Measuring ROI: 20%20%
  • Measuring ROI 20%

Source: Thomson Reuters Institute, 2025[1]

Show the data
ItemValue
Measuring ROI20%

A useful measurement plan is short. It defines a few metrics, records them for a month before launch, records them again after, and accounts for the obvious reasons numbers change on their own.

Define the question first

"Is the tool worth it?" is too vague to measure. Make it specific: "Does answering after-hours calls with an AI intake tool increase signed matters from phone inquiries, and how much staff time does it save or cost?" A specific question tells you which numbers to collect, which period to compare and what result would make you keep, change or drop the tool.

Capture a baseline before switch-on

  1. Pick the period

    Use the 30 days before launch, and note anything unusual, such as a holiday or a marketing campaign.

  2. Count inquiries by channel and hour

    From the phone system and web forms: total new inquiries, and how many arrived outside office hours.

  3. Count missed and abandoned calls

    Calls that went to voicemail or hung up, especially from new numbers.

  4. Record conversion steps

    Consultations booked, consultations held and matters signed from those inquiries.

  5. Measure response time

    Median time from first contact to first meaningful response.

  6. Log staff time

    Ask intake staff to log time spent on calls, callbacks and voicemail for two typical weeks.

The metrics to track

Track the same metrics before and after, over the same length of period.
MetricHow to measureSource
New inquiriesCount by channel and by hourPhone system, forms, CRM
Missed new-inquiry callsCalls to voicemail or abandonedPhone system
Median time to first responseFirst contact to first real replyCRM or call logs
Consultations booked per inquiryBookings divided by new inquiriesCalendar and CRM
Matters signed per consultationSigned engagements divided by consultations heldPractice management
Staff intake timeHours per week on calls, callbacks and voicemailTime log
Review time for AI recordsHours per week reviewing summariesTime log
Errors and complaintsMaterial summary errors and caller complaintsReview log

The cost side

  • The tool's subscription and any usage-based charges, as quoted to your firm.
  • Set-up time: writing scripts, testing and approvals, valued at the rates of the people who did it.
  • Ongoing review time: the daily check of summaries and the monthly supervision review.
  • Phone system changes, if any.
  • Training time for staff who use the outputs.

Review time is the cost most often left out. A tool that saves ten hours of answering calls but adds eight hours of checking summaries is a smaller win than it looks.

Value staff time by role

Time saved is worth what the person's time costs. The median hourly pay was $76.76 for US lawyers and $18.27 for receptionists in May 2025[2][3], and paralegals and legal assistants earned a median of $62,890 a year[4]. Employer costs run higher than wages: benefits made up 30.0% of private employer compensation costs in June 2026[5]. Use your own payroll figures where you can, and value each hour saved at the rate of the person who no longer spends it. An hour a lawyer no longer spends returning voicemails is worth several times an hour of receptionist time.

Value additional signed matters

The larger value is usually in matters you would otherwise have lost. Reachability is part of the reason: Pew found 67% of Americans generally do not answer calls from unknown numbers but will check a voicemail[6], so callbacks to people who reached voicemail often fail. The formula is simple: additional signed matters equals the change in signed matters per inquiry multiplied by inquiries, and value equals additional matters multiplied by your average fee per matter in that practice area. Use net fees after referral fees and costs, and be conservative.

A worked example

Example (fictional): a three-lawyer injury and employment firm records a baseline month: 120 new phone inquiries, 30 missed after hours, 36 consultations and 12 signed matters. Staff log 22 hours a week on calls and voicemail. After launching an AI intake tool for evenings, weekends and overflow, a comparable month shows 124 inquiries, 4 missed, 45 consultations and 15 signed matters. Staff log 14 hours a week on calls plus 4 hours reviewing summaries, a net saving of 4 hours a week. The firm credits the tool with about 3 additional signed matters a month, values them at its conservative average net fee, adds the staff time saved at payroll cost, and compares the total with the tool's cost plus set-up and review time.

Example check: before crediting the tool, the firm compares the same month a year earlier and confirms no marketing campaign started during the test. Signed matters in the prior-year month were flat, which supports the comparison.

Traps that distort the result

  • Seasonality. Injury intake rises in summer; family law has its own peaks. Compare like months or use a control period.
  • Marketing changes. A new ad campaign during the test will raise inquiries regardless of the tool.
  • Counting calls instead of clients. More answered calls only matter if they become consultations and signed matters.
  • Ignoring quality. A tool that books more consultations that should never have been booked adds cost.
  • Short tests. One week is noise; a month is a minimum for most small firms.
  • Forgetting review time on the cost side.

Measures that are not money

Some benefits do not show in fees. Ask staff whether the work is more manageable, track caller complaints and compliments, and note whether lawyers are interrupted less during focused work. Watch for negatives too: callers who dislike speaking to an AI, summaries that need heavy correction, or escalations that failed. A tool that makes the numbers look good while staff quietly redo its work is not a success.

When several things changed at once

Real firms rarely change one thing at a time. If you launched the tool in the same month you redesigned the website or hired a new intake person, you cannot cleanly separate the effects. Two simple techniques help. Compare channels the tool touches with channels it does not, such as after-hours calls against daytime web forms. Or switch the tool on for one practice area or one phone line first, and compare it with the others for a month.

Questions partners will ask

  1. How many more matters did we sign, and how sure are we the tool caused it?
  2. What did it cost in total, including staff time to set it up and review it?
  3. Did anything go wrong, such as a missed escalation or a complaint?
  4. What would happen if we switched it off tomorrow?
  5. What would we change to get more from it?

Answer them in the one-page report before anyone asks. It shortens the meeting and makes the decision easier to revisit later.

Review at 30, 60 and 90 days

  1. Day 30

    Compare with the baseline. Fix configuration problems: scripts, escalations, booking rules. Do not judge value yet.

  2. Day 60

    Compare again. Check the trend in signed matters and staff time. Reduce review time if error rates are low.

  3. Day 90

    Decide: keep, change or stop. Write down the reason and the numbers behind it.

Report it in one page

Give partners a single page: the question, the baseline and after numbers for each metric, the cost, the value of time saved and additional matters, the main caveats, and a recommendation. Keep the raw logs in case anyone wants to check. Repeat the measurement whenever you change the tool's scope, such as adding a practice area or a new language, because the answer for one use does not automatically carry over to the next.

Frequently asked questions

How do I measure the ROI of an AI intake tool?

Record a 30-day baseline of inquiries, missed calls, bookings, signed matters and staff time, then compare a similar period after launch, counting the tool's full cost including review time.

How many firms measure AI ROI?

Only 20% of legal professionals surveyed by Thomson Reuters said they measure the return on generative AI[1].

How long should the test run?

At least a month after launch, with reviews at 30, 60 and 90 days, and a comparison with a similar baseline period.

What is the most common mistake?

Not capturing a baseline before launch. Without one, there is nothing reliable to compare.

References

  1. Thomson Reuters Institute (2025). 2025 GenAI report: Executive summary for legal professionals (opens in a new tab) โ†‘

  2. U.S. Bureau of Labor Statistics (2025). Lawyers: Occupational Outlook Handbook (opens in a new tab) โ†‘

  3. U.S. Bureau of Labor Statistics (2025). Receptionists: Occupational Outlook Handbook (opens in a new tab) โ†‘

  4. U.S. Bureau of Labor Statistics (2025). Paralegals and Legal Assistants: Occupational Outlook Handbook (opens in a new tab) โ†‘

  5. U.S. Bureau of Labor Statistics (2026). Employer Costs for Employee Compensation - June 2026 (opens in a new tab) โ†‘

  6. Pew Research Center (2020). Most Americans don't answer cellphone calls from unknown numbers (opens in a new tab) โ†‘

Book a demo

See Ava handle your calls.

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

Book a demo