No. 02 · Legal services
Plan / 8 business days
Intake call summaries the partner can actually read — small law firm
A vendor wanted $1,400 a month for a "legal AI platform." The partner needed four bullet points. We wrote the plan, the firm built it, the bullets land in his inbox every morning.
Client: 3-attorney personal injury firm (anonymized)
Three-attorney plaintiff firm, mostly auto and slip-and-fall. The managing partner is the bottleneck on every intake — nothing moves until he's read the call notes and decided whether to take the case. He was reading them after hours because he had no time during the day.
The brief
"I get fifteen intake calls a day. Paralegal types up half a page on each. I read all of them at 9pm. A vendor pitched me a $1,400-a-month platform that will, near as I can tell, do the same typing the paralegal already does. I just need a real summary."
What we found
Partner was spending 90 to 110 minutes a day reading paralegal write-ups — most evenings, after the kids were in bed.
The pitched legal-AI platform charged $1,400/mo and replaced the paralegal's notes with longer, structured notes. Did not actually summarize.
Paralegal was the slowest part of the intake loop — calls finished at 4pm, write-up landed at 7pm.
Firm already recorded every intake call (Clio's built-in recording). Audio sat unused in the file.
Partner agreed that what he needed from each call was always the same four things: facts, injuries, fault posture, statute clock.
Partner time on intake review
Daily minutes spent reading intake notes, before and after.
“I got my evenings back. The plan cost less than one billable hour and we built it ourselves.”
What we built
- 01
Define "good summary" before building anything
Pulled the partner's last twenty intake decisions. For each, we asked what one paragraph would have let him decide in under two minutes. The answer was always those same four bullets. The plan locked that format.
- 02
Use what's already there
Clio recordings. OpenAI Whisper for transcription, $0.006/minute. Claude for the four-bullet summary using a fixed prompt. Gmail to deliver it. Zero new platforms.
- 03
Hand the firm a working blueprint
Twelve-page plan: exact prompt, exact API calls, the Make.com scenario diagram, error-handling notes, and a rollback path. Their tech-comfortable paralegal built it in a weekend.
- 04
Two follow-up calls during build
Week 2 to debug the Whisper handoff (their recordings were stereo, Whisper preferred mono). Week 3 to tune the prompt — the partner wanted statute clock first, not last.
- 05
Hold the partner to skim-only
We told him: if you find yourself opening the transcript, the summary failed. After two weeks of tuning, he was opening transcripts on roughly one call in fifteen.
Numbers
- plan delivered
- 8 days
- setup cost (firm built it)
- $0
- ongoing API cost
- ~$22/mo
- to skim a summary (was ~7 min)
- 90 sec
Outcome
Partner reclaimed roughly 7 hours a week. Firm spent $149 once and runs at ~$22/mo in API cost. They've since asked for a Build engagement to do the same thing for medical record summaries — we scoped that as a $549 Build because the medical-record context window is heavier and we wanted to do it right.
What this cost
Plan tier · $149 fixedPlan tier, $149 fixed. The firm pays roughly $22/month in OpenAI Whisper and Claude API usage. No platform subscription.
Stack
- OpenAI Whisper (transcription)
- Claude (summarization)
- Clio call recording (already in place)
- Make.com (glue)
- Gmail
Want a plan like this?
A $149 plan you can build yourself, or hand to your team.
A senior engineer responds personally within one business day. If there isn’t a fit, we’ll tell you and point you somewhere better.