A three-person personal injury firm in Austin, Texas cut their new client intake process from an average of 4.2 hours per case down to just over 2 hours — without adding any staff — by wiring together three AI tools and one automation layer. This is exactly how they did it, what broke, and what you should steal for your own operation.
The Problem That Was Actually Costing Them Money
The firm — call them Reyes Legal, since they asked not to be named — was handling around 40 new potential clients per month. Every intake meant a paralegal manually typing notes from a phone call into their case management system (they used Clio), then copying details into a Google Sheet, then drafting a follow-up email, then scheduling a consultation manually. Four steps, each requiring a human to context-switch.
The paralegal was spending roughly 17 hours a week just on intake-related tasks. At $28/hour, that’s about $476 a week or close to $25,000 annually — on admin that generated zero billable hours. That number hit different when they saw it written out.
They weren’t a tech firm. They weren’t flush with budget. But they were sick of losing potential clients because follow-up emails went out 24 hours late. So they called me.
The Stack They Built (and Why Each Tool Was Chosen)
Fireflies.ai for Call Transcription and Summarization
Fireflies.ai sits inside every intake call and produces a transcript plus an AI-generated summary within minutes of the call ending. For Reyes Legal, this was the anchor of the whole workflow. The bot joins the call automatically once you connect it to Google Calendar — no one has to remember to hit record.
The honest pitch on Fireflies: the transcription accuracy is genuinely good for standard American English, hitting around 90–95% on clean audio. It struggles with heavy accents or bad phone connections, and their personal injury clients calling from construction sites or in emotional distress sometimes produced transcripts that needed a light human pass. The Business plan runs $19/user/month, which for a three-person firm is $57/month — cheap enough that the ROI math takes about 30 seconds to do.
What made it worth it wasn’t just the transcript. Fireflies generates a structured summary with action items. They customized the prompt template in the settings to specifically extract: potential case type, incident date, insurance status, and whether the caller had spoken to another attorney. That last field alone saved the paralegal from writing the same four questions in follow-up emails every single day.
Make.com as the Automation Layer
Once Fireflies drops a summary, something has to actually move that data somewhere useful. This is where Make.com (formerly Integromat) comes in. Make watches for new Fireflies summaries via webhook, parses the structured output, and routes the data into both Clio and a Google Sheet simultaneously.
If you’re comparing Make to Zapier for something like this, Make wins on flexibility and price. The scenario they built had conditional logic — if “insurance status” came back as “uninsured,” it triggered a different email template and flagged the intake in red in the spreadsheet. Zapier can do conditional logic, but you pay per task and it adds up fast at volume. Make’s Core plan at $9/month covered their usage with room to spare. (For a more detailed look at how these platforms stack up, Zapier vs. Make.com is worth reading before you commit to either.)
Setting up the Make scenario took about 90 minutes the first time, most of it figuring out how Fireflies structures its JSON output. Not hard, but not magic either. If you’ve never touched a webhook before, give yourself an extra hour and read this plain-English breakdown of what webhooks actually do before you start.
GPT-4o for Follow-Up Email Drafting
The third tool is OpenAI’s GPT-4o, called via API inside the Make scenario. After the intake data hits the Google Sheet, Make sends a prompt to GPT-4o that includes the extracted details from Fireflies — case type, caller name, incident summary — and asks it to draft a personalized follow-up email in the firm’s voice.
They did not auto-send these emails. That was a deliberate choice, and a smart one. The attorney reviewed each draft in a shared Gmail label, made any edits, and hit send. The draft was already 80% there. Total review time: under three minutes per email, down from 15–20 minutes of writing from scratch.
API costs for GPT-4o at their volume — about 40 emails per month, each prompt around 600 tokens — ran under $3/month. That’s not a typo. At $2.50 per million input tokens, this part of the workflow costs almost nothing.
How the Workflow Actually Runs, Step by Step
- Prospective client calls the firm’s intake line. Fireflies bot is already in the calendar event and joins automatically.
- Call ends. Fireflies generates transcript and structured summary within 3–5 minutes.
- Make.com detects the new summary via webhook trigger, parses the JSON, and extracts the four key fields.
- Make creates a new matter draft in Clio and populates the Google Sheet row simultaneously.
- Make sends extracted data to GPT-4o with a firm-specific prompt template, receives the draft follow-up email.
- Draft appears in a shared Gmail label called “Intake Drafts” for attorney review.
- Attorney approves or edits and sends. Total human time at this stage: under 5 minutes.
End to end, the paralegal’s role in intake dropped from a multi-hour task to under an hour: handling the actual call, doing a quick transcript check, and any follow-up that required legal judgment. The 52% time reduction they measured over the first 90 days was conservative — some weeks it was closer to 60%.
What Broke (Because Something Always Does)
About three weeks in, Fireflies had a brief API outage that lasted most of a Tuesday. Two intake summaries never triggered Make. They only caught it because the Google Sheet didn’t update. The fix was simple: the paralegal now does a quick end-of-day check that the row count in the Sheet matches the day’s call log. One minute of human QA. Automation doesn’t eliminate oversight — it just changes what you’re watching for.
The GPT-4o email drafts occasionally over-promised. One early draft said “we will review your case immediately” — language the firm’s attorney was not comfortable sending. They solved this with a system prompt addition: “Do not make any commitments about case outcomes, review timelines, or attorney availability. Use hedged language.” One prompt edit, problem gone.
Clio’s API integration through Make was also finicky about required fields. If Fireflies failed to capture the caller’s last name (it happened with a few one-name callers), the Clio step would error out. They added a Make error handler that sends a Slack message to the paralegal when any step fails. Now failures are visible in real time instead of discovered days later.
The Numbers After 6 Months
Reyes Legal ran this workflow for six months before I asked them to pull actual data. Their paralegal’s intake-related hours dropped from 17 hours/week to about 8. That’s roughly $12,700 in recovered time annually at her rate. The entire stack — Fireflies, Make, OpenAI API — costs them under $70/month, or about $840/year.
Net first-year savings after tool costs: approximately $11,860. And that’s before you count the cases they stopped losing to slow follow-up. They estimated two or three additional signed clients per quarter directly attributable to faster response time. At their average case value, that’s a significant multiplier on a $70/month investment.
People underestimate this kind of return because they’re looking for flashier automation. If you want to understand why small-scale workflows like this often beat ambitious AI projects on ROI, the common myths about AI automation ROI are worth a careful read before you plan your next project.
What You Can Take From This Right Now
You don’t need to be a law firm for this to apply. Any service business with a repeatable intake or discovery call — accountants, consultants, agencies, healthcare practices — can run an almost identical workflow. Swap Clio for your CRM, keep the rest.
The principle is: transcribe automatically, extract structured data automatically, draft communication automatically, but keep a human in the loop for the send decision. That’s not timidity — that’s how you avoid the errors that damage client relationships before they start.
Start with Fireflies and one Make scenario. Don’t try to automate everything at once. Get the transcript-to-CRM step working reliably for two weeks before you add GPT-4o drafts. Boring advice, but the firms that try to build the whole thing on day one are the ones who end up abandoning it by week three.
FAQ
Does this workflow work if my firm uses a CRM other than Clio?
Yes. Make.com has native integrations for HubSpot, Salesforce, Zoho, and dozens of other CRMs. The logic of the scenario stays identical — you’re just swapping the Clio module for whichever CRM module fits your stack. The Fireflies and GPT-4o steps don’t change at all.
Is it safe to run client intake data through tools like Fireflies and OpenAI?
This is the right question to ask, and the answer requires some homework on your end. Fireflies offers a Business Associate Agreement (BAA) for healthcare use cases, and their data is stored with encryption, but you should review their data retention policy against your jurisdiction’s client confidentiality rules. OpenAI’s API data is not used for training by default when accessed via API, but you should confirm this with legal counsel for your specific profession. Many law firms using AI tools run everything through their own Microsoft Azure OpenAI instance for tighter data control.
How long does it take to set this up if I have no technical background?
Realistically, a full weekend or about 8–10 hours spread across a few evenings. Fireflies setup takes under 30 minutes. The Make scenario is the steepest part of the curve — plan for 3–4 hours including troubleshooting. If you get stuck, Make’s template library has a Fireflies-to-Google Sheets scenario you can use as a starting point rather than building from scratch.
This article was produced with the assistance of AI, and its featured image was AI-generated. We review for accuracy, but please verify critical details.
