Case study · insurance
Building an AI-enabled independent insurance agency.
- Client
- Goodwill Financial: the founder’s independent P&C agency and life MGA, Elmhurst, Illinois
- Status
- Our own business · ongoing
- Systems
- Agency management system, CRM and dialer, Microsoft 365 mailboxes, phone system, carrier portals
The problem
An independent agency runs on disconnected systems: an agency management system, a CRM, phones, a shared service mailbox and a dozen carrier portals. Requests arrive everywhere, and work waits between people and systems. The owner cannot see where time goes or where business leaks, because nothing measures it.
What it cost the business
We did not know, and that was the first finding. No one had measured certificate turnaround, call follow-through or renewal coverage. Where we could, we measured before changing anything; where we could not, we say so.
Before
- Certificate requests answered from the shared mailbox with no clock on them.
- Call outcomes logged, but the CRM did nothing with them: dial lists never got shorter.
- Intake forms re-keyed by hand into the agency system.
- No weekly view of the book: new, lost and renewing business by carrier and line.
What we built, and its status today.
| Component | What it does | Status |
|---|---|---|
| Certificate turnaround measurement | Certificate requests in the shared service mailbox matched to the certificate that answered them where both ends were clear (407 of 621), timed in business hours, split by who handled it. | Measured |
| Call review and coaching | A weekly audit per person: recorded calls transcribed and scored against that person’s own script, coaching drafted for the manager to edit and send, never sent automatically. From April to August it ran nightly. | Running · weekly |
| CRM that acts on call outcomes | Each logged outcome sets the next step in the CRM: a tag, a stage, a task, or the contact leaving the dial lists. | Running |
| Intake into the agency system | Web and emailed intake forms written into the agency management system end to end, then read back from the system to confirm. Paused while the operations machine is replaced. | Built · paused |
| Book-movement board | New, lost and renewing business by carrier, line and client, rebuilt every week. | Running |
| Renewal worklist | Every term sorted into re-shop, review, accept or chase. Paused while the operations machine is replaced. | Built · paused |
| The operating layer | 170 written playbooks, 19 cloud services and 70 enabled scheduled jobs (counted October 2), with written rules for what may run on its own and what needs a person. More jobs are paused while the operations machine is replaced. | Running · some paused |
The certificate tool itself is a vendor's product that the agency chose. What we built is the measurement around it: 12,108 emails rebuilt into 621 requests, of which 407 had a measurable start and finish. The rest are reported as unpaired, not guessed. The manual path includes the unusual requests the tool does not handle, so this is not a like-for-like speed test.
Our agency also uses an outside AI vendor for its main phone line and its certificates. That vendor publishes its own figures for our agency; we report only what we measured ourselves.
AI prepares. People decide.
- Inputs
Service mailbox · phone calls · web and email forms · agency system exports
- Operating layer
Playbooks and rules · scheduled jobs · measurement and boards · drafts and worklists · scheduled jobs log what they read and change
- A person
Approves anything that commits the agency
- Out
Clients and carriers
| The system does | A person does |
|---|---|
| Pairs requests with responses and times them | Decides what to change about the process |
| Transcribes and reviews calls; drafts coaching notes | Reads, edits and sends the coaching |
| Updates CRM stages and lists from logged outcomes | Has the conversation and logs the outcome |
| Writes intake into the agency system and reads it back | Presents prices, binds and advises: always a licensed person |
What is measured, and what is not yet.
The first two rows compare the tool path with the manual path over the same months, not a before-and-after, because nothing was measured before. We say so rather than invent a baseline.
| Measure | Before | After / now | Status |
|---|---|---|---|
| Certificate turnaround, median request to certificate | Not measured before | 6 minutes with the tool we chose · 42 minutes on the manual path | Measured |
| Certificates needing rework | Not measured before | 1.1% with the tool · 5.8% manual | Measured |
| Intake re-keyed by hand | Every intake | 7 of 7 written end to end on the first live run | Measured once |
| Recorded agent calls reviewed | A handful, by ear | 2,208 scored on five parts, April to August 2026 | Measured |
| Hours returned to the team | — | Not measured yet | Baseline being built |
| Retention by household | — | Not measured yet | Next chapter |
| Net financial impact | — | Not measured yet | We will publish it with the method |
What did not work
We tried letting an automated job file incoming documents from the shared mailbox straight into the agency system. It flagged far more documents than it could file with confidence. We changed its job to flag and prepare, with a person filing; it is paused while the operations machine is replaced. We would rather publish that than pretend every automation worked the first time.
What is next
Retention by household, measured from the agency’s own book, and an hours-returned measurement for certificate and renewal work. Both will be published here with their method, whatever they show.
Published October 2026. Certificate figures measured July 1 – September 12, 2026; other counts measured September–October 2026 inside the agency. Sources and dates are kept on file.
Find the one problem worth fixing first.
A free 45-minute discovery call on Zoom with Vytas Skapcevicius, the founder. Four questions about your agency, no pitch deck. You leave knowing whether there is something worth measuring.