Renewal calls that end with a price
An AI phone agent called customers due to renew their car insurance, pulled live prices from several insurers during the call, and read them the best offer — all before the caller hung up.
- Client
- Car insurance broker
- What the AI does
- Calls customers due to renew
- Calls per month
- ~40,000
- Time to launch
- 6 weeks
On 9 in 10 qualified calls, the AI asked insurers for live prices
Calls where a customer picked up
Customer and car details captured correctly
Qualified calls where the AI asked insurers for prices
Of those, customers who heard the best price
Typical call length
Customers who opened the texted price link
Estimated savings once all ~40,000 monthly calls go through the AI — not counting savings on supervisors, office space, or staff turnover
Too many calls, too little time
- About 40,000 comprehensive car insurance renewal calls a month, aiming for ~10% to renew.
- Human agents were paid even for calls that went nowhere.
- Some customers could miss their renewal window before anyone reached them.
- Customers were easiest to reach on evenings and weekends — when the fewest agents were working.
- Customers had to give permission before any personal or car details were collected.
Live prices, right on the call
- Gets consent under the applicable privacy law, then confirms the customer's ID, plate, and registration.
- Gets live prices from several insurers while the customer is still on the phone.
- Reads out the best offer — insurer, price, coverage, and payment plan — then texts a link to the quote.
- Answers concerns, saves each call's outcome and transcript, and hands off to a person on request.
- Payment, card security checks, card details, and issuing the policy stay in the broker's existing online checkout.
The win came from finishing the job on the call — a real price, not just a check on interest. Hearing the price first made the follow-up text a clear next step, not a mystery link.
These are pilot results: 96 supervised test calls plus a limited number of live calls through September 2, 2026. A comparison of purchases against a group called only by human agents was still underway. Client name withheld for confidentiality.
What happens on each call
- 01
Ask permission
Gets the customer's OK before collecting any details.
- 02
Confirm details
Checks identity, ID number, license plate, and registration.
- 03
Get prices
Asks several insurers for prices in real time.
- 04
Share the best deal
Reads out the best insurer, price, coverage, and payment options.
- 05
Wrap up
Texts the quote link, saves the result, and hands off to a person if asked.
From answered call to best price
Calls where a customer picked up
Customer and car records created
Live price requests sent to insurers
92.1% of qualified calls
Best price read to the customer
94.3% of calls with a price request
Minutes on the phone with customers
Times a do-not-contact rule was applied
Calls handed to a live agent
What the $9,100 estimate includes
If the AI handled all ~40,000 calls a month, the broker would save about $9,100 a month in operating costs.
For calls the AI handles, the broker no longer pays for:
- Agents waiting between calls
- After-call wrap-up work
- Breaks
- Minimum call-volume commitments
- Monthly platform fees
Not counted in the $9,100:
- Possible savings on supervisors
- Office space
- Staff turnover
Live in six weeks
- Step 1
Get set up
System access, choosing which customers to call, and approving the call script.
- Step 2
Build
Connected the AI to the quoting system, ranked insurer offers, added retries and backups, and tested the texts.
- Step 3
Test and launch
96 supervised test calls, then a live pilot checked daily.
Four key design choices
- 01
Give the price on the call. Don't end with a callback or a routine hand-off to another agent.
- 02
Say the price before sending the link. Customers hear the price and payment options first.
- 03
Only act after permission. Request prices only after the customer agrees and their details are confirmed, so hang-ups don't set off unwanted texts.
- 04
Track results from day one. Tag every AI call and link, and compare results against a group handled only by human agents.
What got in the way — and the fix
- About half the original customer list was already outside the renewal period.
- Removed those customers before measuring results.
- ID numbers and license plates that mix letters and numbers could lead to a quote for the wrong car.
- The AI reads them back for the customer to confirm — about eight extra seconds, but it prevents wrong-car quotes.
- Insurers took different amounts of time to send prices.
- The AI keeps the conversation going during the wait and has a set plan if an insurer never responds.
- Customers often interrupted the opening disclosure to ask for the price.
- Tuned how the AI handles interruptions.
What the team learned
- Tracking which sales came from the AI has to be built in from the start, not added later.
- Measure results two ways: across all calls, and across calls where a quote was actually possible.
- Reading back IDs and plates is a cheap way to avoid costly mistakes.
- Plan for hang-ups and permission rules before designing the ideal call.
Where it goes from here
- Finish the week-eight sign-off review.
- Compare purchases against the group handled only by human agents.
- Improve on the 31.4% of customers who opened the price link.
- Start tagging customer concerns in a consistent way.
- Explore follow-up calls to win back customers who got a quote but didn't buy.
More from financial services and insurance
Routine member calls, handled safely
A Medicare plan put an AI assistant on routine member and provider calls — reading from enrollment, claims, approvals, and appeals systems, verifying every caller first, and never changing a record.
Read the case studyFinancial & Insurance
How Dottie handles regulated, high-volume phone work for carriers, brokers, health plans, and lenders.
See the overview