An established insurance agency had eight employees and thousands of policyholders. Most customer communication had long since moved to WhatsApp: proof-of-insurance documents, claim questions, detail updates, and renewal reminders. Two WhatsApp numbers, one for the office and one for the owner, received hundreds of messages each week.
Starting point: everyone replied, but nobody owned the case
Messages were answered by whoever was available. A customer who asked a question at noon sometimes received a reply after two hours, sometimes the next day, and sometimes twice from two different employees. When a key employee went on maternity leave, response times rose and the owner found himself answering messages in the evenings.
Our discovery process revealed three clear findings:
- Nearly 60% of messages were recurring questions: proof of insurance, claim status, opening hours, and documents needed for renewal.
- Average initial response time was a little over two hours.
- There was no way to know whether an inquiry had been handled to completion. Anything not answered that day disappeared.
What we built
Our guiding principle was simple: customers should receive an immediate response, and the team should receive only the cases that genuinely require a person.
- One business WhatsApp number on the WhatsApp Business API instead of two personal numbers. Each customer’s history is stored in one place.
- A smart initial response that identifies the type of inquiry. Most common requests are handled end to end without employee involvement: the customer verifies their identity, and the document is sent automatically from the agency’s system.
- Routing by subject. Claims go to the claims coordinator, renewals to the renewals department, and sales inquiries to the owner. Every inquiry opens a card on a CRM service board with a status and an owner.
- Follow-up on open inquiries. A case that remains unresolved for one business day triggers a reminder for its owner. Once a week, the agency owner receives a short summary: how many inquiries arrived, how many were closed, and what is stalled.
Implementation also included two weeks of close support for the team, including writing response scripts together with the employees. The wording is theirs, not ours. That is why customers barely notice the difference between an automated response and an employee response.
Results after two months
- Initial response time fell from an average of two hours to three minutes, including at peak times.
- 55% of inquiries are resolved end to end without employee intervention, mostly requests for certificates and documents.
- Zero inquiries disappear. Every case has a card, a status, and an owner.
- The owner’s evening workload fell almost completely. In his words: “I have the headspace to talk with clients about insurance instead of searching for who answered whom.”
One important point did not change: the number of employees remained the same. The goal was not to reduce staff but to stop wasting their time on manually transferring documents.
What this story teaches
When most inquiries repeat, the problem is not the volume of work but how it is distributed. Good automation does not replace the team. It filters out what does not need them and ensures that what does need them never gets lost.
Questions we are asked about this kind of process
Do customers get frustrated by automated responses? When the response solves the problem in seconds, the opposite is usually true. We set one rule: at any stage, a customer can type “representative” and move to a person. In practice, most customers prefer to receive the document immediately.
What happens with older customers who call by phone? Nothing changed for them. The phone remains available, and the team has more capacity to answer it because WhatsApp no longer consumes all its attention.
How long does it take employees to adjust? Here, it took two weeks. The key was that the team wrote the response scripts themselves and felt ownership of the tool rather than threatened by it.
The client’s identifying details were removed at their request. The data was measured with the client before and after the project.
