Quoting tool for a commercial insurance broker

The brokers priced commercial policies in a spreadsheet that had been extended for a decade by whoever needed a new line of cover. It was accurate in the hands of the two people who understood it and a hazard in anyone else's, and onboarding a new broker took months of shadowing. When a rate changed, someone edited the master and emailed it around, so at any moment there were several versions in use. The firm wanted one tool, an audit trail of which rates produced which quote, and the ability to give a junior broker something they could not silently break.

Contract full-stack developer for about eight months on a team of three, working under a product owner who was himself a former broker — which made requirements sessions unusually fast and unusually opinionated. I did most of the backend and the calculation engine; a second contractor did the front end; the firm's own developer handled the integration into their document generation. Sprints were two weeks and the product owner sat in the same room, which I would take over any amount of written specification.

I owned the rating engine and the versioning around it — rates are data with an effective date range, not code, and a quote records the exact rate set it was priced with so it can be reproduced years later. I built the quote lifecycle, the referral path for cases that need an underwriter's judgement, and the comparison view that shows a broker why two quotes differ. I also did the extraction of the spreadsheet's logic into rules, which meant sitting with the two brokers who understood it and working through cases until we agreed on what it did, including three places where it had been quietly wrong.

The central decision was making rates data rather than code, with an effective-dated table and an approval step, so a rate change is a business action rather than a deployment. The alternative — the one the firm expected — was hard-coded rules and a release per change, which is faster to build and would have handed them the same bottleneck in a new form. I also decided a quote should store a full snapshot of its inputs and rate set rather than references, because a quote is a document that gets argued about eighteen months later. That duplicates data and I would make the same call again.

Every rate change had to be attributable to a named approver, and the firm's compliance officer reviewed the audit design before we built it. There was a hard date: the tool had to be live before the renewal season began, because the brokers would not learn a new tool during it. We could not touch the document generation system, so the tool had to produce exactly the payload it expected, quirks included. And the two brokers who held the knowledge were fee-earners, so their time came in ninety-minute blocks I had to use well.

Quote preparation went from around forty minutes to about twelve for a standard commercial case, measured by the product owner on a sample before and after. More importantly, a new broker was producing quotes in their second week rather than their third month. The three inherited errors we found in the spreadsheet had been mispricing one class of cover slightly low for years, which the firm was not delighted to hear and was glad to stop. It went live three weeks before renewal season and I stayed on-call through the first two weeks of it.

The product owner made every judgement call about how a broker actually works and vetoed several things I thought were improvements. The two senior brokers effectively wrote the rules with me. The front-end contractor built the comparison view, which is the screen brokers spend their day in and is better than my design for it was.