- Sector
- Insurtech · Leadership
- Year
- 2024
- Role
- Fractional CTO
- Outcome
- Stabilised a platform that needed weekly restarts · LLM document pipeline · delivery function built from scratch
What needed fixing.
The symptoms were a portal that needed restarting every few days and long-standing failures in core insurance flows — policy changes, revocations — that had never been properly resolved. Underneath sat the real problem: no technical-PM function, no structured QA, thin documentation, and overlapping vendors whose responsibilities had never been pinned down. Features shipped and regressed in roughly equal measure.
How I built it.
First, stop the bleeding: I stabilised the platform so the portal stopped needing restarts and worked through the years-old bugs in the policy-lifecycle flows. Then the foundations — I stood up the technical-PM function from scratch (Jira, Confluence, structured QA, real documentation across the business-critical areas) and split delivery into parallel development and support streams, so shipping new work stopped breaking the old. I replaced a legacy OCR step with an LLM-based document pipeline that fixed long-running data-quality issues, stood up a read-replica of the upstream Insly database feeding an optimised analytics path and a Power BI layer, imported historical data with a UI to work with it, and pinned down who owned what across the vendors. Then I helped release the platform and saw the migration and follow-up fixes through.
A European insurtech whose customer portal needed restarting every few days. Owned the technical strategy, stopped the restarts, fixed years-old bugs, and built the delivery function that should have existed already.
- 01Stabilise a platform whose customer portal needed restarting every few days
- 02Resolve long-standing bugs in policy-lifecycle flows (changes, revocations)
- 03Build the missing engineering discipline: QA, documentation, delivery process
- 04Improve data quality and give the business usable analytics
- 05Define vendor responsibilities and release the platform
- Platform stabilisation — eliminated multi-day uptime issues and weekly restarts
- Technical-PM function built from scratch: Jira, Confluence, structured QA, documentation
- Delivery split into parallel development and support streams — fewer regressions
- Legacy OCR replaced with an LLM-based document pipeline; data-integrity issues closed
- Read-replica of the upstream Insly database feeding a Power BI analytics layer
- Historical-data import with a purpose-built UI; vendor responsibilities defined
Owning the technical strategy, end to end
A European insurtech platform that needed restarting every few days is not a software problem you can fix by writing more software. It's a leadership problem. I came in as fractional CTO and took the whole technical strategy — which meant the unglamorous first win was simply this: the customer portal stopped falling over. The multi-day uptime issues ended, and then, with the patient stable, the actual work could start.
Two streams, so the firefighting stops eating the roadmap
The first structural change was to stop letting support and delivery fight over the same people. I split the work into two parallel streams — feature delivery on one, support and stabilisation on the other — so a production fire on Tuesday no longer set the roadmap back a fortnight. Fewer regressions, protected velocity, and shipped features that stayed shipped. Around it I stood up the technical-PM function from scratch: Jira, Confluence, structured QA, real documentation across the business-critical flows.
Delivery and support, split into parallel streams. The roadmap stopped paying for every fire.
A read-replica of the upstream platform feeding an optimised analytics path and Power BI.Replacing the guesswork with a pipeline
Two pieces of real engineering stand out. The legacy OCR — the source of a long tail of data-integrity problems — was replaced with an LLM-based document pipeline that materially improved data quality and closed issues that had been open for years. And to get the business its numbers without hammering the live system, I stood up a read-replica of the upstream Insly database feeding an optimised analytics path and a Power BI layer. Years-old policy-lifecycle bugs — the MTA and revocation flows nobody wanted to touch — got fixed along the way. Then we released it, and stayed to clean up after the migration. Strategy is only worth anything if the platform is still standing at the end.
