Core platform modernisation for a Canadian fintech

The platform that carried the company through its first five years could not carry the next five. Deployments took a weekend and were rolled back roughly one time in five. French language support had been retrofitted and was visibly broken in several journeys, which mattered because Quebec was the company's fastest growing market and Law 25 had raised the stakes on consent and disclosure flows that were partly untranslated. Adding a lending product took months. The company's own engineers spent most of their time on maintenance and were leaving because of it.

We placed a dedicated nine person pod, seven in Ontario and two distributed specialists, working inside the client's delivery process rather than alongside it. The pod took a strangler approach, extracting capabilities from the monolith one bounded context at a time behind a stable API layer, so the business kept shipping instead of pausing for a rewrite.

Priority was set by risk rather than architectural preference: ledger and payments first, then identity and onboarding, then the lending engine, then the customer application. The French and bilingual rebuild was treated as a first-class workstream with its own design review, covering not only interface strings but consent flows, disclosure documents and notification templates, which is where the Law 25 exposure actually sat.

We rebuilt the delivery pipeline in parallel, introduced automated testing where there was almost none, and put observability in place before the first extraction went live. SR&ED-eligible development activity was documented contemporaneously in partnership with the client's finance team, which materially improved their claim position.

Dedicated pod, hybrid Ontario, embedded in the client's agile process with shared standups, shared backlog and joint architecture review.

Deployment frequency moved from roughly monthly to several times weekly. Change failure rate fell from around 20 percent to under 4 percent. Time to launch a new lending product reduced from months to weeks. Automated test coverage on extracted services reached 81 percent from a near-zero start. Quebec user engagement rose significantly after the bilingual rebuild. The client's own engineers returned to product work and voluntary attrition on the engineering team fell by more than half.

Two other firms had quoted a full rewrite, which would have frozen the product for a year in a market that was moving. Incremental extraction was slower on paper and faster in reality.