Case study
A refactor question became a strategic rebuild
A regulated market intelligence business whose core platform runs to roughly 500,000 lines of legacy C# and SQL.
The question
The platform underpins the company's core product, and a previous rebuild attempt had ballooned in cost and timeline. So the opening ask was modest: refactor the legacy system cheaply and buy a few more years.
Discover: five weeks, two tracks
We ran discovery on two tracks at once. The functional track mapped processes and ran AI-facilitated interviews across the platform's users. The technical track put agentic analysis through the full codebase and more than 550 tickets. Five weeks later the programme had a grounded picture of the platform's real behaviour and a set of phasing options.
Prove: 12 weeks to an investment decision
The proving phase ran two workstreams in parallel over 12 weeks. One produced the plan: a functional blueprint of 54 features across 10 domains, a target-state architecture, an integration inventory, a phasing proposal and delivery sizing estimated bottom-up from the code itself. The other built a working AI prototype: voice transcription, AI extraction and AI-assisted analysis across three product areas, tested hands-on by the client's own editorial team.
The prototype build, by the numbers
1,000
commits in eight weeks
202
pull requests merged
85%
of the developers' commits AI co-authored
0.2
days median cycle time across 258 tickets
Six milestones, all delivered on schedule, by a team of seven that included three of the client's own engineers. Every merge still required a passing gate of lint, type checks and unit tests with coverage, end-to-end tests ran on every pull request and infrastructure and schema work carried senior sign-off on top of the AI and human review applied to most changes. The prototype was built with the same AI-augmented delivery method proposed for the main programme, so the delivery record was itself part of the evidence for the investment decision.
The outcome
The editorial team tested the prototype hands-on, saw where AI performs and where it needs human judgement, and started shaping the product they will use. The prototype met the client's acceptance criteria, and leadership approved the production build on the strength of it. The approved programme is a full rebuild of the platform, a much larger commitment than the original refactor brief, made on working software and sizing built bottom-up from the organisation's own code.
The production build is now underway: a mobilisation month followed by around 22 weeks of delivery, with go-live targeted for late 2026. The client's engineers remain embedded in the team, which is how the delivery capability stays with the organisation after we leave.
Client identity withheld. The engagement facts above come from the programme's delivery records.