Four pieces of work I keep coming back to - not because they shipped smoothly (they didn't), but because each one turned on a decision that could have gone the other way. That's the part worth showing, so that's what each case study walks through. One note: the visuals throughout are redrawn concept illustrations - the real client work stays confidential, as it should.
From a single-client service to a product four GPs are trialling on live deals.
Holding the tension between serving a design partner and building something the market will actually buy.
The platform I inherited had been built by transcription - one flagship client's process, digitised screen by screen, with no product thinking underneath. We rebuilt from the real co-invest raising process, then used the prototype to widen the room: GPs across the industry, their reactions sharpening what we built. Four GPs have now committed to trials on live deals - one live today, one in implementation, two queued - from a starting position of exactly zero.
The brief asked for a data room. The answer turned out to be a product line.
Reframing the opportunity, not just the solution.
We entered an RFP for a virtual data room - without one to show. Rather than compete on parity with providers who'd had a decade's head start, we pivoted the prototype into a complete pre-trade experience - fund, marketing, data room, access, analytics - modular enough to sell standalone. The prototype became the pitch: demoed to a major private markets investor, warmly received, now integrated into the product suite.
The operational bottleneck that became a revenue line.
Seeing the product hiding inside an operational problem.
ClearGlass brings cost transparency to the pensions industry. The LGPS work arrived as a manual bottleneck - hierarchical fund data, processed by hand, cycle after cycle. I led the build of a multi-scheme platform that automated that processing end to end - mapping structures, visualising data flows, supporting transparent verification - and what had been an internal cost became a new revenue stream, now scaling on the learnings of its first collection cycle.
The integration that made employment data live.
Working at engineering depth - designing the flow, not just writing the spec.
At Salary Finance - a B2B2C fintech serving over 500 UK employers - everything rests on knowing who's employed, at what salary, right now. I led the company's first direct-to-employer API integration, with Sainsburys: real-time employment validation for over 200,000 employees, replacing the lag of periodic files with live salary and leaver data. I wrote the technical and functional specs and designed the backend flow myself - then delivered it on time.
Opinions I hold about the work - formed by doing it, and still being revised by it.
Product management reinvents itself every few years.
You can read the reinventions in the tooling. In 2018 the job lived in Jira - requirements, tickets, translation between business and build. By 2021 it lived in analytics dashboards and user interviews - evidence, discovery, and the authority to say no. Now it lives in AI tooling - and a skilled PM can define, research, build and iterate in weeks what once took a team and months. Each wave stacked on the last rather than replacing it, which is why the job keeps getting bigger. The trick is noticing which version you're in.
The cost of being wrong has collapsed.
It isn't that prototyping got faster, though it did. It's that finding out you were wrong got cheap - an afternoon's work, not a quarter's. That quietly rewrites the economics of every product decision: ideas we'd once have killed on paper are now worth exploring, and answers can change mid-flight, because being wrong on Tuesday means being right by Wednesday. Most teams have updated their tools. Fewer have updated their appetite.
A persuasive prototype is a double-edged thing.
The same concreteness that gets everyone aligned also ends the conversation early - a good prototype quietly forecloses the five solutions nobody got the chance to propose. Alignment and premature lock-in are one mechanism seen from two sides, and knowing which one you're getting in the room is the actual job. The counterweight is breadth: a prototype in one room aligns a client; the same prototype in ten rooms reveals a market. And however good the artefact gets, it only ever shows what could be built - never what should. Prioritisation stays a human act; the most dangerous moment in prototype-driven work is mistaking momentum for a decision.
Trust is architecture.
In markets where the users manage other people's billions, safety isn't a compliance afterthought - it's a first-class product decision, made at design time. Authentication before access, entitlements before answers, audit by construction. The products I'm proudest of led their pitch with the guardrails, and won because of it, not despite it.
Roles don't disappear. They relocate.
When product produces the first concrete artefact, design's job isn't to rubber-stamp it - it's to raise the ceiling. Engineering's job isn't to recreate it - it's to test it against the constraints the prototype was free to ignore. The half-life of any way of working is short, and getting comfortable re-learning the process is the durable skill. What's permanent was never the workflow. It's judgment.