Investigations, not testimonials.

Engagements written up the way they ran: what the client thought, what the numbers said, what moved. Every figure reconciles to the client’s own reporting. Named where the client agreed, anonymised where they asked, and some of this work sits under NDA.

  1. File 01
    A $100M, 60-year engineering consultancyAnonymised at the client’s request

    Seven cascading problems. One question nobody had asked.

    A $3–5M software business had grown almost by accident inside a 60-year consultancy and was straining under its own weight. The code was fine. The decision underneath it had never been made.

    15+Diagnostic interviews5 → 1Competing visions → one owner
  2. File 02
    PerformLine, a marketing-compliance platformNamed with permission · the launch is public record

    Four people. Under 70 days. A new product category.

    A marketing-compliance platform saw a category no one had touched: what AI assistants tell consumers about its clients' financial products. Four people took it from concept to public launch in under 70 days.

    <70 daysConcept to public launch4People on the build, two part-time
  3. File 03
    A 77-year, $20M skincare brandAnonymised at my request

    The agency reported magical gains. The bank statement told a different story.

    Paid traffic converting at a third of the industry floor, and an agency whose numbers didn’t reconcile. We fixed what was leaking before adding a dollar more.

    0.97% → 2.73%Site conversion, same traffic3× → 9–14×Return on ad spend, same budget
  4. File 04
    A $30M creative agencyAnonymised at the client’s request

    From five days to three hours, without new headcount.

    A full-service agency was spending five days turning one design into fifty placements, by hand. The leverage turned out to be sitting in tools they already owned.

    5 days → 3 hrsTime to produce 50 placements−97%Production time cut
  5. File 05
    GrowthLoop, a marketing-automation platformMatt’s own history · an inside role, not a client engagement

    The team tripled. The complexity didn’t.

    When Matt arrived, the core application was a two-person team and every integration was a bespoke build. Three years later the team was 3× the size, the platform processed billions of transactions a day, and infrastructure costs had stayed low.

    3×Engineering team growth across the tenureBillionsTransactions processed per day
  6. File 06
    CBC Digital, Canada’s public broadcasterMatt’s own history · role and organization are public record

    Busy isn’t an outcome. Neither is shipping.

    Large organizations rarely lack things to do. The harder problem is knowing whether the activity produces outcomes that matter. At CBC Digital, every goal had to name its baseline, its target, and its date. That changed everything it touched.

    8Rewrites of one quarter’s goals in two weeks2×Where the final targets landed, vs. the comfortable ones
  7. File 07
    Cloud DX · Qualcomm Tricorder XPRIZEMatt’s own history · team and results are public record

    Building a medical device when the deadline doesn’t move.

    Build a device that diagnoses medical conditions without a physician, in six months, with working hardware arriving four weeks before the deadline. Waiting wasn’t an option, so waiting was designed out.

    Top 3Finish, out of ~330 teams worldwide~6 monthsTo build what sounded like science fiction
On the record

Named where the client agreed, anonymised where they asked. The advisory engagements here run without names, one of them at my own insistence. The files drawn from roles I held name the organisation, because that work is already public record. The numbers are real either way, and a client’s name is theirs to give, not mine.

The first deliverable in every engagement is the truth about the numbers. If a reported figure doesn’t reconcile to the bank, that’s the first finding, before anything gets optimised.

The control case

Files 01–03 are diagnoses. File 04 is what happens when the levers they name are actually pulled.

Four people (two developers, one part-time, a part-time designer, and one product lead) took PerformLine’s AI Response Monitor from concept to public launch in under 70 days. Not with headcount: with process. Research first, rapid iteration, clients reacting to working software while it was built: a discovery loop that’s nearly unheard of inside an enterprise. AI held the guardrails; the process did the work.

It’s the same machinery as File 03: eight products, five visions, no one who could greenlight, run in the right direction. The capacity was already in the building. What it moved fast on was process and ownership, and those are the cheapest things to add.

Know where this has to go, and not how?

Twenty minutes. Tell me where you want this company to be in two years and I’ll tell you what I’d go look at first.

Start a conversation