A three-senior boutique for consequential product and data problems: diagnosed, explained and fixed in one to four weeks. You get working systems and a clear decision, not a deck.
The problem
When a metric that matters stalls, the same scene plays out everywhere: product blames the funnel, growth blames the tracking, data blames the events, and next quarter gets bet on whoever argued best in the meeting.
The market answers this badly. Big consultancies sell a strategy deck that needs another team to implement. Agencies rent you a bench by the day. The people who can read the data usually cannot ship the fix, and the people who can ship rarely sit in the meeting where the number gets used.
We sit exactly in that gap: find out what is actually happening, say why, and build the fix before leaving. Diagnosis and implementation in the same heads.
The wedge
Funnels, cohorts, experiments, and the operational analytics that connect what users do to what it costs and earns. If the metric is lying, we find where and prove it.
dbt, Snowflake, Python, AWS. The tracking plan gets implemented, the tests get written, the pipeline still holds long after the engagement ends.
Head-of-data experience on the team: decisions defended at exec level, and AI adoption grounded in semantic layers instead of demos.
Most shops rent you one of these. Some manage two. The point of this one is all three, at the same table, on the same problem. Hence the name.
The menu
Every offer runs the same way: a fixed fee, a fixed number of weeks, explicit exclusions, and a result that stands alone. The first engagement is never a teaser for a second one. Most problems walk in through one of the three marked start here.
Your north-star metric stalled and every team tells a different story.
We rebuild the funnel from raw events, cut your users by what they do rather than who they are, and stress-test every theory against the counterevidence. You get three to five hypotheses that survive scrutiny, the experiment backlog to test them, and a decision made on evidence instead of volume.
Users swear they trust the product. Then they double-check everything it does.
We map where people verify, override, retry and abandon, against what they tell your surveys, and hand back the product changes that close the gap. Built for the AI-feature adoption problem every roadmap now has.
Every metrics meeting starts with an argument about definitions.
We kill the vanity events, wire the metrics to the decisions they exist to support, build the models that matter, and leave a taxonomy that stays clean after we're gone.
Every channel claims the same conversion and your CPA is fiction.
We audit the UTMs and identity joins, write attribution rules with their limits stated out loud, and deliver channel cost you can budget against. This team has built it before, end to end, at Deel.
Your event data and your user research have never been in the same room.
One analysis that holds both: segments by behaviour, stated versus revealed preference, and the hypotheses your roadmap has been missing because each half only had half the picture.
Reports fail silently, engineers firefight, and the roadmap lost the room.
What is broken, what is merely complicated, and what to fix first. Ranked failure modes, a 30/60/90 plan, quick fixes shipped on the way through, and what not to rebuild, in writing.
Two dashboards, one KPI, different numbers. The board noticed.
We trace the numbers that leave the building, source to report, reconcile them against independent systems, and leave automated tests standing guard on the failure modes we found.
The warehouse bill doubled. The value didn't.
We tear through the workloads, kill the zombie jobs, right-size the compute, and leave guardrails so the bill stays down.
You're running big-data infrastructure on medium-data problems.
A target architecture sized to the problem you actually have, a migration order that doesn't stop the business, and the cost and delivery-speed maths your CFO can read.
A dashboard change takes three weeks and nobody can explain why.
We trace the path from request to production, break the queues, set standard patterns and CI/CD, then prove it by shipping a real change through the new path before we leave.
Everyone is pasting into chatbots. Nothing is governed.
One bounded workflow: the model interprets, classifies and drafts, deterministic code owns the numbers and the writes, a human approves what matters. Evaluation set and audit log included, so it survives contact with your security team.
Leadership wants agents. Your data can't hold them yet.
Proceed, narrow, postpone or reject, decided inside two weeks: data quality, permissions, evaluation data, and the semantic layer your agents will actually read.
You're about to buy a company on the strength of its metrics.
Architecture, data quality, metric credibility, key-person risk, and a decision memo, before the wire goes out.
Method
The team
The whole path, one pair of hands. From the event landing in the cloud to the number a board acts on: dbt, Snowflake, Python, AWS. Rebuilt a hypergrowth fintech's data function as its second data hire, built its marketing attribution, and migrated a UK specialty insurer's platform off Scala. Ships the fix personally.
Reads a product the way an operator reads a P&L. Funnels, cohorts and experiments on one side; the operational analytics that tie user behaviour to cost and revenue on the other. Owns the investigations, and the moment a plausible story meets an inconvenient segment.
Has run the data function, not just advised it. Head-of-data experience: teams built, hires made, roadmaps defended at exec level. Today deep in AI adoption and semantic layers, the groundwork agents need. Owns the executive conversation and the readiness calls.
Track record
DeelDeliverooAirHelpa UK specialty insurer
Bring us the number nobody trusts.
Send the problem, the stack and the outcome you need. We'll tell you straight whether it fits a one-to-four-week engagement, what it would cost, and what we'd explicitly leave out.
Tell us the problem →Not wired yet · prototype