Sales committed a date engineering had already killed. Two teams spent a quarter building the same thing. A decision sat unowned for three weeks while four people waited. The information was there the whole time — held by people who were never asked, in a company with no system for asking.
The Collision Audit asks. Four-minute interviews across a slice of your org, cross-examined against each other and your plan. You get back the collisions — with the verbatim quotes, the cost arithmetic, and an owner for each. In two weeks.
Founding cohort: three engagements at a reduced fee, in exchange for a named case study.
The audit earns its fee when something just changed — and the org map in your head went stale overnight.
You run a company of 120 to 250 people and one of these is true: you just reorganized, you just raised and are hiring fast, you just acquired or merged a team, you're new to the seat, or you were just blindsided by something a board member or a customer knew before you did.
At that size, after that kind of change, no one person can still hold the whole picture — and the gaps cost real money before anyone names them. That's the window this is built for.
Exhibit A — the finding format. Every finding arrives like this. This one is from our worked example.
Sales committed the customer to a July service-level agreement that Platform has scoped for September. Neither team knew the other's date.
“the July SLA is what got us to proposal stage”
“SLA tooling isn't in scope before the September hardening — decided in May”
Quotes are drawn only from summaries each person approved. Attribution is team-level, never individual.
Three shapes, every time. Watch each one happen.
The unbacked promise — a commitment the delivering team had already descoped. Duplicate work — two teams building the same thing without knowing. The silent stall — a decision with no owner, aging while people quietly wait on it.
Two people, typing at the same time, saying things that can't both be true. It surfaces at the customer kickoff — unless someone asks first.
None of these are knowledge problems — the people involved knew. They are coordinationproblems: the information sat one conversation apart, and no one's job was to have the conversation. And they get worse with scale — a company that produces at AI speed collides at AI speed, while its coordination still runs at meeting speed.
No software is installed. No meetings are added — one is retired.
We draft it; they send it. It cancels the status meeting, grants amnesty to unregistered automations, and commits — contractually — to zero individual reporting.
A link, no login. A conversation, not a form. Each person edits and approves their own summary; nothing else is used, and the raw conversation is never stored.
Every summary against every other, and against your stated plan. Six collision types. One rule keeps it honest: no quote, no finding.
Findings reach the responsible executive 48 hours before the readout, with response rights — not a veto. Findings arrive as 'flagged, already being resolved,' never as an ambush.
Three to five findings: the evidence, a capacity range whose arithmetic your CFO can re-derive, an owner, a next step — plus the AI Footprint.
The constraints are the product. Each is contractual — and several are structural: there is no table in our system for an individual score.
Every claim cites verbatim, person-approved text. Findings that fail verification are killed — we publish how many.
Nothing a person didn't approve is ever kept. Raw conversations are never stored — by architecture, not policy.
Quotes attach to teams of a set minimum size; below that, corroborated paraphrase. Individual metrics do not exist, and it can never be used as a performance review.
Cost figures are ranges with their inputs exposed — headcount and calendar, not value-at-stake hand-waving. Your CFO can move the assumptions and watch the range move.
Whether a finding is worth acting on is your call, in writing. If none are, the fee is waived.
Every audit includes one thing your board is starting to ask about and you probably can't answer today: every AI tool and agent with a token to your company's data — what it touches, who owns it, and which ones are running with no owner at all.
From one read-only consent. It's a security and exit-diligence answer most companies your size are missing — and on its own, reason enough for the two weeks.
The only qualifying question
When were you last blindsided by something your own company already knew?
If the answer is a story from this year, the audit will pay for itself on the first finding. If you can't think of one, you don't need us yet — and we'll tell you so.
The audit is the first step. After it, the same engine gives your people something they actually want to use.
The interview asks each person what they're building, what's stuck, and what they need. That becomes a private cockpit for them: their blockers get a named owner and a status they can watch clear, an auto-built board of their work laddered to your goals, and an agent that drafts the busyworkthey hand off. Leadership gets the fleet view — what's automated, what's stuck, how goals are tracking — as the byproduct.
That's how a one-time diagnostic becomes a quarterly pulse no one resents filling in. But we start with the two weeks.
See the employee cockpit →The film — 60 seconds
Sound on
Alan Yang — fifteen years building executive reporting and operating-cadence systems for financial-services leadership at PwC and Westpac; Berkeley MBA. The audits are founder-delivered: the person who designed the method runs your engagement and stands behind the readout.
A. Yang
Founder, Mythral