03 / Data
One number the room trusts
When every tool reports a different figure, the first twenty minutes of the meeting go on working out which file to believe, and whatever was supposed to be decided that morning is not. Merge two sources below into the one number a room stops arguing about and starts deciding from.
Before you merge
Three habits behind the fight
Pretty charts on dirty inputs
If two exports disagree with each other, a dashboard built on top of them is decoration, and a prettier chart only makes the disagreement harder to spot.
Definitions beat decorations
Revenue, churn, and what a customer costs to win each mean two things in most companies I walk into, so the rule gets written down and agreed before anybody paints a tile.
The gap is the story
The distance between two sources for the same number is where the disagreement actually lives, and closing it settles the meeting faster than any chart built on top of it.
Truth Diff
Two numbers. One feed.
Five sample metrics, drawn from the figures two departments most reliably disagree about. Whichever figure your room fights over hardest is usually the first feed I build.
Conflicting sources · Line 01
Revenue
What did we actually book this week?
Leadership arguing over Monday morning exports.
Example figures, not a client’s.
CRM export
€184.2k
Sun 23:00
Includes open quotes
Finance sheet
€161.8k
Mon 08:40
Cash only · manual paste
Two definitions. Two timestamps. The room picks a side — or you merge.
After the merge
Start from the number you fight over.
Every data engagement starts from the number your room already fights over, because that argument is where the cost sits. I wire the feed, we agree the definition and write it somewhere people can find it, and you are left with one daily picture instead of another deck of charts.
Order of work
How one number gets built
5 steps, in this order, and the order is the part people get wrong: if the feed is a lie then the chart is theatre.
Name the fight
Which metric, which two sources are disagreeing, and whose definition is going to win when they do.
Name the one that counts
One system is named as the one that counts, and every other file has to agree with it instead of competing with it.
Build the feed
The numbers come in, get put into one shape using the definition you agreed, and go out as a single daily view.
The odd ones out
Anything that does not fit stays visible on the face of the number, because quietly averaging it away is how a feed starts lying.
Handing it over
The number holds week to week without me in the room, which is the point at which the argument moves back to the work itself.
FAQ
Questions about the data work
What is Truth Diff?
An interactive sample of two sources disagreeing about one metric. You read the gap between them, merge to a single feed, and see how the merged number is built and what it would otherwise hide. The figures in it are illustrative; it is not anybody's live warehouse.
Do you only fix these five metrics?
No. The 5 on the board are the ones that come up most, and any engagement starts from the number your team already argues about, whether that is revenue, pipeline, churn, CAC, NRR, or something specific to how you sell.
Is this a dashboard project?
Not to begin with. The definitions and the plumbing come first, and the tiles come after, because a dashboard built on inputs nobody trusts just moves the argument somewhere prettier. I turn down vanity dashboards on dirty inputs.
Who decides what a metric means?
You do, on the record, and my job is to make you decide it once. Most of these fights are not really about data at all: two teams have been measuring two different things honestly for a year, and nobody has ever written down which one the company is going to use.
How do you handle privacy and GDPR?
The feed can carry rules on how long data is kept and a record of who looked at it, wherever personal data is involved. Anything that turns on interpretation of the law goes to counsel, who owns that opinion: I am not your lawyer, and this practice has no Data Protection Officer.
How long does a data engagement take?
Usually 4 to 6 weeks for one system that counts and the daily picture that comes off it, agreed in writing first. Wiring the feed is rarely the slow part; agreeing the definition is.