Skip to content
woodfrog

·

Data Foundations
Data integration, quality and governancePlaybook, no client result yet

One number every team agrees on.

Two teams bring two figures for the same measure, and both can be right. Say one counts cancellations and the other does not. Nobody wrote either question down, so the meeting is settled by seniority, not evidence.

One set of records, neatly organised
One set of records, neatly organised
1 of 3Grades of quality check that actually stop anything
6 weeksTypical time to a first production-grade result, firm-wide
50+Projects delivered since 2023, firm-wide
WhoNo client. Written for the finance team and the operations team who both need to accept one set of records.
Read this asA playbook, not a client result.
MethodWeek 1 audit, weeks 2 to 6 build, then operate
Before and after

From two figures to one.

Before
  • Definitions nobody wrote down
  • Checks that warn but never stop anything
  • No contract at the boundary between teams
  • No owner on shared data
After
  • Every definition written down, with an owner
  • Graded checks: log it, alert a person, or stop the run
  • Contracts that break in the upstream team's pipeline
  • One name per dataset, with the right to reject a change
The problem

An argument nobody can win.

It never arrives described as a governance problem. It arrives as an argument, usually in front of somebody senior: two teams quote different figures for the same measure. Neither figure can be traced back to what produced it, so the meeting is settled by seniority rather than evidence.

Two figures, and nobody can trace either
Two figures, and nobody can trace either
What was broken
  • Definitions nobody wrote down: what counts as a customer, when an order becomes revenue.
  • Checks that warn instead of fail, so bad data carries on.
  • No contract at the boundary, so changes are found downstream, by a person, in a meeting.
  • No owner: three teams read a dataset, two write to it, and the person who built it has left.
  • Three records for one customer, and no trace from a figure back to its source.
You might be here if
  • Your organisation holds several answers to what counts as a customer.
  • Your checks warn but never stop anything.
  • A dataset you rely on has no owner.
  • Finding where a number came from means asking a person who is on leave.

Why it mattered. If the definition lives only in a document, the next round of the same argument is already scheduled.

The approach

Stop asking which figure is right.

We ask whoever owns each figure to produce it while we watch. Not describe it, produce it, including the spreadsheet step they were not going to mention. Then we trace each one back to the raw records and write down where the definitions differ. People make the three calls that carry weight. Claude helps read, draft and propose, but deciding the grade of a check is the work.

Produce it, trace it, write it down
Produce it, trace it, write it down
Two teams, two figures
Finance figure
Operations figure
woodfrog method
Produce it, watch itBoth teams, live
Trace to raw recordsWhere definitions differ
Write it downOne definition, one owner
What changes
Contracts at the boundary
Graded checks
Named owners
One agreed number
Deliberately not doneSettling it by seniority, or in a document nobody reads
People decide. Claude reads, drafts and proposes.schematic
The solution

Contracts, graded checks, owners and a trace.

What the method puts in
  • Contracts at the boundary: shape, types and meaning for every field. When an upstream team changes something, the contract breaks in their pipeline.
  • Graded checks: grade 1 writes a line in a log, grade 2 alerts a named person, grade 3 fails the load, so downstream keeps yesterday's data rather than today's wrong data.
  • One name per dataset. “A team name is not an owner. A distribution list is not an owner.”
  • One record per customer, supplier or SKU, with matching rules and a review queue. “Nothing is merged on a hunch.”
  • Record-by-record reconciliation, not netted off into a figure that looks close enough.
  • A trace from the figure on the screen back through every job and definition, plus a change process for when a definition has to move.
Everything is documented and versioned in the client's own repositories. “If your team cannot change it without us in the room, we have failed at the part that matters.”
How it would runDefinitions work starts in Week 1 and runs the full length, because it has the longest lead time. Operate is open-ended and drawn at one week only to fit.
The impact

What we would measure.

How two figures become oneExample: one team counts cancellations, the other does not.
Schematic, no values
Three grades of check. Only one stops anything.The business decides the grade, not woodfrog.
1 of 3stops the run
Grade 1Writes a line in a logNothing stops
Grade 2Alerts a named personSomeone is told
Grade 3Fails the loadDownstream keeps yesterday's data, not today's wrong data
A warning “is a decision to carry on with data you have just been told not to trust.”
What we would measure on a real engagementNo client values yet
Disputed measures with one written definitionHeld where the number is produced
Shared datasets with a named ownerA team name is not an owner
Checks by gradeHow many stop, alert or only warn
Reconciliation differencesListed record by record, each with an owner
Duplicate recordsMerged, held for review, or kept apart on purpose
One agreed number
One agreed number
Settle the number once.A short call, or record a message. A person replies within one business day.
Contact usCertified Services Partner, Claude Partner Network