How it works
Data your teams can rely on.
When sales and finance use different customer records, even a simple question takes time. I help your teams connect their systems, fix gaps and duplicates, and reuse the same checked data in daily operations, reporting and AI.
Where we start
You have a reporting system
Show me the last signed-off report and one number people challenge. I trace it through workbook formulas, joins and source transactions. We keep the reliable pieces, repair the break and compare old and new outputs through a reporting cycle before switching over.
You’re starting from scratch
We choose the first report and its owner, identify the system that records each event and manually check a small sample. That gives us a reference total before we pick a platform or automate a feed.
Worked example
Why does £445 become £275?
In this illustrative September sales report, an £80 invoice arrived twice and a £90 invoice was corrected to £95. Select a layer to follow those transactions from the source feed to the regional totals.
Silver · Cleaned
Duplicates removed, formats fixed, records matched across systems. This is where the numbers start to agree.
- Remove duplicates
- Fix dates, currencies and names
- Match records across systems
Checks at this stage
- No duplicate keys
- Dates and amounts parse
- Every row traces to a source row
What the data looks like
Four current transaction lines. The £95 revision replaces £90.
I checked these figures with a synthetic dataset in SQLite. The £275 total includes the latest correction to September sales. If finance needs the figure it signed off at month-end, we also need the cutoff date and a record of which corrections had arrived by then.
What has to pass before go-live?
Agree the figure
Finance signs off the period, currency, treatment of credits and reporting deadline. I record who can approve a correction after month-end.
Reconcile the report
I compare both the total and each regional subtotal with the finance control. Two errors can cancel each other out in a grand total. Every material difference needs an explanation.
Prove it can run again
We replay the £80 invoice. Bronze gains an arrival, while September sales stay at £275. We also test a failed refresh so the report owner can recognise stale figures and recover the load.
Hand it over
Your team gets the definitions, source-to-report map, alerts and recovery instructions. A named operator runs the next refresh. We measure preparation time against the original reporting cycle.
One definition for dashboards, spreadsheets and AI
Power BI can read the approved regional sales measure directly. For teams using Claude or ChatGPT, I can expose the same measure through a read-only Model Context Protocol (MCP) connector that checks the user’s permissions.
Ask for a September workbook and a short sales briefing. The workbook should show the same £195 for North and £80 for South, along with its filters and refresh time. To explain the £170 adjustment, the assistant needs the permitted source lines as evidence. We check the exported file against the dashboard and repeat the test as a user who can see only one region.
Discuss your first report