Overview / Method
The method, written down
Five steps that make the work traceable. At each step you know what we need, what we produce and what you are asked to confirm.
| Step | What we need from you | What we do | What you see |
|---|---|---|---|
| 1. Scope | The question, the decision it supports, the files involved and a named contact | Write a one-page scope with what is in and out | Scope note to approve |
| 2. Profile | A minimised extract with unneeded columns removed | Measure types, gaps, formats and duplicates | Profile summary and first questions |
| 3. Agree rules | Answers about meaning: what a blank is, which name is canonical | Draft rules with worked examples | Rule list to approve line by line |
| 4. Process | Nothing, unless an exception needs your judgement | Apply rules, record counts, set aside failures | Prepared data, change log, exceptions |
| 5. Read together | Thirty to sixty minutes of attention | Present outputs and the interpretation note | Agreed reading and next actions |
Principles that do not move
Minimise first
We ask for the smallest extract that answers the question. If a name or address is not needed to match records, it should not be in the file.
Never change silently
Every rule is agreed and logged. The original file is kept unchanged so any result can be reproduced from it.
Report counts, always
Rows in, rows out, rows set aside. If the numbers do not reconcile, the run is not finished.
Say what the data cannot do
A note that lists its own limits is more useful than one that sounds confident.
Hand over properly
Run notes are written for the next person, so the process does not depend on any one individual, including us.
Typical timescales
A single preparation exercise on a modest export commonly takes from a few days to a couple of weeks, mostly depending on how quickly business rules can be confirmed. Recurring runs are set up once and then follow your reporting calendar. We give a specific estimate after profiling, not before, because an estimate made without seeing the data is a guess.