Preparing data
From scattered files to one honest table.
Data processing, for us, means getting information into a state where it can support a decision, and writing down how it got there.
Gathering
We ask for extracts of the data the question needs, not everything you have. A typical list might be a year of sales by order, a product list with costs, and a staff hours export. For each source we note who owns it, how it was produced and how far it can be trusted. Data is transferred by a route we agree with you in writing, and only once terms covering our processing are in place.
Cleaning
Real business data is untidy in ordinary ways: the same supplier spelled three ways, dates typed as text, test entries left in, refunds recorded as negative sales in one place and separate lines in another. We fix what can be fixed by clear rules and flag what needs your judgement. Nothing is silently dropped.
Joining
Most useful answers need two or more sources side by side. Joining is where definitions matter: is a customer an account, a delivery address or a person? Is a week Monday to Sunday or your trading week? We agree these choices with you before relying on them.
Shaping
The prepared dataset is arranged for the question: one row per order, per customer, per week, whatever the decision needs. It is delivered in an ordinary format your team can open, with a data dictionary that explains every column.
Reading the result
Finally, a short written reading: what the prepared data shows, how confident we are, and what it cannot tell you. This is where the value appears, and we write it for the person who decides rather than for an analyst.
What you receive
- Prepared dataset
- Clean, joined and shaped for the question, in a common file format.
- Data dictionary
- Every column explained, including where it came from.
- Assumptions list
- Every rule, exclusion and judgement call, so nothing is hidden.
- Written reading
- What it shows and what it does not, in plain English.