From data to decision.
This is how every Data & AI Playground project works. Six steps, the same each time, so you can see what we do, what we check, and what you get at the end of each one.
Start with the decision
Before we touch any data, we agree the decision it should inform, who owns it, when it has to be made, and what would change their mind. Then we turn that into a question the data can answer. A vague question gets a vague chart.
- Which decision changes if the answer is yes, and which if it is no?
- Can public data actually answer it, or do we need yours?
- Who signs off, and by when?
Find the primary source
We go to the body that publishes the number first: the statistical office, the regulator, the register, the filing or the official release. We record the link, the publication date, what is counted and what is left out. We keep a dated copy.
- We cite the publisher of the number, never a summary of it.
- We note the definition and coverage, so a “rate” is never confused with a “count”.
- We save a dated copy in case the source changes.
Explore with AI
AI reads the tables we have pulled, compares years, flags outliers and proposes patterns far faster than a person can. It surfaces questions we would not have thought to ask. It never supplies a number of its own.
- AI works only on data we supply from primary sources.
- Every pattern it finds is a hypothesis until step 4 confirms it.
- We keep a record of what we asked it, so the work can be repeated.
Verify every number
This is the step that separates a story you can act on from one you can only share. A person recalculates each figure by hand from the source, reconciles totals, and checks that definitions match across years. Then we write down what the data cannot tell you.
- Each figure recalculated by hand from the source.
- Units, denominators and definitions checked across years.
- Outliers explained or flagged, never quietly dropped.
- A written note on what the data does not show.
Turn it into a visual story
We choose the form that fits the question: a chart, a map, a table, or an explorer you can click through. One idea per chart. The headline states the finding. The source is printed underneath. Colour guides the eye and never decorates.
- A table view next to every chart, so nothing is hidden.
- Meaning never carried by colour alone.
- A “what this does not show” line under every chart.
Decide and act
We walk through what the evidence supports and what it does not, set out the options with their trade-offs, and agree the next move with you. If it helps, we stay on to track whether the numbers moved the way you expected.
- Two or three options, each with its trade-offs.
- What the evidence supports, and what it does not.
- One recommended next step, with the number to watch.
Who does what.
AI is fast and tireless. People are accountable. We use each for what it is good at, and the hand-off between them is where the checking happens.
- Reads the tables we give it
- Compares years and countries
- Flags outliers and unusual changes
- Suggests patterns and questions
- Drafts a first summary for us to edit
- Choose the decision and the question
- Find and cite the primary sources
- Verify every number by hand
- Decide what the finding is, and write it
- Take responsibility for what is published
AI never supplies a number. It only works on numbers we have pulled from a primary source and checked.
Three tiers, and we use only one.
The body that publishes the number first
Statistical offices, regulators, registers, official releases and company filings.
Reports that cite a source
We use them to find the original, then go to the original. We never cite them as the source.
Numbers with no traceable origin
Unsourced statistics, screenshots of charts, and recycled headlines.
Where the work overlaps.
Bring us a decision.
Tell us the decision you face and the data you hold. We will tell you which of these steps you need.