Framing decision questions: turning vague business goals into answerable analytics questions

1. What you will learn

• Why analytics should start from a decision, not from data or dashboards.
• The difference between descriptive, diagnostic, predictive and prescriptive questions.
• A five-part template for a well-framed decision question: decision, owner, options, criteria and deadline.
• How to break a broad goal into a question tree of smaller, answerable questions.
• How to run a quick diagnostic of your business's current decision-making and find the gap that matters most.

2. The idea explained

Decision intelligence is the practice of designing how decisions are made — what information is used, how options are compared, who decides and how outcomes are reviewed — rather than simply producing reports. Many small and growing businesses have plenty of data (billing software, e-commerce dashboards, ad accounts, spreadsheets) but still make key decisions on instinct, because nobody has turned the data into answers to the questions that matter.

The starting point is always the decision. A report without a decision behind it is a vanity report: it may be interesting, but nobody acts differently because of it.

Four types of analytics question. A widely used classification helps you see what kind of answer you need:

1. Descriptive — What happened? Sales by month, customers by city, returns by product.
2. Diagnostic — Why did it happen? Why did repeat orders fall in March? Which factor changed?
3. Predictive — What is likely to happen? What will demand be next quarter? Which customers are likely to lapse?
4. Prescriptive — What should we do? Which price, which stock level, which channel budget gives the best expected result within our constraints?

Each step needs more data discipline than the one before. Most small businesses should first make their descriptive numbers reliable; prediction built on unreliable history simply produces confident errors.

The five-part decision question. A well-framed question names:

• Decision — the choice to be made, in one sentence. "Should we open a second outlet in Whitefield this financial year?"
• Owner — the one person accountable for deciding (others may advise).
• Options — at least two real alternatives, including the status quo.
• Criteria — how options will be judged: payback period, margin, risk, strategic fit, with any thresholds.
• Deadline — when the decision must be made, and why then (lease offer expiry, season start, budget cycle).

When these five are clear, the analysis needed usually becomes obvious — and so does the analysis that is not needed.

The question tree. A broad goal ("grow profit by 20%") cannot be answered directly. Break it into branches that are mutually exclusive and collectively exhaustive — often abbreviated MECE — meaning the branches do not overlap and together cover the whole goal. Profit, for example, splits into revenue and costs; revenue splits into number of customers × orders per customer × average order value; costs split into variable and fixed costs. Each leaf of the tree becomes a smaller question you can answer with data and act on.

Diagnostic of current decision-making. Before designing new analytics, assess the present state. For the five to ten most important recurring decisions in the business (pricing, purchasing, hiring, marketing budget, credit to customers, new products), ask: How often is it made? Who makes it? What information is used? How long does the information take to prepare? How often, looking back, was the decision wrong or late? The decision that is frequent, high-value and poorly informed is your biggest capability gap and the right place to start.

3. Let us work through it

Step 1 — List recurring decisions. Write down the ten most important decisions made in the business each month, quarter or year.

Step 2 — Score each decision from 1 to 5 on value (money at stake), frequency, and current information quality (5 = excellent, 1 = guesswork).

Step 3 — Calculate a priority score = value × frequency × (6 − information quality). High scores show important, frequent decisions made with poor information.

Step 4 — Frame the top decision using the five-part template.

Step 5 — Build a question tree under it, splitting the decision into MECE branches down to questions you can answer with available or collectable data.

Step 6 — Classify each leaf question as descriptive, diagnostic, predictive or prescriptive, and note the data it needs.

Step 7 — Agree the frame with the decision owner before any analysis starts.

5. Common mistakes and how to fix them

• Starting with a dashboard instead of a decision. Fix: write the decision and owner first; build only what that decision needs.
• Questions with no real options. Fix: list at least two alternatives, including the status quo.
• Jumping to prediction before descriptive numbers are reliable. Fix: make basic historical data accurate first.
• Overlapping or incomplete question trees. Fix: check that branches are MECE — no overlap, nothing missing.
• No deadline, so analysis never ends. Fix: set a decision date tied to a real business event.
• Analysts and decision-makers not aligned. Fix: agree the five-part frame with the owner before analysis starts.