Data-driven decision-making

Why it matters

Big decisions made on instinct tend to be made on the most recent or most emotional evidence: a bad month, a customer who left loudly, a competitor's new shopfront. A short evidence step corrects for that at very low cost. Over time, a business that writes down why it decided things and reviews the results learns faster than one that does not, because it finds out which of its beliefs are true. That compounding is the real pay-off, more than any single better decision.

Frame the decision before touching data

Analysis without a decision attached produces interesting charts and no action. Start by writing the decision as a question with options and a deadline: "Should we open a third café in Belconnen by March, or invest the same money in extending hours at the existing two?" Then list what would have to be true for each option to be the right one. Those statements tell you exactly which data to gather and spare you from the rest.

Agree in advance what result would change your mind. If no evidence could move you, you have already decided and should save the analysis effort. If you cannot say what evidence would settle it, the question is not sharp enough yet.

Match the effort to the decision

Not every decision deserves a project. Two questions set the right level of effort: how much is at stake, and how easily can it be reversed? A low-stakes, easily reversed choice, such as trying a new supplier for one order, should simply be made. A high-stakes choice that is easy to reverse, such as a price change you can roll back, is best made as a small test. A high-stakes choice that is hard to undo, such as a lease or a major hire, deserves the full loop and probably outside advice.

Most owners spend too long on the easy decisions and too little on the hard ones, because the easy ones feel comfortable to analyse. The grid below is a quick way to notice which quadrant you are in.

Gather just enough, then test where you can

The goal of the gathering step is to make the choice clear, not to be complete. Start with what the business already records: sales by product, customer location, margin by job type. Add external evidence only where an internal number is missing, such as foot traffic for a site you do not yet have. Stop when more data would not change the decision.

Where a decision can be tried small, a test beats any amount of analysis. A pop-up, a pilot with one team, a price change on one product line, a campaign in one suburb. The test needs a success measure agreed in advance and a fixed end date; otherwise it becomes a permanent half-decision. Record the result against the prediction you made before the test. Being wrong is useful information; being vague is not.

Build the habit in the team

A business becomes data-driven through routine, not announcements. Keep a decision log: one line per significant decision with the date, the options, the evidence used, the choice and the expected result. Review it quarterly. Nothing teaches a team faster than seeing which confident predictions came true and which did not.

Make the numbers available to the people who make decisions, and reward the behaviour you want: asking for the data, running a small test, admitting a decision did not work and changing course. Owners who punish being wrong get teams that stop measuring. For the biggest decisions, an outside view from an accountant or adviser is part of the evidence, not a sign of weakness.

• Keep a one-line decision log and review it every quarter.
• Give managers access to the numbers for their own decisions.
• Agree the success measure and end date for every test before it starts.
• Treat a wrong prediction as learning and a vague one as the real failure.

How to run a data-driven decision

1. Write the question with options and a deadline One sentence. Include the option of doing nothing.
2. List what would have to be true For each option, the two or three facts that would make it the right call.
3. Place it on the grid Stakes and reversibility decide how much effort it deserves.
4. Gather just enough evidence Internal records first, external only for gaps. Stop when the choice is clear or more data would not change it.
5. Test small if you can Agree the success measure and end date first. Predict the result before you run it.
6. Decide and record why Date, options, evidence, choice, expected result. One line in the log.
7. Review on the date you set Compare the result with the prediction. Update what the business believes.

Real-world example

A Canberra café group decides whether to open a third site

Brindabella Coffee Co. runs two cafés. The owners, Leah and Dan, are offered a lease in a suburb they like and feel ready to grow. They frame the question: open a third site by March, or spend the same capital extending trading hours at the existing two? For the new site to be right, three things would have to be true: the suburb has enough morning foot traffic, enough of their existing customers already live there to seed the opening, and the site can reach break-even within six months. Their point-of-sale data shows site one averages about 640 transactions a day at $11.40 and site two about 480 at $12.10. The rent, wages and cost of goods for the new site mean break-even at roughly 410 transactions a day.

They gather two weeks of manual foot-traffic counts at the proposed site in three morning windows, and their loyalty app shows that 14 per cent of site-one members live in the candidate suburb's postcodes. Then they test: a pop-up coffee cart at the suburb's Saturday market for six weekends, predicting at least 250 transactions a Saturday would be a good sign. It averages 290. They decide to proceed, but the evidence also shapes the deal: they negotiate a break clause at 18 months and set the review KPI at 410 transactions a day by month six, recorded in their decision log alongside the forecast so the result can be judged honestly.

Common mistakes

• Analysis without a decision Charts that answer no specific question produce no action. Frame the decision and deadline first.
• Treating every decision the same Agonising over reversible choices and rushing permanent ones. Use stakes and reversibility to set the effort.
• Gathering data to confirm, not to test If no evidence could change your mind, save the effort. Agree in advance what would.
• Tests with no end date or measure A pilot without a success criterion becomes a permanent half-decision. Set both before you start.
• Never reviewing Without a review on a fixed date, the business cannot learn which beliefs were right. Keep the log and read it.

KPIs to track

• Decisions logged Significant decisions recorded with evidence and expected result. Aim for every one that lands in the top two quadrants of the grid.
• Prediction accuracy The share of logged decisions whose expected result was achieved. The trend matters more than the level.
• Time to decide Days from question framed to decision made for the high-stakes quadrant. Falling as the habit forms.
• Tests run per quarter Small pilots with an agreed measure. A business that runs none is deciding everything on instinct or not at all.

Frequently asked questions

Does data-driven mean ignoring experience?

No. Experience frames the question, judges the evidence and makes the final call. Data corrects for the biases in memory and lets you explain and delegate the decision.

What if I do not have the data?

Decide what would have to be true, then look for the cheapest evidence: a two-week manual count, a short survey, a small test. Often the gap itself tells you what to start recording.

How long should a test run?

Long enough to see the measure you agreed on move, and no longer. For most SMB tests that is a few weeks to a couple of months. Set the end date before you start.

How do I get my team to use data?

Give them access to the numbers for their own area, ask "what does the data say" in meetings, keep a decision log and treat wrong predictions as learning. The habit follows the behaviour the owner models.

Key takeaways

• Frame the decision as a question with options and a deadline before gathering anything.
• Use stakes and reversibility to decide how much evidence a decision deserves.
• Gather just enough, then test small with a success measure and end date agreed in advance.
• Write down why you decided and the result you expect; review on the date you set.
• The loop compounds: a business that reviews its decisions learns faster than one that does not.