Data-driven decisions:
a guide for managers

Data-driven decision-making (DDDM) is using evidence systematically alongside judgement. It covers framing the question, choosing metrics, gathering and weighing evidence, acknowledging uncertainty, deciding and learning from the result.

It matters because it works. In a study of 179 large firms, Brynjolfsson, Hitt and Kim found that companies adopting data-driven decision-making had output and productivity about 5–6% higher than their other investments would predict.¹ And it is increasingly expected of managers: the World Economic Forum's 2025 report lists analytical thinking as the most-wanted core skill and leadership and social influence among the fastest rising.⁴

Productivity edge of data-driven firms (Brynjolfsson, Hitt and Kim, 2011)

5–6%

Large firms in that study

179

Information Jeff Bezos suggests is often enough to decide (Amazon, 2016)

~70%

Analytical thinking among core skills employers want (WEF, 2025)

#1

Table: Data-driven decision-making by role (editorial summary)

1. Frame the decision — What are we deciding, by when, and what are the realistic options?
2. Define success — Which metric reflects the goal? Which guardrails must not get worse?
3. Gather the most useful evidence — Existing data, quick analysis, customer conversations, a small test.
4. Weigh it honestly — How strong is it? What's biased or missing? What would change our mind?
5. Decide at the right speed — Fast for reversible "two-way door" decisions, careful for one-way doors.²
6. Record and review — Write down the decision, evidence and expected result; check back on a set date.

Where a reversible change can be tested, a controlled experiment beats debate: Kohavi and Thomke describe how a small, tested change at Bing produced a 12% revenue lift that intuition had overlooked.³

Many decision errors are behavioural; our Laws of UX pages on cognitive bias and anchoring explain how they work.

"Tell me about a decision you made with incomplete data."

Use a real example: the decision, what evidence you had, why you couldn't wait, how you limited the risk, what happened and what you learned.

"Your team's main metric dropped 10%. What do you do?"

Check data quality, scope the change, segment it, look for internal and external causes, decide whether action is urgent, and communicate what you know and don't.

"Two senior colleagues disagree, each citing data. How do you decide?"

Get both analyses on agreed definitions, check the evidence quality, identify the real disagreement (data or values), and, if the decision is reversible, test it.

"How do you choose metrics for your team?"

Start from goals, choose outcome and driver metrics, add guardrails, keep it to a handful, and make sure the team can influence them.

"When would you not use data to decide?"

When the decision is quick and reversible and data would cost more than a mistake, or when it's about values and ethics that data can't settle — though evidence can still inform it.

"How do you avoid confirmation bias?"

Pre-commit to criteria, ask someone to argue the other side, look for disconfirming evidence and run tests where possible.

"How do you know a decision was good?"

Separate decision quality from outcome luck: was the process sound given what was known? Review both.

"A dashboard shows a great result that seems too good. What do you do?"

Assume something's wrong until checked: definitions, data freshness, duplicates, a tracking change.

Table: How data-driven management has changed (editorial summary)

• Three decision memos — real or realistic decisions with framing, evidence, choice and review date.
• A team metrics set — outcomes, drivers and guardrails for a team you know.
• One test you designed — even a small pilot with a comparison group.
• A verified decision-making score to link.

1. Week 1: Framing — Write decision questions and options for five real choices at work.
2. Week 2: Metrics — Build outcome, driver and guardrail metrics for your team.
3. Week 3: Evidence and bias — Review three past decisions for bias and evidence quality.
4. Week 4: Prove it — Start a decision log and take the Data-Driven Decision-Making Test.

Related tests: Statistical Reasoning, Decision-Making and Judgment and OKRs and Performance Management.

Decision: Should support move to a 24×7 rota for overseas customers?

Options: (A) keep current hours; (B) 24×7 for all; (C) 24×7 pilot for the Middle East region only.

Success metric: churn among overseas customers over 90 days. Guardrails: support cost per ticket, agent attrition.

Evidence so far: 18% of overseas tickets arrive out of hours; overseas customers who waited more than 8 hours for a reply churned at twice the rate of others (correlation, not proof); two main competitors offer 24×7.

What would change our mind: if the pilot shows no difference in churn after 90 days, or cost per ticket rises more than 25%.

Decision: Option C — a reversible, low-cost test. Review date: 15 February 2027.

One page, written before the data comes in, with the criteria fixed in advance — that is what data-driven decision-making looks like in practice.

1. Decision log — For four weeks, record each significant decision, the evidence, your confidence and a review date.
2. Pre-commit — Before your next data review, write what result would lead to each option.
3. Reversibility sort — List ten pending decisions and mark each as one-way or two-way; speed up the two-way ones.
4. Metric audit — For your team's metrics, mark each as outcome, driver or guardrail, and remove one that doesn't inform decisions.
5. Small test — Turn one debated change into a two-week pilot with a comparison group.

Data-driven judgement is what managers are increasingly hired and promoted on. These analyst-side roles show how the same skill is paid when it's the core of the job (AmbitionBox, October 2026):

Table: Decision-support roles and their typical pay in India (AmbitionBox, October 2026; indicative)

• Waiting for perfect data on reversible decisions.
• Choosing metrics that are easy to count rather than meaningful.
• Cherry-picking supporting numbers.
• Confusing outcomes with decision quality.
• Never reviewing whether a decision worked.

Research references are Brynjolfsson, Hitt and Kim (2011), Kohavi and Thomke (HBR, 2017) and Kaplan and Norton (HBR, 1992); decision-speed guidance comes from Amazon's 2016 letter to shareholders; skills context from the World Economic Forum. Interview content and the history table are editorial summaries. Science of Apps publishes the free test linked here. Last reviewed October 2026.

• Data-driven decision-making — using evidence systematically alongside judgement to decide.
• Two-way door decision — a reversible decision that can be made quickly and changed later.
• Guardrail metric — a measure that must not worsen while you pursue a goal.
• Confirmation bias — favouring information that supports what you already believe.
• Decision log — a record of decisions, the evidence behind them and expected outcomes.
• HiPPO — the "highest-paid person's opinion", which can override evidence.