01 / PRACTICE MANUAL
Start with a decision
The task
Choose one recurring business decision. A dashboard is useful when someone can name the decision it changes. Start with a sentence: “Every Monday, the store manager decides which products to replenish for the next seven days.” Name the decision-maker, timing, available actions, and constraints before choosing a chart.
Worked example
A fictional retailer sees weekly sales fall from 1,000 units to 900. “Increase advertising” is one possible response, but sales alone cannot establish the cause. Stock shortages, fewer visitors, a changed product mix, and a shorter trading week could produce different actions. Treat these as competing explanations to investigate.
Decision brief
Write: Decision owner; decision deadline; available actions; objective; guardrail; information missing. An objective could be fewer stockouts. A guardrail could be a fixed inventory budget. This keeps the analysis tied to an action instead of an open-ended search for interesting numbers.
Exercise
Select a decision you can revisit within two weeks. List three actions, including “make no change.” For each action, write what evidence would make it reasonable. If every answer is “more data,” narrow the decision until a specific observation could change your choice.
02 / PRACTICE MANUAL
Define the measures
Make a metric contract
For each measure, record its name, purpose, formula, unit, time window, exclusions, source, and owner. Define whether revenue includes returns, taxes, and delivery fees. Define whether a customer means an account, a person, or a household. Two reports can disagree without either calculation being wrong if their definitions differ.
Worked example
Using fictional figures, conversion is 180 purchases divided by 3,000 eligible visits: 6%. If another report divides by 2,400 visits after excluding staff and bots, it shows 7.5%. The difference is the denominator. The reports should not be compared until the population and exclusions match.
Useful companions
Pair an outcome with a diagnostic measure and a guardrail. For replenishment, consider stockout rate, supplier lead time, and inventory value. A diagnostic helps investigate an outcome; it does not prove a cause. Record missing data rather than silently treating missing values as zero.
Exercise
Create a metric contract for one measure. Ask a colleague to compute it from the definition without your help. Record each question they ask. Revise the definition until the calculation can be repeated consistently.
03 / PRACTICE MANUAL
Check before interpreting
Five checks
Check freshness, completeness, uniqueness, validity, and consistency. Freshness asks when the data last arrived. Completeness asks what is missing. Uniqueness checks duplicate business events. Validity checks allowed ranges and types. Consistency compares definitions and totals across sources.
Worked example
Suppose 50 stores normally submit daily sales, but only 45 submitted yesterday. A lower total may reflect incomplete reporting. Show “45 of 50 stores received” beside the total and compare the same reporting stores where appropriate. A like-for-like comparison answers a narrower question; it does not replace the missing reports.
Comparisons that help
Compare equivalent periods and segments. Check trading days, holidays, promotions, and store openings. Report absolute change and percentage change together. A rise from 2 to 4 is 100%, but only two additional events. Avoid implying that a large percentage necessarily means a large business effect.
Exercise
Build a five-row quality log: check, observed result, business consequence, owner, resolution. Mark each issue as blocking or non-blocking for the specific decision. State the reason; do not use a blanket “data is good” label.
04 / PRACTICE MANUAL
Choose a small test
Separate evidence from explanation
Write three columns: observed, inferred, unknown. “Sales fell 10%” may be observed. “Customers disliked the new price” is an inference unless investigated. Listing alternatives keeps a plausible story from becoming an unquestioned conclusion.
Worked example
A retailer tests a replenishment change in four fictional stores. Choose comparison stores with similar patterns before starting. Record the test dates, primary measure, budget guardrail, and stopping conditions. Changes in weather, promotions, or deliveries can complicate interpretation; a simple before-and-after comparison cannot isolate every influence.
Decision rule
Agree in advance what would justify adopting, revising, or stopping the change. For example: adopt only if stockouts improve, the inventory limit is respected, and reporting is complete enough to evaluate the result. This is an illustrative rule, not a statistically validated threshold.
Exercise
Write a one-page test plan. Include the smallest reversible action, who implements it, how the result will be measured, what else might explain it, and when the owner will review it. Avoid declaring success from a single favorable day.
05 / PRACTICE MANUAL
Run the review
A 20-minute meeting
Spend five minutes checking data quality and context, five interpreting the measures, five selecting an action, and five recording ownership. Keep the discussion anchored to the decision brief. If evidence is insufficient, record what is needed and when it will be available.
Decision log template
Date | Question | Options | Evidence | Assumptions | Chosen action | Owner | Review date | Result. Keep the original entry when the result arrives. Revising the original rationale after seeing the outcome removes the opportunity to learn.
What to publish on a dashboard
Show the decision question, a small number of defined measures, relevant comparisons, freshness, limitations, and an action owner. Provide a path to supporting detail. A beautiful display without a clear owner or decision can become reporting overhead.
Your next step
Use this process for one decision cycle. At review, ask: Did the evidence change our action? Was the definition repeatable? What information arrived too late? Improve the process before adding more charts. All numerical examples in this manual are fictional teaching examples.