DATIMORE · BUSINESS ANALYSIS WORKSHEET

Turn a business question into a useful next step

Choose a method that answers your question, then write down what the answer can and cannot tell you. This worksheet is a planning aid. All three examples and every number below are invented, not client results.

Save this HTML file for offline use, or print it. You can type in the planning fields before printing. Entries stay in this page; they are not sent anywhere or saved automatically. Keep a printed or PDF copy before closing or refreshing.

Choose the question first

What happened? → Description: Use counts, totals, rates or a typical value to describe the records you have. Define what is counted and which records are included.
Where should we investigate? → Diagnosis: Break down a result by a relevant stage, product or period and inspect the underlying records. A pattern suggests where to look; it does not prove a cause.
How do groups differ? → Comparison: Use the same measure and follow each group for the same length of time. Check how many records each group has, what it includes and which records are missing. A difference alone does not explain why it exists.
What might happen next? → Prediction: Use earlier dated records to make a forecast. Test it on later results not used to make the forecast. Also try a simple forecast, such as using last week's count again. Compare how far each forecast was from the actual result. A forecast is uncertain, not a commitment.

Choose records that fit the question. Numeric amounts support calculations; categories help group records; dates establish order and duration. Written comments need careful interpretation. A few selected records may tell you about those records, without showing what happened across the whole business.

Three synthetic worked examples

1. Retail: focus the returns review

Question: Which branch has the higher recorded return rate?

Inputs: For the same sales period, Branch A: 200 units sold, 20 returned. Branch B: 600 sold, 30 returned. Count distinct sold units once, with one return maximum per unit. Exclude cancelled sales. Both sets have complete records covering 30 days after each sale.
Method and output: Comparison: returned units ÷ sold units × 100. A: 20 ÷ 200 × 100 = 10%. B: 30 ÷ 600 × 100 = 5%. B has more returns but A has the higher rate. This describes these sales, not future sales.
Invalid or incomplete case: If B's follow-up window has not finished, its 5% is provisional. Stop the branch ranking until both windows are complete. If sold units are zero, the rate is undefined, not 0%. Do not mix unit counts with order counts.
Limit and next step: Different products or customers may explain the difference. The rates do not prove worse service or establish that the difference will persist. The retail manager checks product mix and recorded return reasons before choosing a change.

2. Manufacturing: explain the time behind a promise

Question: What was a typical completion time for these five finished orders, and which one needs investigation?

Inputs: Five selected orders took 2, 3, 3, 4, 18 working days from accepted order to completion. All use the same working-day calendar and start/end definitions.
Method and output: Description: after sorting, the middle value (median) is 3 days. The arithmetic average (mean) is (2 + 3 + 3 + 4 + 18) ÷ 5 = 6 days. Show both: the long order raises the average. Start investigating the 18-day order by checking the dates and notes for each stage.
Invalid or incomplete case: A sixth order is still unfinished and has no completion date. Keep it visible as unfinished; do not enter zero days or include it as a completed order. Do not delete the 18-day value merely because it is unusual. Check for a recording error first and explain any correction.
Limit and next step: These five selected, completed orders do not represent all orders or establish a delivery promise. Neither average identifies the cause of delay. The production planner reviews unfinished work, order types and a wider relevant history before making a customer commitment.

3. Professional services: test a simple starting forecast

Question: How close would last week's enquiry count have been as a forecast for the following week?

Inputs: Complete enquiry counts for four consecutive weeks: 8, 10, 9, 11. Each enquiry is counted once under the same definition. Keep week 5 out of the forecast calculation; its eventual count is 12.
Method and output: Prediction: use the last known count, 11, as the week-5 forecast. The forecast was 1 enquiry below the actual count: 12 − 11 = 1. This shows how to check one prediction on a later observation.
Invalid or incomplete case: If week 4 is only a partial week, do not use 11 as the complete-week starting value. If week 5 is used to choose or adjust this forecast, it is no longer an independent check.
Limit and next step: One checked week cannot establish reliable accuracy or a staffing requirement. Enquiries are not confirmed projects. The service manager needs a longer relevant history, repeated checks on later periods and evidence about workload per project before planning staffing. Campaigns, holidays or changed services may make past patterns unsuitable.

Plan one analysis for your team

Keep the question narrow enough to answer with the information you can verify. Assign a person to check the result and decide the next step.

What improvement do we want, and what is the exact question?

Which method fits, and why?

Which records, dates, definitions and group sizes will we use?

What must we check for missing or unsuitable data?

What is the result, with its calculation or evidence?

What does this result NOT establish?

Who checks it, what happens next, and when do we review it?