What data should you prepare for a business review?

For a question such as “why are deliveries running late?”, order records need to be read alongside promised and actual delivery dates. Monthly sales totals alone cannot locate the delay. The aim is to prepare the smallest coherent set of records that makes the question examinable.

Fix the question and the period

Write down the question, its owner and the date by which a decision is needed. Identify the business unit, product group or customer population involved. Replace a broad ambition such as “improve operations” with an observable question: “why has delivery time increased for this product group?”

Choose the period to suit the question. Seasonality, a product change or a system migration can affect whether periods are comparable. There is no universal rule that three months of data will be enough. If older records use different definitions, make the change visible.

Give each dataset a short data card

Someone opening a spreadsheet should not have to guess what its columns mean. For each source, record:

• Scope: The facility, products, customers and period included.
• Unit of observation: Whether a row represents an order, an order line or a delivery.
• Origin: The source system, extraction date and filters used.
• Definitions: What dates, quantities, statuses and units mean.
• Linking key: How records can be matched across sources.
• Known gaps: Missing periods, changing codes and manual additions.
• Owner: Who can explain the definitions and how the records were created.

Data quality depends on its intended use. A populated field still needs to be checked for accuracy. Government Data Quality Framework.

Trace a total back to individual records

Match a management-report total to a sample of the records behind it. Different identifiers for the same order, split deliveries and returns posted in a later period can all change how totals should be read.

Suppose the sales table contains one row per order, while the warehouse table contains one row per delivery. If an order is delivered in two parts, joining the tables without accounting for this can count its value twice. Explaining the unit of observation makes that risk visible before analysis starts.

Preserve the original records. Use a separate working copy for corrections and record what changed. Do not fill unknown values with zero. Missing information may itself be relevant to the investigation.

Share an inventory before sharing detailed records

An initial conversation can often be framed using the types of records available, their periods and their coverage. Avoid attaching confidential contracts, employee information or a full customer list to the first message. Once the scope and appropriate sharing route are agreed, prepare only the records needed.

External benchmarks can provide context, but their populations and dates need to be checked. In Türkiye, for example, KOSGEB's Business Assessment Report provides historical and sector context. A current question about a particular business still requires relevant operating records. KOSGEB's description of the report.

Know when the package is ready

A starting package is useful when another person can understand the question and the limits of the records. Every gap does not have to be closed first. What matters is distinguishing questions that the available data can answer from those that need additional evidence.

If you have a defined business question, Probe frames an initial review around that question and the available data. Briefly describe the question and the types of records you hold.

Agree the sharing conditions

Before sending records, settle access, tools and retention. Use the data security guide to prepare these questions.