When your business numbers disagree: a practical data quality check

Treat the disagreement as useful evidence

Conflicting figures are frustrating, but they reveal something valuable: two parts of the business are answering different questions, using different records or taking their snapshot at different times. The aim is not to prove that one colleague has the ‘wrong spreadsheet’. It is to understand what each number actually represents.

Start with one number that matters to a real decision. It might be revenue expected this month, open work, active customers, sales conversion or jobs delivered on time. Do not try to reconcile every report in the business at once. A single disputed measure gives the team a manageable place to learn.

Write down the decision the number should support. ‘How much cash will arrive?’ needs a different definition from ‘how much work have we won?’ even if both have been labelled sales. Once the decision is explicit, the useful measure becomes easier to define.

Put the competing numbers on one page

Bring together the people who produce, use and understand the figures. For each version, record five things: the number, its source, the time it was taken, the rule used to include a record and the person who owns that rule. This is a working comparison, not a fault-finding exercise.

The differences often appear quickly. One report may count an opportunity when a quote is accepted; another may wait for an invoice. One may include cancelled jobs until month-end; another may remove them immediately. A report refreshed at 8am will not match one that includes changes made at lunchtime.

• Purpose: the decision or conversation the number is meant to support.
• Source: the system, file or manual step from which it is produced.
• Cut-off: the date, time and reporting period represented.
• Rule: what is included, excluded or treated as an exception.
• Owner: the person who can explain and approve that definition.

Test the difference against six common causes

The UK Government Data Quality Framework describes quality as fitness for purpose and uses six dimensions: completeness, uniqueness, consistency, timeliness, validity and accuracy. The framework is written for government, but the questions translate well to a small-business report.

Check whether expected records are missing, the same customer or job appears twice, or two systems contradict each other. Look for information that arrived too late for the reporting cut-off, values stored in an unexpected format, and records that simply do not match reality. These are different problems and need different fixes.

Take a sample of five to ten records that contribute to the disputed figure and trace them back to the original event: the accepted order, completed job, invoice, payment or customer record. A small, deliberate sample is often more revealing than scanning a whole spreadsheet for something that looks odd.

Agree a definition that a colleague can repeat

Write the chosen definition in one sentence, using the language of the business. For example: ‘Won work is the value of customer orders accepted in writing during the calendar month, excluding VAT, cancellations and draft renewals.’ Then add the source, update frequency, owner and any known limitation.

The definition should be specific enough for two people to produce the same answer. If phrases such as ‘active customer’, ‘completed job’ or ‘qualified lead’ still depend on personal judgement, name the event or status that marks the boundary.

Do not force unlike measures into one number just to remove disagreement. Finance, sales and operations may each need a different view. The useful outcome is a clear label for each view and one agreed measure for the decision in front of you.

One trusted number starts with one shared sentence about what it means.

Fix the earliest controllable cause

If staff repeatedly correct a report after export, the problem probably begins earlier. Change the first controllable point: make a field required, replace free text with a short set of statuses, agree when a record moves stage, or remove a duplicate place where the same fact is maintained.

Keep the correction proportionate. A weekly exception list may be enough while a team learns a new definition. A validation rule may prevent an impossible date or blank owner. A system integration is useful only after the fields and rules on both sides mean the same thing.

Where the records contain personal data, the accuracy question also has a legal dimension. The ICO says organisations should take reasonable steps to ensure personal data is not incorrect or misleading and should correct or erase inaccurate data as soon as possible where appropriate. The action needed depends on what the information is used for and the risk to the person.

Leave the room with a 30-day repair

End the hour with a small repair that can be checked within a month. Name the measure owner, the source of record, the definition, the change to make and the date the team will compare the reports again. If an old report will remain in circulation, relabel it or retire it so the ambiguity is not quietly reintroduced.

At the review, ask whether the figures now reconcile closely enough for the decision, whether exceptions are visible and whether producing the number takes less manual explanation. Do not claim victory because two totals happen to match once; check that the definition and process make the agreement repeatable.

Only then decide whether a dashboard, automated feed or wider data project is worthwhile. Better presentation cannot rescue a number the team has not defined. A reliable definition, owned at the point where the data is created, is a much stronger foundation.

• One decision the measure will support.
• One plain-English definition and source of record.
• One named owner for the rule and its exceptions.
• One upstream change to test for 30 days.
• One review date to confirm the fix is repeatable.

Sources and further guidance

The Government Data Quality Framework offers a detailed way to think about fitness for purpose and common quality dimensions. The ICO guidance explains the accuracy principle where personal data is involved. This article is practical guidance, not legal advice.

• GOV.UK: The Government Data Quality Framework
• ICO: Principle (d) — accuracy