To reconcile conflicting management reports, preserve both versions, agree the precise measure being compared and trace the difference to the underlying records. Check the reporting period, included population, calculation and refresh time. Record which differences are explained, which require a correction and who owns the next action.
Start with one disputed number that affects a real decision. If a service director cannot tell whether the team cleared its backlog, reconcile the case count before rebuilding the whole dashboard. Keep the decision and its deadline visible throughout the investigation.
Keep a copy of what people actually saw
Save the two reports with their extraction or refresh timestamps, selected filters and source versions. Include manual adjustments and any notes used to prepare the management pack. Keep these copies unchanged while investigating.
If the live dashboard refreshes during the discussion, a later result may no longer reproduce the discrepancy. A screenshot can show the figure and filters, but you will also need the permitted record-level extract or a way to reproduce the query. Use the organisation’s approved access and storage arrangements.
Write the issue narrowly: “The service dashboard shows 128 closed cases for the week; the management pack shows 112.” Add the decision affected, such as whether backlog-reduction work can end. “The data is unreliable” is too broad to investigate efficiently.
Agree what the number is meant to count
Ask the business owner to describe the measure in plain language. Then ask each report preparer to explain how their report implements it.
Check these details together:
- Business event: an enquiry received, a case closed, an order confirmed or a payment received.
- Unit: cases, customers, transactions, hours or a monetary amount.
- Period: the date field, start and end boundaries and any relevant time zone.
- Population: business entities, teams, products, statuses and exclusions.
- Calculation: a distinct count, a sum or a ratio with a defined numerator and denominator.
- Data state: the source version, last successful refresh and treatment of late corrections.
Two useful measures can describe different events. Give each a clear name and show how they relate. If both reports claim to show the same measure, these details become the agreed basis for testing them.
For a rate, compare the numerator and denominator separately. The same rounded percentage can conceal different populations, while averaging team percentages can produce a different answer from calculating the rate across all eligible records.
Find where the two record sets diverge
Work with the data team to compare records using the appropriate stable identifier. If the intended unit is a case, establish which records belong to each case before comparing totals.
Separate the findings into records present in both extracts, records present only in one and records whose relevant values differ. Examine the first point in the data flow where the results stop agreeing. That might be a source export, a transformation, a relationship between tables or a report filter.
Check what one row represents. A case joined to several contact or category records can appear on several rows. Counting those rows answers a different question from counting cases. Do not remove repeated identifiers automatically: repeated events or invoice lines may be valid for the measure being used.
Microsoft’s Power BI modelling guidance explains how the level of detail in fact tables and their relationships affect reporting. It is product-specific guidance, but checking what each record represents is a useful question in any reporting investigation.
Keep unexplained items visible. A balancing adjustment that forces the total to match removes the warning without establishing the cause.
A worked example: 128 closed cases or 112?
The following figures are illustrative, not client results.
The team agrees to count distinct eligible cases with a closure date within the reporting week. Both reports are supposed to use that definition. Comparing the saved extracts reveals:
- The dashboard counts 128 rows after cases have been joined to contact-category data.
- Seven extra rows repeat cases already counted because those cases have more than one category. Removing that multiplication leaves 121 distinct cases.
- Of those distinct cases, nine closed before the reporting week. The dashboard filtered the extract date instead of the closure date.
- Applying the agreed closure-date rule leaves 112 eligible cases, matching the management pack.
The seven repeated rows and nine out-of-period cases are separate findings, so they are not subtracted twice. The reconciliation explains the full difference: 128 minus 7 minus 9 equals 112.
The fix needs to address both the calculation and the date rule. The reporting owner should also check whether earlier management decisions used the affected figure. Matching this week’s total alone does not establish that every other period is correct.
Give the correction an owner and an acceptance check
The business owner agrees what the measure means. The person responsible for the source or model makes the relevant correction. The reporting owner checks that the published result and its explanation meet the agreed need. One person may hold several roles, but the responsibilities should be explicit.
The Government Data Quality Framework guidance recommends investigating causes, improving data close to source and monitoring quality against its intended use. Its context is government data; the same questions help a business distinguish a temporary reporting fix from a recurring quality problem.
For the example above, acceptance checks could include a case with several categories, one at the reporting-period boundary and a normal single-category case. Re-run a second period as well. Confirm that totals and permitted breakdowns behave as expected and that access restrictions still apply. Record who reviewed the evidence.
If a correction changes a previously circulated figure, issue a labelled revision through the normal reporting process. Explain the change and identify any decision that needs reconsideration. Keep the original version available for traceability.
A reconciliation note your team can reuse
Use the reporting issue log you already have. A useful entry contains:
- Decision affected: what is waiting for a dependable number and by when.
- Reports compared: versions, timestamps, filters and the two values.
- Agreed definition: event, unit, period, population and calculation.
- Difference explained: record groups and amounts or counts, with unresolved items separate.
- Cause and correction: where the error arose, the proposed change and the owner.
- Acceptance evidence: test cases, comparison periods, reviewer and result.
- Publication action: revised reports, people to inform and the next quality check.
Close the issue when the cause is explained, the correction is checked and the people using the number know what changed. For the wider reporting cycle, the management reporting pack guide covers how to organise information around leadership decisions. The decision-rights guide can help when nobody is authorised to settle a definition or approve a correction.
When this needs a wider BI review
If the same discrepancy returns every month, trace the shared source and calculation rather than repeating the manual adjustment in each pack. LUKiN’s Business Intelligence reporting case describes work on multiple sources, KPI definitions and a consistent reporting model.
Business Intelligence consulting is the relevant next step when you need to connect sources, agree definitions and validate the reporting with its users. For an initial conversation, describe the disputed measure, the systems involved and the decision it is holding up. You do not need to send sensitive extracts through the website.
If the reporting problem is mixed with unclear processes or responsibilities, the Operational Performance Diagnostic can help identify what to address first.
Sources
- Government Data Quality Framework: guidance, published 3 December 2020.
- Microsoft Learn: Understand star schema and the importance for Power BI, last updated 30 December 2024.
Sources checked on 21 November 2026. The worked figures and reconciliation note are LUKiN’s practical examples.
