By Tom Shuttleworth, Co-Founder, IllumiFi Advisory. 20+ years in financial and commercial data leadership.
This KPI data validation checklist has eight checks to run on each of your core KPIs and each check has a clear failure signal. When one fails, you will know which KPI to stop trusting and what you should do next, and you'll have a written test of KPI data quality.
KPI data validation means checking a reported figure against a rule or against its source before anyone needs to rely on it. It is the data version of proofreading. This checklist is written for business leaders or founders who rely on KPIs to assess business performance and progress and want proof of KPI reporting accuracy before the next board pack is produced.
Each KPI should have an owner and this individual should be responsible for its reported accuracy. They will need read access to the source system, the last management pack, and time with whoever builds it.
Start with the KPIs your board, lender or investors actually focus on, and run all eight checks on each. If you are short of time and only need to verify KPIs are accurate, begin with checks 3, 4 and 5. Write down what you find as you go, because the written record gives you the audit trail of what has been fixed.
If you want to know what unreliable numbers could cost, read how much bad data is costing your business.
A KPI must mean the same thing to everyone in the business.
Pick one KPI, such as revenue or active customers, and ask three people who use it to write down its definition separately, including what is in and what is out. Keep those people apart while they write. You have a failure if you get three different answers.
Solve this by writing a single definition for each KPI, including the date logic and the exclusions, and store it in a shared data dictionary: a list of every KPI and the meaning of each. If you are building the list from scratch, what is a KPI framework explains how definitions connect to objectives and decisions.
Every KPI needs one named person who can defend its definition and answer when it looks wrong.
List each KPI with the name of the person responsible for it. If the owner column is blank, or it says "finance", "the team" or "marketing", the check has failed. A team is not an owner, it needs to be an individual.
Assign one person per KPI. Their job is to own the definition and be the first call when the figure moves unexpectedly. A data owner is a named person who answers for a defined area of data, and it is the smallest unit of data governance.
You can trust a KPI when you can rebuild it from source data and land on the same answer shown against it in the reporting pack. This is KPI reconciliation.
Take one KPI for one closed month and rebuild it from the ledger or source system: pull the underlying transactions and apply the written definition. Compare your figure with the one in the reporting pack. A gap you cannot explain item by item is a failure. A difference you can explain, such as timing or tax treatment, is acceptable as long as you write the explanation down.
Then trace the gap to its cause, whether that is a missing feed, a filter nobody remembers or a manual adjustment. Leave the report alone until you know which side is wrong.
Two correct figures can still disagree if they were cut off at different moments.
List the cut-off date and time for each source behind your KPIs: the sales system, the billing system, the ledger. Compare them against the reporting period in your pack. If month-end totals differ by timing, the check has failed. Sales might close at midnight on the last actual day of the month, finance on the last working day, with returns only posting a week later.
Agree one cut-off for each KPI and put it in the definition. Where a timing difference is unavoidable, label it in the pack so nobody mistakes it for an error.
If a KPI is reported in more than one place, each must show the same value for the same period.
Choose one KPI and one closed month, then find it in two places, such as the board pack and an operational dashboard. Different values for the same month mean the check has failed.
Find out what each report does differently. It is usually a different definition or a different source. Decide which is the authoritative version, then point the other at it or retire the metric from one report.
Blanks and repeats in the data records behind a KPI make it drift no matter how carefully it is calculated.
Pull a sample of records from the table behind one KPI and look for blank fields and repeated entries. Start with about 25 records and widen the sample if you find a problem. Blanks where a value is required, or the same customer or order recorded twice, are the failure signal.
Find where the record is created, because defects found will typically start at the point of data entry. Tell the person who owns that area, set a rule for how the data is entered, reduce free text fields where possible and correct the existing records.
A restatement is a change to a figure that has already been reported. Restatements happen. Record each one, with the date, the reason and who approved it.
Ask to see the log of restated figures. If there isn't one, compare last month's pack with this month's: does the prior-period figure in this month's pack equal what was reported last month? No log, and prior-period numbers that have moved without a note, is a failure.
Start a simple log. Record the date, the KPI, the old value, the new value, the reason and who approved the change. A log stops arguments about which version was right.
A number should have a named approver before the pack goes out.
Look at the last pack that went out and ask who approved it, when, and on what basis. It has failed if there is no approval record, or if the approver did not see the results of checks 3 to 5.
Name an approver for each pack and record a one-line sign-off before it circulates. Make checks 3, 4 and 5 the evidence the approver sees.
The data foundation module that typically fixes each failure is in the last column.
| Check | What you do | What failure looks like | Data Foundation Module that typically fixes it |
|---|---|---|---|
| 1. One written definition | Ask three people to define one KPI | Three different answers | KPI Insight and Design |
| 2. Named owner | List each KPI with its owner | Blank, or "the team" | Data Governance |
| 3. Reconciles to source | Rebuild one KPI for one closed month from the ledger | An unexplained gap | Data Integrity |
| 4. Periods and cut-offs line up | Compare cut-off dates across sources and reports | Month-end totals differ by timing | Data Integrity (how reports are produced), or KPI Insight and Design (the definition) |
| 5. Same number everywhere | Compare one KPI in two reports | Different values for the same month | KPI Insight and Design |
| 6. Complete, no duplicates | Sample records behind one KPI | Blanks and repeats | Data Integrity |
| 7. Log of restated figures | Ask to see the log, or compare two packs | Numbers change silently | A rule you set; Data Governance owners and standards support it |
| 8. Sign-off before circulation | Check the last pack for an approval record | No approval record | A rule you set; Data Governance owners and standards support it |
A checklist tells you that something is wrong and where. It does not rebuild every figure from source across several systems, and it does not measure your data against defined quality dimensions.
Go further when any of these are true:
There are two deeper reads. A data quality audit for KPI reporting explains how data is measured against quality dimensions. A KPI reliability audit explains how a reported figure is traced back to its source.
Reporting data validation usually turns up faults that three of the five Data Foundations modules fix. Each is fixed-scope, ends in a named deliverable, and can be taken alone or as a set.
A UK multi-brand e-commerce retailer with annual revenue of £25m to £30m had reporting built on manual extraction from its e-commerce platform. We built an automated platform with a unified KPI layer, so definitions are consistent across all reporting, and the build was completed in under three months. The case study highlights the 750+ hours saved a year, a 71% twelve-month ROI and a seven-month payback. Read the e-commerce retailer case study.
We run your data and analytics, so you run the business. Our Analytics Platform gives you one documented data model and metric glossary, automated reporting and data connections that we maintain as your systems change.
If some of your checks failed and you would like a hand fixing the cause first, start with Data Foundations. If three or more failed, book a call and bring your last two packs.
Run all eight checks before a pack circulates for the first time, and again after any change to a core system or a KPI definition. Between those moments, repeat checks 3, 4 and 5 on one KPI each reporting cycle, since they are the quickest.
Finance should run them, because finance answers for the numbers and can read the ledger. Whoever administers your systems helps with access and exports but should not decide whether a figure passes. Ideally the person who built a report does not mark their own work. In a small business a founder can run the checks too.
The checklist is a first pass you run yourself to find out whether and where your reporting is unreliable. A full audit rebuilds figures from source across systems and measures the data against defined quality dimensions, ideally by someone who did not build the reports. Use the checklist first, and commission the deeper work if checks keep failing.
Do not just correct the figure. Note which KPI failed and how, name an owner, fix the definition or the source, then run the same check again. If the cause sits in how your systems connect, who owns the data or how KPIs are defined, the Data Integrity, Data Governance and KPI Insight and Design modules in Data Foundations cover those causes.