An operational report can show a sudden improvement because the business changed what it counts. Completed work may now exclude reopened cases; qualified enquiries may replace all enquiries; a new source may add previously unrecorded events. The values can be correct under each rule while the comparison remains misleading.
The proposed change log below records definitions and comparability before AI writes the explanation. It does not define the business's preferred metrics or claim a performance outcome. The information owner approves the actual rules and any restatement. The practical output is a measurement-change record that makes the break visible in both the data and the narrative.
Describe the definition beyond its display label
Create a metric ID and version with its purpose, population, unit, period basis, inclusion/exclusion rules, source systems and calculation. Lead, active customer and completed job are labels, not full definitions. Two reports using the same word may measure different populations.
Record the accepted event or record identities used by the calculation. A distinct-ticket measure differs from closure-event volume. A financial total needs a currency and sign convention. An average needs its denominator and treatment of missing values. Keep the owner's approval reference with the definition.
Document source coverage changes as well as formula changes. Adding another branch's data or a new collector can alter the measure without any change in underlying behaviour. The owner should decide whether the result is a new version of the metric or a different metric entirely.
Record the effective period and reason for change
Keep the old definition and its reports accessible. Record the new version's effective date, approving owner, reason and expected interpretation difference. A change made midway through a reporting period needs a specific transition rule; it should not be silently applied to whichever records produce a preferred total.
Distinguish corrections from intentional redefinition. Fixing a duplicate import may restate a value under the same accepted definition, while narrowing the population changes what the measure means. Both need history, but they require different explanatory language. The change log should retain that distinction.
Source: Google Sheets batch updates describes grouped spreadsheet changes and related operations. It can be part of a reporting implementation, but it does not version business meaning automatically. The team must preserve definition IDs, source snapshots and review decisions in its own accepted design.
Test whether earlier periods can be compared
Build a comparability decision for each proposed period comparison. Equivalent definitions with accepted coverage can be compared. Different definitions may be restated only when the underlying records contain the evidence needed to apply the accepted rule and the owner reviews the calculation.
If old records lack a newly required field, retrospective classification may be impossible. Do not ask AI to infer that field from a vague note just to produce a continuous chart. Keep the break visible and report the old and new series separately. An estimated bridge, if the organisation chooses one through appropriate analysis, needs its own accepted methodology and limitations.
A restated series should retain original values, revised values, calculation version and reason. It must not overwrite previously issued reports without a clear replacement record. Readers need to understand whether they are seeing historical data as originally reported or as later restated.
Reusable reporting-definition change log
Use this proposed record whenever a definition, population, source or calculation changes.
| Field | Required content | Review question |
|---|---|---|
| Metric identity | Stable ID, old and new definition versions | Same metric or a new measure? |
| Meaning | Purpose, population, unit and denominator | What does the figure now represent? |
| Rule change | Inclusion/exclusion, formula or source difference | Measurement change or data correction? |
| Effective period | Accepted date and transition rule | Which records use which version? |
| Authority | Owner and approval reference | Was the change accepted? |
| Historical coverage | Required old fields and available records | Can earlier periods be restated faithfully? |
| Comparability | Equivalent, reviewed restatement or break | What comparison is permitted? |
| Restatement | Original/revised values, method and source snapshots | Can a reviewer reproduce it? |
| Narrative note | Plain explanation of the measurement limit | Does the report avoid a false performance claim? |
| Release history | Report versions and changed sections | Is the earlier record preserved? |
A reusable narrative note is: “From [accepted period], metric [ID] uses definition [version], which [plain change]. Earlier values used [old version]. [Comparison is supported by reviewed restatement / direct comparison is limited]. The difference must not be interpreted as business improvement or deterioration without comparable evidence.”
The completion check is that the report names the relevant definition versions and supports each comparison with an accepted equivalence or restatement decision. A change log kept internally but omitted from a misleading chart is insufficient.
Walk through narrower qualification and missing history
Consider a hypothetical enquiry measure. Under definition A, an earlier period has 60 accepted enquiries. Definition B counts only enquiries meeting newly documented qualification criteria, and the later period has 40 qualified enquiries out of 80 total accepted enquiries. The 60 and 40 cannot be compared as one unchanged metric merely because both reports display enquiries.
If the old records contain all required qualification evidence and the owner approves the restatement, the team can calculate an earlier definition-B value and compare it with the later 40. That restated result needs its own source and calculation references. It should not be guessed from the earlier total.
If the qualification fields were not collected earlier, the report retains a comparability break. It can still report 60 earlier total enquiries and 80 later total enquiries under definition A where those scopes genuinely match, with the separate later qualified count. It does not call the qualified-count change a decline in demand or an improvement in lead quality without supporting evidence.
A duplicate version entry or repeated update request should link to the accepted definition-change record. Two definitions with the same label but different rules remain distinct. A late correction to one period creates a new report version rather than silently changing the basis of an already accepted narrative.
Give AI the definition contract with every value
Source: OpenAI Structured Outputs describes schema-constrained responses. A proposed narrative input can require metric ID, definition version, period, comparability state and accepted value. Schema conformity does not guarantee that the generated prose respects the measurement break. Review statements suggesting improvement, decline or causal change against the accepted comparison decision.
Source: OpenAI function calling describes application execution of requested functions. A report lookup can enforce the accepted metric/version pair and return the relevant limitation. The drafting step should not be able to alter the definition or overwrite a historical value to make a comparison possible.
Evaluate the process with fixtures covering narrower populations, added sources, changed denominators, corrected duplicates and incomplete historical fields. Inspect the chart labels and prose as well as calculations. A valid sum under the new rule does not establish an honest old-to-new comparison.
Questions about changed definitions
Can we retain the old metric name after changing the rule?
The display name can remain if the accepted reporting design makes the version and meaning clear, but the name alone must not imply continuity. Show the change and comparability limit where readers assess the figure. The owner should decide whether a separate name would better prevent misunderstanding.
Should we always recalculate all historical periods?
Only when accepted source evidence supports the new rule and the owner reviews the restatement. Missing historical fields may make it impossible. Keep the break visible rather than inventing data to preserve a continuous series, and retain original reports alongside any accepted replacements.
Does a lower count under a stricter rule prove better quality?
No. It shows the result of the accepted counting rule for that population. Quality or business improvement requires its own comparable evidence and interpretation. The narrative should distinguish the definition change from outcomes instead of treating a narrower measure as a successful intervention.
If your business needs help defining this process, explore Workflow automation, the wider AI automation services, and our custom-agent workflow guide. The agents and automation comparison and custom AI agent glossary explain the terms. To discuss your records and approval rules, get in touch.