How to Handle Missing and Stale Data in a Weekly Marketing Report

The weekly report is due and one export is still missing. Last week's numbers are available, but copying them into this week's column would misrepresent the reporting period.

Leave unavailable current values empty and explain why. Put older figures in a separate column with their dates. You can then share the observations the evidence supports and hold back decisions that depend on missing data.

If you combine store, advertising and web-analytics reports in a spreadsheet, the review method below gives you a way to handle those gaps. It includes a working calculator and editable templates. The figures are synthetic, and the method has no measured improvement claim. It is not a platform integration or an accounting standard. Source documentation was checked on 1 November 2026.

Five rules for the next weekly report

• A reviewed zero is a measurement. A missing value is an unresolved question.
• Keep the original period and export time beside every figure. Recent download time does not prove current coverage.
• Count unavailable measures in your readiness inventory instead of deleting them from the denominator.
• One missing decision input can matter more than twenty completed rows.
• Use AI to draft a note from checked evidence, then verify it. Do not let the model invent a replacement value.

Why an export can exist and still be unready

First, find out why the figure is unavailable. You might need access to the account, or the source might still be processing data. Check whether the export covers the right dates and uses the definition your report needs. Sometimes the file is correct and simply waiting for review. The next action depends on which of these problems you find.

Google describes Analytics freshness as the time between collection, processing and reporting. Its documentation says processing can take 24–48 hours and figures may change during that period. It also distinguishes quicker intraday data from more complete daily processing. Treat these as documented characteristics to review, not a promise that every export is final at a fixed hour. Google Analytics data freshness.

Shopify separates dashboard summaries from more detailed reports and requires appropriate analytics access for staff. Record the actual report and scope you used, rather than writing only “Shopify” in the source column. Shopify Analytics documentation.

Neither source establishes that two similarly named metrics are equivalent. Before comparing, check their definitions, filters, currency, timezone and attribution rules. A mismatch may explain a difference; it is not automatically evidence that a campaign or connector failed.

Start with the decision and a fixed inventory

Write the decision at the top of the report. “Explain last week's order volume” is narrower than “Increase acquisition spend.” The latter needs cost evidence as well as an agreed outcome measure. If cost is missing, a complete order count cannot answer the budget question.

Next, list each required measure as a combination of measure, source and scope. For example, store orders for the agreed week are one item; web sessions for that week are another. Do not split one ready metric into ten rows to inflate completion. Do not remove a missing metric because it makes the report look untidy. Freeze the list before looking at the result.

Keep the inventory small enough to review yourself. Include the measures needed for the report, assign an owner to each one and record which decision depends on it. The readiness framework and worksheet provide a copyable starting point.

Use four statuses with clear meanings

Status	What it means	How to show it
Current and reviewed	The agreed complete period, definition and source-processing state have been checked	Show the value with its scope and review note
Stale	Only an older usable value is available	Leave the current cell empty; show the older figure separately with its dates
Missing	No usable value exists for the agreed scope	Show unavailable, its reason, owner and next check
Unchecked	A current export exists but still needs review	Mark it provisional and withhold dependent conclusions

Assign one status to each required item. If a current export exists but is unchecked, use unchecked even if an older reviewed figure also exists. Keep the older figure as a reference, not as a second inventory item. If the current export uses an incompatible definition, it is not a usable current value; record the mismatch and investigate.

A source-confirmed zero belongs with current reviewed data. The number zero should not be the default value of a missing spreadsheet cell or a failed API response. Keep the distinction in storage as well as presentation: use a nullable value field and a separate status and reason. Formatting alone cannot repair an ambiguous data model.

A small data contract that a marketer can maintain

For each item, retain the measure name, source report, period start and end, timezone, units, definition, filters, export time, value, status, review time and evidence reference. Add a baseline status and a decision dependency. Use roles rather than personal names in templates you plan to share.

The evidence reference can be an internal file identifier. It does not need to expose a private dashboard URL publicly. Keep raw customer records, contact details and credentials out of the reporting prompt. Aggregate figures are often enough to write the update, although aggregates can still be commercially sensitive and should follow your team's data policy.

For an integration, separate retrieval from eligibility. A successful response means the request returned; it does not establish that the requested period is complete, the account is right or the report is ready for a decision. Preserve the response's scope, validate its shape and check the row's business meaning before marking it reviewed.

In a spreadsheet, use a controlled status column and protect the original export tabs from casual edits. Put adjustments in a separate, explained column. A reviewer should be able to trace a sentence in the summary back to a row and then to the source evidence without guessing which copy was used.

Worked example: most rows are ready, but the decision is not

A fictional team has ten required measures. Six are current and reviewed. Two are available only for an older period, one is missing and one is awaiting review. Four of the six ready measures have a baseline with the same definition and equivalent complete window. One of the unready measures is acquisition spend, required for a proposed spend increase.

Ready coverage is 6 ÷ 10 = 60%. Comparable coverage is 4 ÷ 10 = 40%. Four measures remain unready. The affected spend decision stays on hold because its required cost evidence is missing. These percentages describe the inventory; they do not estimate the share of revenue explained or the probability that the report is correct.

Suppose reviewed orders are 100 this week and 90 in the equivalent baseline. The difference is ten orders, about 11.1%. That supports a descriptive comparison of orders. It does not establish that advertising caused the increase, that acquisition became more efficient or that the team should spend more. Each of those statements needs additional evidence.

A useful note would read: “Reviewed orders increased from 90 to 100 in the comparable reporting windows. Current acquisition spend is unavailable, so return on ad spend and the proposed spend increase are withheld. Older spend remains in a separate dated reference column. The analytics owner will recheck the export at the agreed review time.” Add the actual dates and time before sharing; do not let a template invent them.

Calculate coverage without turning it into a quality score

The free readiness calculator uses counts, not raw business values. It runs in the browser without an AI request or an account connection. Current, stale, missing and unchecked are disjoint groups. Comparable is a subset of current; unready decision measures are a subset of the other three groups.

Total = current + stale + missing + unchecked
Ready coverage = current / total * 100
Comparable coverage = comparable / total * 100
Unready = stale + missing + unchecked
Hold affected decision if unready decision measures > 0

A zero-item inventory returns undefined percentages. Calling it 100% complete would hide the fact that nothing was assessed; calling it 0% would suggest a measured failure. The calculator also rejects negative counts, decimals, non-finite input and impossible subsets. Its one-million-per-field ceiling is an input guard, not a recommended reporting scale.

The following JavaScript example uses the same pure calculation function as the tool. It runs from this project's root with a modern Node.js runtime. It has no network call, no credentials and no paid dependency. The surrounding tests also cover invalid values, empty inventories and scaling invariants.

import assert from 'node:assert/strict';
import { reportReadiness } from './src/lib/report-readiness.mjs';
const result = reportReadiness({
  current: 6, stale: 2, missing: 1, unchecked: 1,
  comparable: 4, critical: 1
}, true);
assert.equal(result.total, 10);
assert.equal(result.readyPercent, 60);
assert.equal(result.comparablePercent, 40);
assert.equal(result.unready, 4);
assert.equal(result.decision, 'Hold the affected decision');

Where AI helps, and where it should stop

Once the inventory is reviewed, a model can turn approved aggregate facts into a short update. Ask for four things: what is known, what cannot be concluded, who checks next and when. Give it explicit null values and status labels. Tell it to preserve them rather than “complete” the table.

The missing-data review prompt includes named inputs, an output shape, a synthetic example and a failure rubric. A fabricated current value fails the review regardless of how polished the rest of the output sounds. A person still checks every figure, date and recommendation against the evidence.

Writing those four sentences manually is the fallback. Copying exports, matching periods and calculating a percentage are ordinary reporting tasks; they do not become AI-assisted just because a model described them. If you need to map the wider process, the Workflow Architect can help draft a brief, but its suggestions also require practical review before implementation.

How to share a partial report

Give the report a visible version and review time. Separate current reviewed observations from dated reference data and unresolved items. Do not bury the limitation in a footnote while putting a complete-looking scorecard at the top. Place the unavailable label beside the affected measure and the withheld claim beside the affected decision.

Agree when the next check will happen and who will do it. That person needs access to the source. Close the item only when evidence for the correct period has been reviewed and included in the report. If the source is still processing, leave the item unresolved and record the next check; do not keep changing the number to make it appear final.

When data changes materially, publish a correction note: which value changed, the reason, the affected comparison and whether a previous decision needs review. Keep the earlier snapshot as evidence. Quietly overwriting it makes subsequent conversations about “what we knew at the time” unnecessarily difficult.

What this method will not solve

Equal item counts are not equal business importance. Coverage cannot certify accuracy, remove consent-related measurement limits, reconcile different attribution systems or establish causality. The decision-dependency check prevents a large percentage from hiding a known missing input, but it still depends on a person defining that dependency honestly.

Nor is current-reviewed a promise that a platform will never revise the data. Document the review boundary and revisit it when the evidence changes. For regulated financial reporting or a formal close, use the relevant professional controls rather than treating this marketing worksheet as a substitute.

For your next weekly report, agree the measure list before reviewing the numbers. Assign one of the four statuses to each measure, then identify any unresolved item that prevents the main decision. For help implementing the source checks and handoffs, see the forward-deployed marketer service.