croissant
Home / Blog / Reporting

Missing data in agency reports: when zero is the wrong answer

Croissant editorial team · · 6 min read

Keep missing, delayed and withheld marketing data separate from zero. A practical agency reporting guide with a worked example and human review checks.

Show zero only when a checked source supports a zero for the agreed metric and reporting window. Show pending when processing is still expected, unavailable when collection failed, and withheld when the source explicitly limits disclosure. Keep those states visible in charts, calculations and AI summaries. A blank cell cannot tell your account lead which decision is safe.

Give every number a reporting state

An agency report can look complete while one connector has returned nothing. If the workflow replaces that gap with zero, the client may read a collection problem as a performance collapse. Start by giving each metric a value and a separate state. The following conventions are Croissant’s recommended workflow design; they are not standardized labels shared by every reporting product.

Use observed for a value retrieved and checked against the agreed account, period, filters and metric definition. Zero can be an observed value. Use pending for a known processing wait, unavailable for a failed or unresolved collection, and withheld only when source evidence supports that explanation. Use not applicable when the metric does not belong in this report. An empty response with no explanation stays unresolved.

Keep freshness separate. A retrieved value can be observed but provisional. Google Analytics documents processing delays and figures that can change during processing; its timing guidance is specific to Analytics and is not a deadline for every platform. Record the extraction time and the source’s relevant processing limits rather than declaring every recent number final. [1]

Check why a row is absent before choosing its label

First confirm the correct account, reporting dates, time zone, filters and requested metric. Then inspect whether collection succeeded and whether the response covers the intended report. A successful request alone does not prove that a missing campaign row means no activity. The reporting owner must understand the source’s response rules before converting absence into zero.

Privacy limits are another possible explanation, but they need evidence. Google Analytics says thresholds can withhold data in reports and explorations, including certain demographic and search-query contexts. Its data-quality indicator can identify thresholding. That does not mean every absent campaign row is thresholded. Preserve a confirmed withholding state without trying to reconstruct hidden individuals or counts. [2]

If you cannot identify the cause, label the metric unavailable with a short factual note, such as “No usable value returned; reporting owner investigating.” Avoid guessing “tracking broken” or “no conversions.” Those explanations can change what the client does next.

Worked example: one missing channel changes the decision

This example is entirely synthetic. A fictional agency tracks accepted enquiries from three mutually exclusive channels under one approved definition. Last week, Channel A had 12 enquiries, B had 8 and C had 5, for a total of 25. This week, A has a checked count of 10 and B has a checked count of zero. Channel C’s export fails, so its count is unavailable.

Replacing the failed export with zero produces an apparent total of 10 and an apparent decline of 60% against 25. The arithmetic is correct for those substituted inputs, but the all-channel performance claim is unsupported. Channel C’s actual count is unknown. The report should show “10 observed enquiries across A and B; C unavailable; all-channel total and change withheld pending collection.” Here, “withheld pending collection” describes the agency’s display decision, not vendor privacy thresholding.

If the account lead needs a like-for-like view, the checked A-and-B subtotal is 10 this week versus 20 last week, a 50% decline for those two channels only. Label the restricted scope beside the figure. It cannot establish the change across all three channels. The account lead decides whether that partial view is useful enough for the client discussion.

B’s zero remains visible because the example assumes a verified source value for that scope. It should not disappear merely because another channel failed. Keep the missing state attached to C, then hold only the totals and conclusions that depend on C.

Carry the state through calculations and AI text

Begin with a small metric record: source reference, account label, metric definition, period, time zone, value, state, extraction time, reason and investigation owner. Keep private source records in the team’s approved system. The client-facing report usually needs a readable limitation, not raw connector diagnostics.

Have the calculation step check required inputs before computing a total or trend. A required unavailable input should make the dependent result unavailable. A deliberately partial subtotal needs an explicit list of included channels and a matching comparison period. Never silently reuse last week’s number or interpolate a missing value to keep a chart smooth.

Pass AI the checked figures, allowed scope and unresolved states. Ask it to preserve those limits. In the synthetic example, an acceptable draft says that A and B recorded 10 enquiries and C remains unavailable. A draft saying “overall lead generation fell sharply” needs rejection because it broadens a partial observation into an overall conclusion.

For collecting sources and reconciling definitions across a complete report, use our weekly agency reporting guide. The decision here is narrower: what each empty or delayed input permits the report to say.

Assign the next check and test the release boundary

Give each unresolved metric one owner and a next check tied to the actual cause. A processing wait needs an agreed recheck. A collection error needs investigation. A confirmed privacy threshold needs a transparent limitation; repeated refreshes may not resolve it. A person should approve changes to reporting scope rather than allowing the automation to widen dates or drop filters silently.

Test with synthetic inputs before connecting client reports. Include an explicit zero, an empty response, a failed export, a provisional value and a confirmed withheld row. Verify the displayed state, dependent calculation and generated sentence for each. Also test recovery: a later successful export should create a candidate report for review without silently replacing an issued copy.

Keep budget changes and client distribution with an authorized person. A reporting exception is not permission to pause campaigns, send explanations or alter tracking. If figures arrive after an issued report, follow the separate client report corrections guide.

A protected spreadsheet may be enough for a small report. Consider a connected workflow when collecting state evidence and routing exceptions recur across accounts. Use our reporting demo as context, then bring one fictional report and its missing-data rules through the Croissant contact form. Keep client exports and credentials out of the initial enquiry.

Implementation checklist

  • Define when a source value qualifies as an observed zero.
  • Keep value, availability state and freshness separate.
  • Check account, period, filters and collection evidence before interpreting an absent row.
  • Hold dependent totals when required inputs are unavailable.
  • Label partial subtotals and compare the same included population.
  • Test missing states in calculations, charts and AI summaries.
  • Assign an investigation owner and next check for unresolved inputs.
  • Require human approval for client distribution and consequential decisions.

The sources behind the sauce.

Primary research and official documentation behind the cited points. The workflow examples and checklists are Croissant’s own guidance.

  1. Google Analytics Help — Data freshness and processing intervals
  2. Google Analytics Help — About data thresholds and withheld rows