Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics › Business Intelligence for Marketing

Correcting a Number After a Report Has Gone Out

What to do when a figure in a sent report or live dashboard is wrong: confirm and size it, fix the source, restate it and tell the people who acted on it.
Tutorialintermediate9 min readUpdated

Why a quiet fix makes trust worse

When an analyst finds their own mistake in a sent report, the easiest path is to fix the query and let next month's figures be right. Nobody has to be told. The trouble is that somebody will notice: the finance partner who compares the new month with the old one, the manager who quoted last month's figure in a budget request, the agency whose results suddenly look different. A difference nobody explained looks like either incompetence or concealment, and the troubleshooting page on finance distrusting marketing numbers describes how quickly one unexplained figure spreads doubt to every other. The alternative is an open correction: the figure restated with a dated note saying what was wrong.

An open correction does the opposite. It shows that the team checks its numbers after sending them, that it says when they were wrong and that it changes how it works so the error does not recur. Corrections handled this way are usually remembered as evidence the numbers can be trusted. The academy exercise on restating a stationery brand's report after a tracking outage practises sizing a wrong figure that led to a budget cut, restating the report and telling the people who acted on it, in the right order.

Not every surprising number is this kind of error. The first job is to place it:

What happenedWhich process handles it
Tracking stopped or double-fired, so data was missing or inflated at collectionThe tracking outage SOP, with annotation
A definition was changed on purposeVersioning in the metrics layer guide
A dashboard's logic was changed through its normal processThe change process in dashboard governance
A calculation, join, filter, export, copy or early load produced a wrong figure that readers have seenThis tutorial
A figure was caught before the report was sentThe pack's own checks, as the monthly reporting pack SOP sets them

Before you start

  • The report or dashboard as it was sent, from the archive, not the live version that may already have changed.
  • The source of truth for the measure in question: the order system, the ledger, the CRM or the platform's own report.
  • The working files: the queries, exports and spreadsheets that produced the figure.
  • The distribution list for the report and any scheduled deliveries, and the viewers of the dashboard if the tool lists them.
  • The measurement change log, open at a new entry.
  • Someone senior who knows the correction is coming. A correction that surprises the head of the team is a second problem.

The steps

  1. Confirm the error against the source of truth

    Before telling anyone, prove that the number is wrong and that the replacement is right. Reproduce the sent figure from the saved query or export, then reproduce the correct figure from the source of truth by a route that does not share the faulty step. The guide to explaining metric changes starts every investigation by doubting the change; apply the same doubt to the correction. Check that the two figures use the same definition, period and time zone, and that the source itself was complete when it was read. A correction that later turns out to be wrong as well does more damage than the original error.

  2. Size the difference for every period and segment it touches

    Work out how far the error reaches. Find when the faulty step started, from the query history, the change log or the date a table or export changed, and recompute every affected period. Recompute every split that the report shows, such as channels, markets and products, because an error that is small in total can be large in one segment. Note which derived figures move with it: an inflated order count lowers cost per order, raises conversion rate and changes every comparison against those months. The guide to comparing periods is a reminder that an error in a base period distorts every later comparison with it. Write the old and new values side by side in one table.

  3. Find everyone who received the number and what they decided with it

    List every place the figure went: the report's distribution list, forwarded copies, decks that quoted it, the dashboard's viewers, and any scheduled email delivery. Looker Studio's scheduled delivery sends a PDF to the recipients its owner chose, rendered with the creator's access, as the sharing and embedding guide explains, so the schedule's recipient list is part of the audience; the Looker Studio help centre describes where schedules are managed. Then ask the main readers what they did with the number. Budget moved, a target set, a figure reported to the board or a client, a campaign scaled or cut: each of those is a decision the correction has to reach. The analytics request process records who asked for one-off analyses and why, which makes this step faster.

  4. Fix the cause at the source rather than in the report

    Correct the query, the model, the export or the procedure that produced the error, not the cell in the report. Use the five whys to reach a cause that can be changed: the figure was wrong because orders were counted twice, because the join was on order lines, because a new line type was added to the orders table, because nobody checked the table's grain after the change. The SQL guide describes the commonest causes, joins that fan out and filters that drop rows; others are exports taken before a source had finished updating, which is a data freshness problem, and figures copied by hand from the wrong tab or period. If the fix changes how a certified dashboard calculates anything, it goes through the dashboard governance change process like any other change.

  5. Restate the affected figures and keep the old ones findable

    Rerun the affected periods with the fix and publish the restated figures in the places the old ones lived: the live dashboard, the reporting tables and the archive. Do not overwrite the sent version. The metrics layer guide's rules for deliberate changes apply here too: the old value stays available, marked as superseded with the date, and a before-and-after is published for the affected periods. The monthly reporting pack SOP archives each pack as final so that later corrections are made as a new version; save the corrected report beside the original with a version name that says so. A certified dashboard can carry a notice in its header for a period, as the governance guide suggests for any change that moves its numbers.

  6. Write the correction note

    Write a short note that a reader can understand without the analyst in the room:

    PartWhat it says
    What was wrongThe figure, the report or dashboard, and the periods affected
    Old and newThe sent and the correct values side by side, for every affected figure that readers saw
    WhyThe cause in one or two plain sentences, without jargon or blame
    What it changesWhich conclusions in the original report still stand and which do not
    What has been doneThe fix, the restatement and the check added so it does not recur

    Lead with the corrected figure and what it changes, not with the apology or the technical story. Whether the note goes out as a reissued report or as a footnote in the next one follows from what readers decided with the number: a decision made on it needs a reissue and a direct conversation, a figure nobody acted on can be footnoted. The monthly marketing report template has a place for data notes where a footnoted correction can sit.

  7. Tell the people who acted on it first and everyone else second

    Speak to the people who made decisions on the number before the general note goes out, in person or by direct message, with the old and new figures and what it means for their decision. They should never learn of the error from a distribution list or a colleague. Agree with them whether the decision stands, needs revisiting or needs reversing; that is their call, informed by the corrected analysis. Then send the correction note to everyone else who received the report, and to the owners of any decks or documents that quoted the figure, so the old number stops circulating.

  8. Add the test or check that would have caught it

    Ask what check, run before the report went out, would have stopped this error, and add it. For a table fault, the data tests tutorial turns a rule into a query that fails loudly with each load: a unique key test catches duplicated orders, which is the Deduplication rule applied as data arrives; a freshness test catches a table that had not finished loading; a reconciliation test catches a total that no longer ties to the source. For a manual fault, such as a figure copied from the wrong tab, the fix is a step in the reporting procedure: a reconciliation against the source of truth before sending, or a second person's check, as the routine for reviewing an analysis before it is shared sets out for one-off work.

  9. Record the correction in the change log

    Write one dated change log entry with the error, the periods affected, the old and new values, the cause, the fix, the people told and the test added. Link the entry from the restated report and from the dashboard's own change log if it has one. The change log is what the next analyst reads when a trend chart shows a step they cannot explain, and the record of corrections is also evidence, over time, of which parts of the reporting need more checks. If the restated figures change a series that other teams model or forecast from, tell those teams through the same log.

A bakery chain corrects its cost per order (illustrative)

Illustrative example — not a real result

Sulmonde Bakeries restates three months of orders

Sulmonde Bakeries, an invented regional bakery chain, adds a collection-slot line to click-and-collect orders in April. The monthly report's query joins orders to order lines, so from April every click-and-collect order is counted twice. In July a finance analyst notices that reported orders exceed the till and web order totals.

MonthOrders reportedOrders correctMarketing spendCost per order reportedCost per order correct
April9,1208,050£24,150£2.65£3.00
May9,5108,370£24,900£2.62£2.97
June9,8408,610£25,830£2.63£3.00

On the strength of a cost per order that seemed to fall from about £3.00 in March to £2.65, the marketing director had moved £4,000 a month from local radio into a click-and-collect promotion. The analyst tells the director first, with the corrected table; the director keeps the promotion for one more month as a measured test rather than a proven win. The April to June reports are reissued as corrected versions with a note, the originals stay in the archive marked superseded, and a unique order key test now runs with every load. The change log gains 1 entry, linked from all 3 reports.

Common mistakes

  • Fixing it quietly. The next report differs from the last and nobody explains why.
  • Correcting before confirming. A second wrong number is worse than the first.
  • Sizing only the total. The error that is small overall can be large in the segment someone acted on.
  • Patching the report. A cell typed over in the report leaves the faulty query to produce the error again.
  • Overwriting the original. Readers who kept the old figure cannot see what changed.
  • Telling everyone at once. The person who moved budget on the number learns from a distribution list.
  • Leading with the apology. Readers need the corrected figure and what it changes first.
  • No new check. Without a test, the same join fails the same way next quarter.

The business intelligence hub lists the other guides in this section. The academy capstone on a measurement governance plan for a used camera dealer practises setting how wrong figures are corrected and communicated as one part of a year's plan, beside the outage procedure and the audit.

Frequently asked questions

Should we reissue the report or just add a footnote?

It depends on what readers did with the number. If someone moved budget, changed a target or reported the figure upward, reissue the affected pages with a correction note and tell those people directly. If the error was small enough that no decision would have differed, a footnote in the next report, with the old and new figures and the cause, is honest and proportionate. The test is the decision, not the size of the percentage change.

Is it better to wait until the next scheduled report to correct the number?

Only when nobody is about to act on it. A wrong number keeps being quoted for as long as it stands: in decks, in budget requests and in the next comparison against it. If a decision is pending or the figure is live on a dashboard, correct it as soon as the error is confirmed and sized, then carry the restated figures into the next scheduled report as well.

How is this different from a tracking outage or a definition change?

A tracking outage means data was never collected, and the outage response procedure annotates the gap and explains it. A deliberate definition change, such as adding fees to cost per lead, is versioned and announced in advance through the metrics layer. This tutorial covers the remaining case: data that existed and was correct at source, but was calculated, joined, filtered, exported or copied wrongly on its way into a report that people have already read.

Terms used on this page

Frameworks

Where this page is used

Part of these learning paths

Sources

  1. Looker Studio Help — Looker Studio Help
  2. Google Analytics Help — Google Analytics Help

Related