The report may point to a real problem. To decide what to do, we need to understand how it reached its conclusion.
- What data did it receive?
- What period did it analyze?
- What information does it not have?
- Which part of the result is calculation and which part is interpretation?
A useful report makes the path from data to observation, explanation and recommendation visible. The same standard applies to a bot, a dashboard and an analysis written by a human team.
Four levels worth separating
| Level | Fictional example | What needs checking |
|---|---|---|
| Data | The export contains no changes in the selected period | The account, the period, the filters and the coverage of the source |
| Observation | No interventions appear in the data analyzed | Whether the wording describes exactly what was extracted |
| Possible explanation | The setup may have been kept deliberately | The record of analysis and the reason for the decision |
| Recommendation | Ask for clarity on the management and check the results | Which decision the available information can support |
Jumping straight from an export to the verdict “no work was done” skips important questions. Equally, the answer “of course work was done, it just doesn’t show” needs evidence. Missing data should not be turned into a certainty that suits either side.
What does the tool actually see?
A system may have access to Google Ads but not to lead qualification in the CRM. It may see attributed conversions without knowing how many appointments were confirmed. It may receive only some of the campaigns, or an export with a forgotten filter.
That is why we define the sources, accounts and periods. We check the currency, the time zone, when the data was extracted and what each metric means. The difference between the date of the interaction and the date of the conversion can change the comparison. In the same way, an event sent twice can change the number of results reported.
A report that receives no sales data cannot, on its own, prove full commercial profitability. It can analyze the metrics available and say what information would be needed for the next level of assessment.
The API and the interface are not always the same view
In Google Ads, the documentation for change_event states that the resource does not necessarily include every change available in the interface history. Queries are limited to the last 30 days and to at most 10,000 results per query; retrieving more has to be handled accordingly. Source: Google Ads API, Change Event.
The implication for a monthly report is concrete: we check the window actually available, the types of changes tracked and whether the extraction was complete. If there is a history collected earlier and stored separately, the tool can use that record as well. The limit of one query does not automatically describe everything the system knows.
These details do not prove that a particular bot got it wrong. A verdict needs its implementation and its data. They show why a conclusion has to come with the source and coverage it relies on.
The metric can be right while the business question is a different one
A report can calculate the cost per submitted form perfectly. The client may need to know the cost per eligible inquiry or per signed contract. The relationship between these metrics has to be built and checked.
A fictional example. Campaign A produces more forms at a low cost. Campaign B produces fewer, but the sales team confirms a higher share of relevant inquiries. The choice depends on the cost and value of confirmed results, on volume and on the capacity to handle them. The label “best campaign” is not enough without the criterion behind it.
We can keep the intermediate metrics. They help with diagnosis. What matters is not presenting them as equivalent to the final result. The topic is developed in “A lead is not a customer.” and “The platform optimizes. But for what?”.
A simultaneous change does not explain the result on its own
If a drop in conversions coincides with a campaign update, we have a hypothesis to investigate. Perhaps the offer changed, perhaps measurement has a problem, or perhaps part of the demand moved to another period.
A tool can quickly generate possible explanations. The analysis has to show what evidence supports each explanation and what would contradict it. When the data cannot separate the causes, we say what remains uncertain and propose the right check.
A recommendation phrased with confidence can seem more convincing than one that states its limits. For the business, though, what counts is how well the conclusion holds up when we check what it rests on.
What a report that helps the decision looks like
Beyond the metrics, the report should answer a few questions: what changed, compared with what benchmark, what explanation the data supports, what we do not know yet and what action we propose. For important recommendations, we add the person responsible and the criterion for reassessment.
A fictional example of wording: “The export analyzed shows no changes in this period. We will check the export’s coverage and the management record. Recent results are not yet complete, and inquiry quality has to be confirmed with the sales team. After these checks, we will decide whether an intervention is needed.”
This wording neither clears the team of responsibility nor condemns the tool. It sets out what we know and turns the signal into a concrete check.
Automation can make the analysis more useful
Repetitive collection, checking thresholds and flagging inconsistencies are areas where automation can save time. Its value grows when the system keeps track of where the data comes from and separates facts from generated hypotheses.
The recommendation in a report and carrying it out are separate responsibilities. If a system is also given the right to change an account, the limits, approvals and checking of the result have to be defined separately. Access for reporting should not be confused with a mandate to intervene.
NOT MORE. can build or check the data flows, connect reporting with commercial information and turn conclusions into implementation. We can work with existing tools and specialists, including automations built by the client. Our role is set by the concrete problem and the access agreed.
For the case of campaign history, continue with “The campaign has not changed. Has it been managed?”. A good conclusion has to be explainable and verifiable, whoever or whatever wrote it.