Monday’s marketing report: Check the numbers before AI writes the draft

By Greg Nowak. Last updated 2026-09-11.

If Monday’s report shows a decline, it is natural to look at the campaign. But the initial data extract is not, by itself, evidence that the campaign is performing worse. The report also needs to tell you whether the figures are ready to be compared. Otherwise, you risk discussing budgets on the basis of data that is still changing.

For Danish business owners, operations managers and agencies, the weekly marketing report is a manageable starting point for incorporating AI into an automated workflow. The scope can be defined quite precisely: retrieve the agreed figures from GA4 and Google Ads, check the underlying data and calculate the changes in code. Then ask AI to write a short draft for the Monday meeting.

Choose metrics to support the decisions you need to make

Start with what the recipient needs to assess. Are you tracking ad spend, evaluating traffic or reviewing recorded results? Each metric should serve a clear purpose in the meeting. Keep the selection small enough for the report to be useful.

Write down the definitions before building the integration. Each metric needs a name, a data source, a defined scope, a unit and a calculation rule. For conversions, this means specifying which actions count. For revenue, identify the specific measurement and its source. A shared column labelled “results” becomes difficult to use if the word means different things in the two systems.

Comparing the most recent Monday-to-Sunday period with the previous week is a practical starting point. Show the actual dates in the report. Also agree on how to flag campaign launches, pauses and changes to measurement definitions, so the recipient can see when a comparison needs an explanation.

Use fixed queries to extract the data

Google’s guide to the Analytics Data API describes how to build reports using selected date ranges, dimensions and metrics. Dimensions determine how the figures are broken down, for example by date. Metrics are the figures you retrieve. This provides a documented basis for repeating the same data extraction week after week.

Google Ads also uses queries for reporting. Search and SearchStream support the same queries and return equivalent results. The difference is in delivery: Search splits the results into pages, while SearchStream delivers them as a stream.

Use fixed, reviewed queries for the agreed fields and accounts. Record the extraction time, query version and filters used. The integration must retrieve the full result, even when it spans several pages. Otherwise, the report may look complete despite missing figures.

Before combining data, check the periods, time zones and currencies. If the systems use different day boundaries, the integration needs to establish whether the available data can support comparable periods. Amounts in different currencies must be shown separately unless you have agreed on and documented a conversion.

Show how reliable Monday’s figures are

Google states that data processing in GA4 can take 24–48 hours and that report data may change during processing. There may also be temporary gaps in certain event-scoped traffic source dimensions, including source, medium and campaign. Sunday’s data may therefore still be provisional when the report is retrieved early on Monday.

Show the reporting period, extraction time and data status at the top. It should be easy to see whether a decline needs follow-up or whether the latest figures first need to be retrieved again. A technically successful extraction does not guarantee that the data is sufficient to comment on performance trends.

You can use the table below to agree on what the report should do when a check identifies an issue. The status labels are suggestions for your workflow, not official Google categories.

“Ready for drafting” means that your validation rules have been met. The figures may still change later. Save the version you send out, and clearly flag any later report that uses revised data.

Let code handle the calculations

Totals, differences and percentage changes should be calculated in a separate step. The standard percentage change is the difference between the current and previous values divided by the previous value, multiplied by 100. If the previous value is zero, the percentage change is undefined. The system must handle that situation explicitly.

Zero and missing data must also be kept distinct. A failed extraction must never appear in the report as zero ad spend or zero results. Leave the value missing and state which information could not be retrieved. This is a significant distinction for anyone assessing the week’s marketing activity.

For ratios, agree on the numerator, denominator and scope. Calculate the weekly ratio from the relevant totals; a simple average of daily percentages is unsuitable when the days have different volumes of data. Define rounding in one place as well, so readers see the same value in the table and the text.

Give AI a clearly defined writing task

Once the calculations have been checked, AI can receive a compact data package containing periods, metrics, calculated changes and caveats. Include short definitions so the model can label the results correctly.

Basic LLM Chain in n8n supports a fixed instruction with dynamic content. The documentation also describes how to connect a parser to control the response format. This can be used for the part of the workflow that writes the report text.

Ask for a short draft covering key observations, data limitations and points for follow-up. The instruction must require AI to reproduce only the supplied figures, retain the caveats and avoid new calculations. An observed change must be clearly distinguished from a possible explanation that someone needs to investigate.

A fixed response format controls the structure. The content still needs checking. Add a check to ensure that the values mentioned exist in the data package and that all mandatory caveats are included. Also have a designated reviewer read the draft before it is sent out, particularly while you are getting the workflow established.

Trial one report before the autumn campaigns

Start with one company, one report template and a named recipient. Run the report for a few weeks and check the calculations, data status and text. Also test what happens when data extractions are missing and values are zero. These cases need to be handled before the report becomes a regular part of the Monday meeting.

Then assess whether the report works in practice. How much editing does the text need? Does the recipient understand the caveats? Is it clear what needs investigating before you decide to change the budget? Use the answers to refine the workflow before adding more accounts or metrics.

Through nowa.dk, Greg’s AI automation service for Danish companies, Greg can help build this workflow: the API integrations, alignment of periods and currencies, and flags for missing or provisional data. Start with the Monday report you already use. It provides a concrete basis for agreeing on the figures, checks and approvals the automated version should include.

Related on GrN.dk

• Montørens talenote skal blive til en arbejdsordre – ikke rå lyd
• Fra salgsmøde til CRM: Automatisér opfølgningen uden datarod

Need help with this kind of work?

Let’s talk about your weekly marketing report Get in touch with Greg.

Sources

• Oprettelse af rapporter med Google Analytics Data API
• Rapportudtræk med Google Ads API
• Databehandling og aktualitet i GA4
• En enkel sprogmodelarbejdsgang i n8n