Numbers describe; sentences explain
Weekly analytics exports are skimmed and then forgotten. In some cases the figure has risen by 18 percent, and no one knows whether this is due to a campaign, a seasonal trend, or a change in tracking. So no action is taken, which is the appropriate response when a number is unexplained.
To explain things properly, you need context the analytics tool does not provide, for example what was shipped, what campaign was started, and what broke. Joining those two datasets is straightforward, but no one does it every week.
What makes this safe is the distinction between observation and hypothesis. "Organic sessions decreased by nine per cent" is data. "Because the migration changed the URLs" is a guess. It is wrong to draw confident conclusions when a report presents both in the same tone.
Preparing a report for a channel that no one is working on is a habit, not a way to make decisions, so the scope of this exercise is smaller than the scope most teams set.
How the ActionFlow runs on the canvas
When a Start Scheduled trigger fires, an HTTP Request node retrieves the current window and the comparison window.
The Google Sheets node supplies the campaign calendar and the release log. If that second source is not included, the flow would only show movement, which the dashboard already does.
The Generate Object node returns the analysis as a structure, with the observation and the hypothesis kept in separate fields and each hypothesis having a confidence value. That is why they stay separate in the writing as well.
The Filter node removes any movements below the threshold you set, which is why the report shows three changes rather than thirty.
The Generate Text node produces the narrative from the filtered structure, provided that each hypothesis is stated in a single form. Then the Slack node posts it.
A Google Sheets node records each hypothesis. The only way to find out whether the explanations are worth reading is to check afterward which guesses were correct.
Nodes this ActionFlow uses
- Start Scheduled: runs on the reporting cadence.
- HTTP Request (Util): pulls the current and comparison analytics windows.
- Google Sheets (Integration): supplies the campaign calendar and release log.
- Generate Object (AI Core): returns observation and hypothesis as separate fields with confidence.
- Filter (Data): drops movement below threshold so the report stays short.
- Generate Text (AI Core): writes the narrative with speculation phrased as speculation.
- Slack (Communication): posts the report.
- Google Sheets (Integration): logs hypotheses so they can be checked later.
Frequently asked questions
Start building AI workflows
Create a free account, open a template or a blank canvas, and run your first ActionFlow.
Newsletter
Get product updates
New nodes, agents, and product notes. We send mail only when we have something worth opening.