AI Runs My Reports
I used to spend Sunday evenings assembling the week’s reports for Monday morning. Three platforms, a spreadsheet, and 2 hours of copy-pasting data I was already too tired to properly interpret. Now the reports arrive in my inbox before I wake up — assembled, narrated, and ready to act on.
What Changed
| Report | Previous Method | Now | Time Saved Weekly |
|---|---|---|---|
| Weekly client updates | Manual data pull + written summary per client | Make.com + Claude narrative, auto-delivered | 3 hrs/week for 4 clients |
| Sales pipeline report | CRM export + Excel manipulation + email | GHL data + AI narrative, delivered Monday 7am | 90 min/week |
| Marketing performance | GA4 + social + ad platform, manual | Make.com multi-source + AI analysis | 2 hrs/week |
| Cash flow summary | Xero export + manual commentary | Xero API + Claude forecast narrative | 60 min/week |
| Team utilisation | Time tracking export + manual calc | Automated from PM tool + AI insight | 45 min/week |
| Monthly board pack | All of the above + PowerPoint | All sources + AI narrative + template | 6 hrs/month |
How It Actually Works
Data collection layer
The foundation of every report automation is reliable data collection. Make.com scenarios run on schedule — most at 6am on the appropriate day — and collect data from each source via API: the Google Analytics 4 API returns session counts, new users, top pages, and conversion events. The GoHighLevel API returns pipeline value by stage, new contacts, and emails sent. The Xero API returns revenue received, outstanding invoices, and bank balance. Each API call produces structured data that is stored in a Bubble.io MetricRecord database for historical comparison. The data collection runs without any human involvement — the APIs return the numbers, Make.com stores them.
AI narrative generation
After data collection, a second Make.com module calls Claude with the collected data and a context-specific prompt. The Monday client update prompt: You are generating a client project update for [client name]. This week’s project data: [data]. Last week’s data for comparison: [comparison data]. Generate a professional 3-paragraph client update covering what was accomplished this week, what is planned for next week, and any decisions needed from the client. Tone: confident and specific — reference actual numbers and actual deliverables, never vague phrases like good progress. The report narrative emerges in seconds — specific, professional, and ready to deliver.
Formatting and delivery
The narrative plus the raw data is formatted and delivered. For client-facing reports: the AI narrative is formatted into the client portal (Bubble.io — the client can read it when they log in) and emailed from the account manager’s address. For internal reports: the AI narrative is posted to the relevant Slack channel and emailed to the report owner. For board-level reports: the narrative and data are formatted into a Google Doc template via the Google Docs API and shared with the board via a secure link. The delivery is as automated as the generation — nobody needs to copy-paste, format, or send.
The Monday morning experience
What used to be a 2-hour Sunday evening of report preparation is now a 15-minute Monday morning review. The reports are already prepared and delivered. My job is to read them — to actually absorb the insights rather than being too absorbed in production to notice what the data is saying. The quality of decisions made with properly absorbed reporting data is demonstrably higher than decisions made after 2 hours of tired data manipulation. The AI reports are not just faster — they are better inputs to better decisions.
📌 The most important investment in a reporting automation system is the prompt design for each report type. Spend 30 minutes designing the prompt before building the Make.com scenario — test 5 variations of the prompt with real data and select the one that consistently produces the most useful narrative. The prompt is the intellectual capital of the system; the Make.com build is the infrastructure that runs it. Get the prompt right and the infrastructure delivers value indefinitely.
How do I handle reports that require context AI cannot know?
Some report narratives benefit from context that is not in the data — a client who mentioned in passing that they are going through an acquisition, a team member who has been dealing with a personal situation, a market development that explains a traffic anomaly. Build a notes field into the report generation workflow: a Slack or email prompt sent 30 minutes before the report generates (hey, any context to add to this week’s report?) that allows you to add 2 to 3 sentences of human context. Claude incorporates the context into the narrative. The report has the analytical rigour of AI and the contextual depth of human knowledge.
What if one of the data source APIs is down when the report should generate?
Build error handling into every reporting scenario: if any data source fails to return data, the scenario sends an alert (Slack message or email) rather than generating a partial report. The alert specifies which source failed and provides the manual fallback (how to get the data manually in 5 minutes). The report is delayed until the data is available — either by automatic retry when the API recovers, or by manual data entry into the fallback form. A partial report with missing data is more dangerous than a slightly delayed complete report — decisions made on incomplete data produce unreliable outcomes.
Want Your Reports Automated?
SA Solutions builds Make.com reporting pipelines that collect data from all your platforms, generate AI narrative, and deliver formatted reports on schedule.
