Weekly client reports now put themselves together and save managers 4–6 hours a week
An ad agency works with 8 clients. Every week its managers used to spend 4–6 hours on reports built from Facebook, Yandex Direct, Yandex Metrica and Google Analytics data. Now a workflow in n8n, an automation tool, builds the reports, and AI comments on sharp changes in the numbers. Nobody assembles reports by hand anymore.
Every week the managers built eight reports by hand
The agency works with eight clients at once: online stores, service businesses and projects that collect customer inquiries. Every client needs a report for the past week on Monday morning: ad spend, cost per thousand impressions and per click, click-through rate, orders and inquiries, return on ad spend, how each creative did, sharp changes and a manager's comment.
Each report took a manager 30–45 minutes on average. The manager had to export data from the Facebook ad account, switch to Yandex Direct, pull Metrica, add Google Analytics data, put it all together by hand in a Notion or Google Slides template, explain the spikes and dips, and send the report.
In total, from Friday to Monday the team spent 4–6 hours of manager time every week, and that was if the client asked no follow-up questions.
A workflow in n8n builds the report, and AI writes the commentary
n8n runs on its own server and starts the chain of steps every Monday at 06:00. By 09:00 the reports for all eight clients are in Notion and have been sent out by email or Telegram. A manager can look them over and add a paragraph if needed.
Every Monday at 06:00, for all 8 clients one after another
Facebook, Yandex Direct, Metrica and Google Analytics are queried at the same time
All metrics go into one table; the workflow calculates week-over-week changes and flags sharp deviations
Claude gets the data and writes 3–5 paragraphs explaining the charts
Each client gets a Notion page or a PDF through the channel they prefer
Data from four platforms is collected at once
n8n queries four sources at the same time: the Facebook Graph API (ad account and campaign data), the Yandex Direct API (campaign reports), the Metrica API (visits, goals, sources) and Google Analytics (traffic sources and conversions). All the responses go into one shared data set and are brought to a single format.
AI comments on the numbers the way a manager would
Claude gets the current week's data plus the three previous weeks for comparison. The instructions for the model require specifics with numbers: "cost per lead rose from X to Y because of this, return on ad spend dipped because of that, creative #3 is doing best". The comment must not contain vague general phrases.
Sharp changes are found automatically
Before passing the data to the model, the workflow computes a z-score: how far a metric has moved from its usual level over the last 4 weeks. This is checked for the cost per thousand impressions, click-through rate and cost per lead. If the deviation is more than two standard deviations (±2σ), the report puts that metric in a separate block, so the manager doesn't miss what needs attention.
Each client gets the report the way they like it
Some clients want a Notion page with embedded charts, some a PDF by email, and some a short summary in Telegram with a link to the full version. n8n builds the right format and sends it to the address in the client's card.
The n8n workflow connects to the ad platforms through their APIs
Runs every step of the workflow and starts it on schedule
Data on Facebook ad accounts and campaigns
Reports on campaigns and ad groups (the Reports service)
Visits, goals, traffic sources and visitor behavior
Traffic source and conversion reports from GA4
Commentary on the data and sharp changes, 3–5 paragraphs per report
Builds the report into a page with charts and attachments
Starts the run on Monday at 06:00 and retries it after a failure
What changed in numbers
a week on building reports for 8 clients
to all clients together, with no "I'll send it by lunch" delays
counting the value of the manager hours freed up
The main result: 4–6 hours a week went back to client work. Managers now spend that time on hypotheses, tests and creatives.
An unexpected bonus: the AI commentary sometimes catches sharp changes that a tired manager could miss while building a report on a Friday evening. The workflow doesn't get tired.
Where else this setup works
The general pattern: collect data from several sources, bring it into one format, have AI write a commentary and send out the report. In n8n only the data sources change, so the same setup suits:
- → Social media agencies: a client report on Instagram, TikTok, VK and Telegram
- → Marketplace sellers: a daily summary of Wildberries, Ozon, Yandex Market and Avito sent to the owner in Telegram
- → SEO agencies: rankings from Topvisor or Allpositions, Metrica and Search Console in one Notion report
- → Teams that work with ad trackers: a daily summary from Keitaro, Binom and Voluum with AI commentary
- → Sales teams: a weekly summary for the director from amoCRM or Bitrix24, the phone system and email
- The main n8n workflow with separate settings for each client: a new client can be added in 20 minutes
- Instructions for the AI that require specifics with numbers
- Finding sharp changes by z-score, which can be set up for any metric
- Reports in Notion, PDF or Telegram: each client gets the format they prefer
If your people spend hours building reports, a workflow can take that work over
This setup works wherever there are several data sources and someone who expects a regular report: marketing reports, sales reports, summaries for management, updates on factory output. A typical project goes live in 1–2 weeks.
Blog article on this topic: What AI implementation costs: the estimate for client reports and where to start
Related cases
From speech to a finished ad: transcription, translation, a new voice-over and lip-sync
Whisper transcribes the speech, GPT-4o translates the script and rewrites it for the market, ElevenLabs…
GMBSPYLAB: from the Facebook Ad Library to a Telegram feed
Async Python and Postgres 16. Every day 30–75k ads come in from the public Facebook Ad Library, and each…
End-to-end attribution for two Shopify stores: from click to order
My own service on the Python standard library receives Shopify order notifications and checks their signature…
Start with the 5,000 ₽ audit: a written report and a payback estimate
I'll tell you what to automate first, when it will pay for itself and whether your task needs AI at all. Sometimes it doesn't.
Or just ask a question. During business hours I reply within 2 hours