Automated client reporting for agencies: the n8n build

A weekly client report built by hand costs the same every week, for every client, for as long as the retainer lasts. Automated client reporting turns that recurring cost into a one-off build. The build is not hard. What catches agencies out is the part nobody draws on the diagram: what the pipeline does on the morning something breaks.

The short version

• ▸A reporting pipeline has four layers: pull the data, store it, render it, deliver it.
• ▸n8n orchestrates the run; Google Sheets or BigQuery holds the numbers; Looker Studio draws the report.
• ▸Most failures produce no error. The run goes green and a client gets a wrong report, or none.
• ▸Design for timezones, partial data, rate limits and expiring credentials before the first send.

Manual reporting lasts because each instance is small. Someone logs into Google Ads, exports a range, pastes it into a sheet, fixes the formatting, writes two lines of commentary and sends it. No single Friday feels like a problem, and the cost only shows when you add it up across clients and weeks. By then the report is part of someone's job description, not a task anyone questions.

This guide shows the architecture, the build in n8n node by node, and the failure modes that decide whether you can trust the system without checking every output by hand.

What automated client reporting actually involves

Automated client reporting is a scheduled pipeline. It pulls each client's performance data from the ad and analytics platforms, writes it to one store, renders it into a report template and delivers it. Every client it could not serve gets a log row and an alert. The schedule replaces the person. The log and the alert replace their judgement.

That last part separates automation from a liability. A person who copies numbers by hand notices when Meta returns nothing. A workflow does not, unless you tell it what "nothing" looks like.

To model whether the build pays off for your agency, time one report end to end, multiply by your client count and by the weeks in a year. Our guide on connecting your CRM to n8n for automated reports covers the CRM half of the same data, if pipeline and deal numbers belong in the report too.

The four layers of the pipeline

Every automated client reporting build separates into the same layers, however simple. Keeping them separate is what lets you change one without breaking the others.

The orchestration across all four is the workflow tool. This guide uses n8n. Zapier and Make can run simpler versions of the same design, and our comparison of n8n, Zapier and Make for agencies covers where each one stops being the right fit.

Building it in n8n, node by node

The schedule and the client list

The whole run, in order:

1. Schedule Trigger fires and reads the client config sheet.
2. Loop Over Items takes the active clients a few at a time.
3. One API call per platform fetches each client's numbers.
4. A Code node normalises them, checks them and writes one row.
5. A delivery node sends the report, and a final node logs the outcome.

Start with a Schedule Trigger and a master config sheet with one row per client: account IDs for each platform, recipients, the preferred format and whether the client is active. The workflow reads the sheet first, so adding a client is a new row, not a change to the workflow.

The Schedule Trigger uses the workflow's own timezone if one is set, otherwise the instance timezone, which defaults to America/New York on a self-hosted install. Set the workflow timezone explicitly. Then compute the reporting date range once, at the top of the run, and pass it down as data rather than recalculating it in every node.

Looping over clients

Use Loop Over Items (the node formerly called Split in Batches) to process clients in small groups. It returns a set number of items per pass, which gives you two things at once. You get a natural place to pause between batches, and isolation, so one client's failure does not decide what happens to the next.

Set the API nodes to continue on error and route that error output to a log row. Then attach an error workflow in Workflow Settings for failures outside the loop. An error workflow must start with the Error Trigger node.

Fetching the data

Each client gets one call per platform. GA4 goes through the Google Analytics Data API, Google Ads through its own API, Meta through the Marketing API. Store credentials per platform rather than per client: an agency's Google Ads manager account can read every account it manages.

Quotas decide how big your batches can be. A standard GA4 property gets 200,000 core tokens per day, 40,000 per hour and 10 concurrent requests. A Google Ads developer token with Basic access is allowed 15,000 API operations per day. Meta applies Ads Insights rate limits per app, so once your app is throttled, every Insights call it makes is throttled, whichever client it is for.

Normalising and storing

A Code node maps each platform's response onto one schema: spend, impressions, clicks, conversions and revenue. It also computes the derived figures you report on, such as week-over-week change and blended return on ad spend. Write one row per client per period to the client's sheet, or to a BigQuery table if the volume justifies it, with the period's start date as a column.

Before the row is written, check that every required metric is present. A missing value should stop that client's report and raise an alert. Writing a zero instead is how a client ends up reading that their spend was nothing this week.

Delivering the report

Branch on the client's preferred format. Email goes through the Gmail or SMTP node, with the dashboard link and a short written summary in the body. Slack goes through the Slack node to the client's shared channel. If you add an AI-written summary, generate it from the stored row, never from the raw API responses, so the prose and the numbers come from the same source.

The last node writes the outcome to a log sheet: client, period, status, and the error if there was one.

Where Looker Studio fits, and where it stops

Looker Studio is the rendering layer most agencies already know, and it reads straight from Google Sheets or BigQuery. Build one master template, copy it per client and repoint the data source. Set the date control to a rolling range so the report is current whenever the link is opened.

Know one limit before you promise a client a PDF. Looker Studio's own scheduled email delivery sends the report as a PDF attachment, at most once a day without Pro, to no more than 50 recipient addresses. It runs on its own schedule, outside n8n, so your workflow cannot hold it back when the data is incomplete.

That gives you two honest options for automated reporting for clients who want a file:

1. Let n8n send the dashboard link and summary, and let Looker Studio's scheduled delivery send the PDF later, after the data has landed.
2. Render the PDF yourself with a headless browser on a small server, triggered by n8n after the checks pass.

The first is simpler. The second is the only one where a failed data check can stop the PDF from going out.

How to roll it out one client at a time

Roll out automated client reporting for one client first. Run the workflow by hand, compare every figure against the platform dashboards for the same date range, and only then add the next client. Agencies that move every client at once spend the first month debugging edge cases they could have met one at a time.

A sequence that works:

1. Create the config sheet and the credentials, and copy the Looker Studio template once.
2. Build the workflow for a single, well-tracked client and run it manually.
3. Check the numbers against each platform for the same date range and timezone.
4. Add the remaining clients to the config sheet and run the whole set by hand.
5. Turn on the schedule, and read the log after the first unattended run.

Send the first automated reports alongside a short note to each client explaining the new format. A report that changes without warning reads as a mistake, even when it is better.

Where EsperaStudio fits

We build automated client reporting pipelines like this one as part of a longer engagement, not as a one-off export script. Each client's numbers flow through a workflow that answers for itself: logged, alerted and checked before anything leaves.

We run every system on infrastructure the client owns, hosted inside the EU, so their operational data never sits inside a third party's workflow tool and nothing is billed per task.

We start every engagement with a €500 automation audit, delivered as a written diagnostic rather than a sales call, and we credit the fee toward the build if the client goes ahead. For reporting, that audit maps every manual step in your current process and tells you which ones are worth automating first.

If your team would rather build this in-house, the honest comparison of running n8n yourself covers what that takes.

Common mistakes to avoid

Automating messy tracking. If campaigns lack UTM parameters or the Meta pixel misfires, the automated report repeats those errors every week, on time. Fix the tracking before you build on top of it.

Treating "the workflow finished" as "the work was done". In automated client reporting, a green run says nothing about whether each client got a complete report. Count delivered reports against active clients in the log after every run.

Leaving the Google Cloud app in Testing. A Google OAuth app with a publishing status of Testing is issued refresh tokens that expire in 7 days. The pipeline works for a week, then every Google node fails at once. Move the consent screen to production before you depend on it.

Firing every call at once. Requests sent for every client at the same minute run into each platform's limits together. Depending on the node, you get an error, a retry or a shorter result that looks complete. Batch the loop, add a Wait node between batches and size the batch to the strictest platform.

Removing the human layer entirely. The pipeline should deliver the numbers. Someone on your team should still read them and add the line of judgement a client is paying for. An AI summary can draft that line; it should not be the only reader.

Frequently asked questions

What are the best automated reporting tools for an agency?

For most agencies the stack has three parts: a workflow tool to orchestrate, a spreadsheet or warehouse to store, and a dashboard tool to render. For a small team, n8n, Google Sheets and Looker Studio cover all three, and BigQuery takes over from Sheets when the volume grows. The tool matters less than the checks: logging, alerts and a rule that incomplete data is never sent.

How do you automate Google Analytics reports for clients?

To automate Google Analytics reports, call the GA4 Data API from your workflow for each client's property, write the results to a store, and point a Looker Studio template at that store. Batch the requests, because each property has a daily and hourly token quota and a cap on concurrent requests, and compute the date range once so every platform reports the same week.

Can Looker Studio send automated marketing reports by email?

Yes. Looker Studio can email a report as a PDF on a schedule, and the PDF covers every page you select. Without a Pro subscription it sends at most once a day, and a report can have one schedule. It runs independently of your workflow, so it cannot be told to wait when the underlying data is incomplete.

Is report automation worth it for an agency with only a few clients?

Often, yes. Automated client reporting has a build cost that is mostly fixed, while the manual cost repeats every week. Time one report end to end and multiply it by your clients and the weeks in a year. If the result is more than the build and a year of upkeep, automate. If your reports change shape every month, fix the format first.

Sources

• ▸Using OAuth 2.0 to Access Google APIs — Google for Developers, accessed 2026-10-09
• ▸Data API limits and quotas — Google for Developers, accessed 2026-10-09
• ▸Google Ads API quotas and limits — Google for Developers, accessed 2026-10-09
• ▸Marketing API rate limiting — Meta for Developers, accessed 2026-10-09
• ▸Schedule automatic report delivery — Google Cloud, accessed 2026-10-09
• ▸Schedule Trigger — n8n Docs, accessed 2026-10-09
• ▸Loop Over Items (Split in Batches) — n8n Docs, accessed 2026-10-09
• ▸Handle errors gracefully — n8n Docs, accessed 2026-10-09