How to Automate Marketing Reporting and Check the Results

1. Define One Report Before You Schedule It

Choose one decision and one recurring reporting period. A weekly source review, for example, asks which recorded sources received revenue credit and which unexplained changes need investigation. It does not promise to prove which campaign caused the sales.

Write down the website, reporting timezone, closed date range, comparison, attribution model, filters, currency and revenue definition. Decide who reviews the draft and what happens if retrieval fails. A marketing report template gives that agreement a home before you connect another tool.

Keep the first version small: a headline result, a source table, one explanation and one next action. Add a measure only when someone uses it to make a decision. Selecting relevant marketing KPIs is a separate task from automating their retrieval.

2. Choose API, CLI or MCP for the Job

Use the REST API for a repeatable server-side job, the CLI for a terminal or export workflow, and MCP for a compatible AI assistant that retrieves authorized reports while you ask questions. These are different ways to access reporting, not three different kinds of revenue.

Attriboo’s CLI is available now. Its report commands use the same public report API, and browser sign-in requests report access by default. Website setup is a separate approval. Installing a terminal command does not create a recurring reporting job or schedule client delivery.

For an AI workflow, start with list_sites, then get_overview and get_breakdown for the selected website and dates. Ask the assistant to preserve the result fields before writing a narrative. Check the MCP tool reference for exact arguments; report data is returned directly within MCP text content rather than inside the REST status/data wrapper.

3. Give the Workflow Appropriate Access

Prefer a report-only OAuth connection when the workflow only needs reports. In Attriboo, analytics:read reads reports; optional sites:write adds separately approved owner/admin website setup. Do not request setup access just to prepare a weekly summary.

A workspace API key is broader: it can read analytics and use supported data-write routes. If an existing server workflow uses that key, keep it in private server configuration, restrict who can use the job, and keep it out of browser bundles, prompts, screenshots and exported reports. The public website ID in a tracking tag is not API authentication.

Owners and admins find the workspace key in Settings → API & MCP. Separately approved AI apps appear under Connected AI apps, where they can be disconnected. Regenerating the workspace key stops the old key; it does not revoke separately approved OAuth connections.

Read-only access still exposes whatever report data the connection can read. Use aggregate overview and source reports when individual sales are unnecessary. The MCP maintainers explain that tool annotations are hints, not a security boundary. Check the actual server, authorization and client behavior rather than trusting a read-only label alone.

4. Retrieve a Small Report With Explicit Scope

In the developer reference, open Overview and read its query parameters. List websites first when the intended website is unclear. Use the returned ID and its saved currency and timezone, then request the overview and source breakdown with the same selections.

For a server job, a request path can be /api/v1/stats/overview?site=site_example&from=2026-11-01&to=2026-11-07&compare=none&model=first_touch. site_example is a placeholder: replace it with the returned website ID. Your server supplies the private Authorization header. The API reference provides cURL, JavaScript and Python examples without asking you to put a secret into this article’s browser worksheet.

In a signed-in terminal, the corresponding command is attriboo reports overview --site shop.example.com --from 2026-11-01 --to 2026-11-07 --compare none --model first_touch --json. Replace the domain with your website. Use reports sources for the source breakdown and inspect its cursor fields before treating an export as complete.

Store requested selections, retrieval time and returned period metadata beside the result. Attriboo accepts inclusive website-local from/to dates, while period.from and period.to are UTC timestamps with an exclusive end. Read period.timezone rather than assuming UTC or the computer’s timezone.

A compatible assistant can use this instruction: “Read only the selected website’s overview and sources for November 1–7 with compare none and First-Touch. Return the website, currency, timezone, date boundaries, metric definitions and any coverage or continuation fields. Separate observations from possible explanations. Stop on an error or incomplete result, and do not send the report.” Replace the website and dates before using it.

5. Check Numbers Before Writing the Summary

Compare the automated overview with Dashboard using the same website, dates, model and filters. Check the raw fields before judging the prose. Google’s reporting guidance likewise identifies mismatched date ranges, dimensions, metrics, filters and reporting identity as reasons API and interface results can differ.

For an example USD response, revenue=120000 means $1,200.00 and ad_spend=30000 means $300.00. Their ratio is 4. That is recorded revenue per unit of ad spend, not a claim of $900 accounting profit. Attriboo’s net_profit field subtracts ad spend from recorded revenue; product costs, fees and other expenses still need their own treatment.

Keep missing values distinct from measured zero. ROAS is null when spend is unavailable; a comparison change can be null when no comparison data exists. A failed request is not a report of zero visitors or zero sales. Investigate an unexplained mismatch before the narrative tells someone to cut a channel.

The worksheet checks an aggregate REST overview shape and selected scope fields. It does not validate every report type or prove that an input is authentic. Keep hidden filters and provider coverage in the manual review, and do not paste customer-level sales or credentials into it.

6. Make Incomplete and Failed Runs Visible

Set a clear exception outcome: hold the current draft, identify the failed retrieval and notify the person responsible for the workflow. If you show the last successful report, display its original dates and retrieval time. Reusing old figures without that label makes a failed run look like fresh evidence.

Attriboo report history covers available records in the latest 365 days. That window does not reconstruct visits from before installation or guarantee that every imported payment matches a visitor. Provider-specific import limits, missing records and refunds remain relevant even when retrieval itself succeeds.

For row reports that return next_cursor, continue with that exact value and preserve the original website, dates, model and filters. Inspect optional coverage fields when supplied. Headline totals and a limited row page have different scopes; do not add the first 50 source rows and silently label the sum a complete website total.

The documented REST report limit is 120 requests per minute per key per server instance. Handle 429 responses using the returned rate-limit information and bounded retries. Treat authentication errors as an access problem and capacity errors as failed retrievals. Do not retry forever or turn an error into a blank success.

7. Review the Explanation Before Client Delivery

Ask the reviewer to distinguish an observation, a possible explanation and an action. “Recorded paid-source revenue declined” can be an observation. “The new creative caused the decline” needs additional evidence. A missing source tag, delayed payment or changed reporting selection can produce a different explanation.

1. Confirm the data and reporting scope, including remaining gaps.
2. Explain the largest relevant change without inventing a cause.
3. Assign one investigation or test, an owner and a review date.
4. Approve the audience and delivery separately from data access.

For budget decisions, attribution credit and incremental impact answer different questions. Use a suitable experiment when you need evidence of causal lift. Automated collection can support that work, but an AI summary cannot turn recorded source credit into an experiment.

Start with one website and two manually reviewed runs. Resolve any unexplained differences, then decide whether a scheduler saves enough work to maintain. If your report is still changing every week, a consistent spreadsheet and a named reviewer may be the more useful next step.

Common questions