Use cases
How to automate reports with AI: daily and weekly summaries
To automate a report with AI, a workflow runs on a schedule, fetches the numbers from your tools, asks a model to explain what changed, and posts the summary where your team reads it. When the report needs judgment about where to look next, an agent does the fetching and the digging instead.
Updated · 4 min read
Which reports can AI write for you?
- A Monday morning traffic summary from Google Analytics: visits, top pages, what rose and what fell.
- A daily support digest from Zendesk: new tickets, the biggest themes, anything waiting too long.
- A weekly engineering update from Jira or Linear: what shipped, what is stuck.
- A sales pipeline note from your CRM or a Google Sheets tracker.
- A business summary straight from your own database, with a read-only query.
The model’s job is the part people skip: reading the numbers, comparing them with last time and saying in two sentences what matters.
Should a report be a workflow or an AI agent?
| Workflow with an Ask AI step | Workflow that runs an agent | |
|---|---|---|
| The steps | Always the same: fetch, summarize, post | Decided by the agent as it reads |
| Cost per run | One model call | Several model calls |
| Good for | Fixed dashboards and digests | Explaining surprises, following a thread |
| Predictability | High | Lower: watch the first runs |
The steps
- Workflow with an Ask AI step
- Always the same: fetch, summarize, post
- Workflow that runs an agent
- Decided by the agent as it reads
Cost per run
- Workflow with an Ask AI step
- One model call
- Workflow that runs an agent
- Several model calls
Good for
- Workflow with an Ask AI step
- Fixed dashboards and digests
- Workflow that runs an agent
- Explaining surprises, following a thread
Predictability
- Workflow with an Ask AI step
- High
- Workflow that runs an agent
- Lower: watch the first runs
Start with a workflow. Switch the summary step to an agent when you find yourself wishing the report would look into why a number moved. See AI agents vs workflows.
How do you build a weekly AI report, step by step?
In MyDeepAgents, a weekly traffic report posted to Slack:
- Connect Google Analytics by signing in with Google, and Slack with an incoming webhook for the channel.
- Create a workflow with a schedule trigger: Mondays at 8:00 in your time zone.
- Add an action step: Google Analytics’ Run a report, for the last 7 days and the 7 days before.
- Add an Ask AI step. Choose a model, and write a prompt that includes the report’s output: what to compare, what counts as a big change, how long the summary should be.
- Add an action step: Slack’s Post a message, with the Ask AI step’s text.
- Run it once by hand, read the message, and tune the prompt.
Post to Microsoft Teams or send it with Gmail or Outlook instead, with the same steps.
How do you make sure the AI summary is accurate?
- Give the model the numbers, not the job of finding them. The action step fetches the data; the model only explains it.
- Ask for structure. The Ask AI step can return JSON that must match a schema, so a missing field fails loudly instead of quietly.
- Keep the raw numbers in the message, next to the explanation, so readers can check it.
- Name a fallback model: up to three, tried in order if the first is unavailable.
- Redact personal data before it reaches the model when the report draws on customer records.
What should you know before you schedule a report?
If the platform is down at the scheduled moment, the schedule can catch up with one run afterwards, or skip it. A run that fails is in the run history with the step that failed and why.
Key points
- A scheduled workflow fetches the numbers, an Ask AI step explains them, and an action posts the summary.
- Use an agent instead when the report should dig into why something changed.
- Schedules take any cron expression in your own time zone.
- Accuracy comes from fetched data, structured output and the raw numbers beside the summary.
- Slack, Microsoft Teams, Gmail and Outlook can all deliver the report.
Questions people also ask
Can AI write a weekly report from Google Analytics?
Yes. A workflow runs Google Analytics’ report for the week, an Ask AI step compares it with the week before, and the summary is posted to Slack or sent by email.
How much does an AI-written report cost to run?
With a workflow, one model call per report, billed by your model provider on your own key. An agent makes several calls, so it costs more.
Can the report read from my own database?
Yes. A database connection’s query action runs inside a read-only transaction, so a report step can query PostgreSQL, MySQL, SQL Server and others without being able to change anything.
Can I run the report on demand as well as on a schedule?
Yes. A workflow can be run by hand from the console or started through the API.