Report automation: four ways to build a report without hands, and how to pick yours

Closing the month looks roughly the same everywhere. Accounting exports the bank statement, a sales manager sends over their own spreadsheet, marketing pastes numbers from the ad dashboard, and one person stitches it all together by hand. By the third round you discover that one file uses 01/03, another 2026-03-01, and the third has a column called "total" that quietly includes tax. Report automation starts right there, not with picking a tool.

Below: what it actually consists of, the four ways to do it, and how to tell which one fits you.

Heads up. If your exports contain names, phone numbers or salaries, read the section on data first.

Contents:

• What report automation is, and what gets mistaken for it
• Four ways to do it
• How to choose
• The four ways in detail
• How to check the numbers
• What you should not send out
• What automation will not do
• FAQ
• The short version

What report automation is, and what gets mistaken for it

Report automation means the report builds itself and arrives without a human at every step. It has three parts, and most of the confusion comes from automating one of them while expecting results from all three.

First, the source: where the numbers come from. A CRM export, a bank statement, a spreadsheet someone fills in by hand. Second, assembly: how those pieces become a total. Third, delivery: where the finished report lands and when.

Now the common mistake. Dropping a spreadsheet into a chat with an AI model and asking it to total things up is not automation. It speeds up manual work, and only the middle part. You still pull the export yourself, you still move the result yourself, and tomorrow it all repeats. Useful, sure. But if a month later the report still eats half your day, you automated the wrong thing.

Automation is when the summary is already sitting in your chat on Monday morning and all you have to do is read it. The difference is not the quality of the tool, it is who starts the process.

Four ways to do it

There really are four, and they differ less in what they can do than in what you have to put in up front.

"Runs itself" means nobody has to launch anything: the report shows up on a schedule or on an event. "Breaks when" is not about reliability, it is the situation where the approach stops fitting.

How to choose

Check each of these before you configure anything. Half of failed rollouts die on one of these questions rather than on the technology.

That last one stings, but check it first. A report that takes forty minutes once a month is not worth automating.

The four ways in detail

Formulas and pivot tables

The cheapest entry. Formulas calculate, pivots group, Power Query in Excel or an import in Google Sheets pulls data from an external file. Hit refresh and the numbers update.

This works while there are few sources and their structure holds still.

Then a fourth department shows up with its own spreadsheet. Different date formats, columns named their own way, someone inserted a row in the middle and broke a range. From then on reconciliation costs you more than the calculation ever saved. We covered the mechanics separately in our guide to automating Google Sheets.

BI platform

Power BI, Tableau, Looker and the like. You connect the sources once, the platform refreshes them and draws the dashboard. This is a genuinely automatic report, with history and access control.

The price is setup. Connecting sources, describing the data model, building the views: that is weeks of work, and usually not yours. BI also lives well where the data is already in a database. If half your numbers are in email threads and in a manager's head, BI will not save you.

Who it fits: mid-size and large companies with real bookkeeping and a budget for implementation. Though I would not be dogmatic here. I have seen small companies where BI took root, simply because somebody there found it interesting.

Off-the-shelf AI agent

Different principle. The agent does not build a data model, it does what a person would: goes into the sources, picks out what it needs, calculates, and writes the summary in plain words. What separates it from a chat window is that it starts itself and fetches the data without you.

The ASCN marketplace has several of these. The KPI report agent pulls metrics from your spreadsheets, Search Console and GitHub and sends a weekly summary. The CFO agent builds a monthly financial package and a cash runway forecast. The agency agent prepares reports for every client at once.

The honest limit: an off-the-shelf agent is built for a standard report. Your own format with its own calculation logic means modifying it, which is the next option.

A note on integrations, since this is what people ask about most. There is no one-click "connect your ERP" button. The connection goes through an API or an MCP server, or the data gets exported into a spreadsheet and the agent works with that. The second option looks crude and in practice gets running in an evening and then works for years.

Your own AI agent

For when the report is not standard. You describe the task in words: where to pull from, how to calculate, what counts as an anomaly, who gets it and when. Connect the sources, test against historical data, switch it on.

No programming needed, but you do need to state clearly how your report works. That usually turns out to be the hard part. People calculate things by feel for years and write the rule down for the first time only when explaining it to a machine.

And here something tends to surface that nobody planned for. As you sit down to write the rule, you find that sales counts revenue on shipment while finance counts it on payment, and both sides spent years assuming they counted the same way and that the numbers just "drift a bit from rounding". At that point the automation conversation turns into an accounting conversation and can take longer than the setup itself. That is not a flaw in the method. But budget time for it, or the rollout stalls on flat ground.

How to check the numbers

You do not need to recalculate the whole report by hand, that defeats the point. A few spot checks are enough, and they are worth doing for at least the first month.

• Match the total against a number you already know. Monthly revenue should agree with the bank statement.
• Check the period boundaries. A March report that pulls the first through the thirtieth looks correct right up until somebody asks where the thirty-first went. Or the other way round, the first of April creeps in and the month runs a day long. A stupid error, a persistent one, and the most common thing on this list.
• Look at what happens to empty fields. A deal with no amount should land in the report as a zero or as its own line, not vanish.
• Run a month you have already closed by hand.
• Cut access to one of the sources and see what happens. The system should complain. If it assembles a report from whatever is left and says nothing, that is a landmine.

The last one is the nastiest, because a report like that looks fine. It arrives on time, the numbers are plausible, and you may not notice for a quarter, until somebody happens to reconcile it against the bank.

What you should not send out

Reports almost always contain things that do not belong on someone else's servers: customer names, phone numbers, salaries, unit costs, supplier terms.

If your data covers people in the EU or the UK, uploading a file with names and phone numbers to an external service is a transfer of personal data and falls under GDPR, with the usual questions about lawful basis and where processing happens. Other countries have their own rules, and several require certain categories of data to stay within national borders. Before you connect an external service, check where its servers physically sit and whether it uses uploaded content to train models.

Trade secrets are a separate matter. Your report may contain no personal data at all and still spell out your price structure and supplier terms.

The good news is that you usually do not need to hand over the real data in the first place.

A machine does not need real surnames to debug a report. Swapping them for fakes does not change the shape of the task.

What automation will not do

Worth being blunt here, because expectations usually run high.

• It will not clean up your sources. If half the deals in your CRM have no amount, the automated report will be just as incomplete, only faster.
• It will not explain why a number is what it is. It will show you the dip in March. That it was a large shipment moved to April rather than a system fault is something only you know.
• It will not take responsibility. A person signs the report, and a person checks it.
• It will not replace management accounting. If you do not have it, automation will not invent it for you.
• It will not survive unattended. Change the export structure in your CRM and Monday's report arrives empty. Somebody has to notice.

One more observation, not a technical one. Resistance usually comes not from people being unwilling to learn something new, but from an automatic report showing numbers that used to be presentable with a little framing.

FAQ

Which report should I start with?

The one that eats the most time and runs most often. Usually a weekly sales summary or a monthly management pack. Starting with an annual report is a bad idea: you see the result once a year and never get to tune anything.

How long does setup take?

A spreadsheet with formulas, a few hours. An off-the-shelf agent, an evening, assuming your sources are in order. Your own agent, several days, most of which goes into agreeing internally how the metrics are actually calculated. A BI platform, weeks, usually with an integrator.

Is a chat assistant not simpler?

Simpler, but it solves a different problem. A chat is good for pulling apart one file once. For a recurring report it does not work: you export the data yourself every time, explain the structure again, and move the result by hand.

Can I connect my ERP or CRM?

Yes, though not with a button. The connection runs through an API or MCP, or through an intermediate export into a spreadsheet. Which one fits depends on how your system is deployed and who has access to it.

What if the data lives in people's heads?

Then first give it somewhere to live. An ordinary spreadsheet with clear rules for filling it in will do. You cannot automate what does not exist in digital form, and this is probably the most common reason rollouts end in nothing.

What does it cost?

Formulas are free if you already have Excel or Google Sheets. An off-the-shelf ASCN agent comes with a subscription from $29 a month, which also includes several other agents. A BI platform means licences plus implementation, and that is where the numbers get serious.

The short version

• Report automation has three parts: source, assembly, delivery. Speeding up assembly alone means still losing half a day to the report a month from now.
• Pick the approach by how many sources you have and who will fix it, not by a feature list.
• Reconcile against the manual version for the first month. Above all, check that a broken source makes the system complain rather than quietly undercount.
• Strip personal data and trade secrets before sending anything to an external service. For setup, the real data is almost never needed.
• Without management accounting in place, automation will not substitute for it. Sources first, everything else after.

If you want to see how this looks in practice, start with the KPI report agent: you can run it on your own data and drop it if it does not fit. For what else in office routine comes off the same way, see our overview of AI tools for office automation. If your report is a non-standard one, message a manager and they will tell you whether it can be built at all.