Reports from your own systems, written in plain language.
Automated report generation pulls figures from the systems you already use, calculates them with fixed definitions and writes a short plain-language summary of what changed. Reports can arrive on a schedule or on request, and the numbers come from queries, not from the AI’s guesswork.
What kind of reports can be automated?
Most businesses already produce reports by hand: someone exports a spreadsheet on Monday, copies numbers into a template and writes a paragraph about the week. Report generation automates the export, the arithmetic and the first draft of the paragraph.
There are two shapes. Scheduled reports arrive at a fixed time, such as a weekly sales summary or a monthly arrears list. On-demand reports answer a question when someone asks it: “how many trial lessons did we book this month?” or “which clients haven’t paid since September?”.
Decide where reports arrive. A weekly summary can go by email, but people are more likely to read a few lines in the chat they already use, such as Slack or WhatsApp, with a link to the full figures. The format should suit the reader: an owner wants the three numbers that moved, a site manager wants their own site.
| Report | Usual sources |
|---|---|
| Weekly bookings and cancellations | Booking system or calendar, plus no-show records. |
| Cash and arrears | Accounting software such as [Xero](/integrations/xero/), invoices and payments. |
| Sales pipeline | CRM stages, deal values and last-contact dates. |
| Staff or tutor activity | Timesheets, session logs or check-ins. |
| Customer reports, such as parent updates | Session notes, attendance and progress records. |
Is report generation a job for an agent?
Tick what is true of your report generation work.
Tick the statements that apply and the suggestion appears here: an AI agent, ordinary automation, or a person with software support.
Who calculates the numbers?
This is the most important design decision. Language models are good at explaining figures and poor at adding up long columns reliably. In a well-built system, ordinary code or database queries calculate every number, and the model only writes the words around them.
Definitions matter as much as code. Agree what counts as “revenue” (invoiced or received, with or without VAT), when a customer becomes “inactive” and which week a Sunday booking belongs to. Write those definitions down once and use them everywhere.
Keep the generated text short and anchored to the figures. “Bookings fell 12% on last week, mostly at the Cork site” is useful if both numbers come from the query. Avoid asking the model to predict next month or explain causes it cannot see in the data. If you want forecasts, build them as a separate, tested calculation and label them clearly as estimates.
Numbers on demand, in the chat
In the tutoring operations agent we built for a UK business, the owner can ask for a complete report on any part of the business in WhatsApp and get it straight away, including that week’s tutor payouts. The same agent sends personalised parent reports on schedule: twelve in a week, none missed.
Both work because the agent already holds clean records: tutors check in each session and payments are matched as they arrive. Reporting is easy when the underlying data is captured properly, and hard when it is not.
How much time could report generation give back?
Put your own numbers in.
The starting values are an illustrative example, not a benchmark or a neu figure. Use your own volume, the time each one takes today and the share you expect the system to handle without a person.
Time saved is only part of the case: response times, missed work and errors matter too. The ROI guide covers how to set a baseline, and the cost guide what the build and running costs include.
Report, dashboard or both?
A dashboard suits numbers you check constantly, such as today’s bookings across sites. For Smurfs FC, the owner sees players, attendance and fees across 100+ centres on one live dashboard instead of reconciling separate returns. A written report suits a weekly or monthly review, where the useful part is what changed and why.
Many businesses need both, built on the same definitions. If your data lives in several places, a small operations platform that becomes the single record can make reporting far simpler than stitching exports together.
What can go wrong with automated reports?
- Partial data: a sync failed overnight and the report shows a bad week that did not happen. Reports should state when each source was last updated and refuse to run on stale data.
- Quiet definition changes: someone adds a new booking type and it falls outside every category.
- Overconfident commentary: the summary invents a reason for a dip. Keep explanations to what the data shows, and leave the “why” to people.
- Wrong audience: a report with site-level detail sent to every manager can show people figures they should not see. Apply the same access rules as your other systems.
- Too many reports: if nobody reads it, stop sending it.
Where to start
- Pick the one report someone builds by hand every week, and time how long it takes.
- List its sources and write down the definition of each figure.
- Automate the figures first, and check them against the manual version for a few weeks.
- Add the written summary once the numbers match.
- Then add on-demand questions, starting with the ones people ask most.
The integrations guide explains what access each source typically needs, and measuring AI agent ROI covers how to judge whether it was worth it.
Common questions
Can it report from spreadsheets?
Yes, if the spreadsheet has a consistent structure: one row per record, fixed column names and no merged cells or totals mixed into the data. Google Sheets and Excel files in Microsoft 365 can both be read through their APIs. Loosely kept sheets usually need tidying first, which is worth doing anyway.
How do we know the figures are right?
Run the automated report alongside the manual one for a few cycles and compare them line by line. Differences usually reveal a definition nobody had written down. After that, keep a small set of check figures, such as total invoices for the month, that the report must reconcile to before it is sent.
Who can ask for on-demand reports?
Whoever you allow. Access should follow roles: a manager might see their own site, while finance figures stay with the owner. The agent should check who is asking before answering, the same way your existing systems restrict who sees what.
Can reports be sent to customers?
Yes, such as parent progress updates or monthly client summaries. Customer-facing reports deserve more care: use approved templates, check personal data goes only to the right person, and consider a review step until the content is consistently accurate. Our tutoring agent’s parent reports are a working example.
Bring us the task.
Tell us what happens today, which tools are involved and where it goes wrong. On a free 30-minute review, we’ll work out whether an agent, an automation or neither is the right fix, and what a sensible first release would be.
Start a project.

Book a call or send us a message. We read everything.
tash@neu.ie
Working with teams in Ireland and worldwide.
Free 30-minute AI review.