Key takeaways
- Start with 8 to 12 numbers you would act on, not everything you can track.
- Write down exactly how each number is defined.
- Automate the pulling with code or connectors; let AI write the words.
- Check totals against the source. AI can misread numbers.
- Most small businesses need a weekly one-page report more than a live dashboard.
On this page
Pick the 8 to 12 numbers that matter
The most common reporting mistake is tracking everything. A dashboard with 60 charts looks impressive for a week, then nobody opens it. Start small.
Here's the test for each number: if it moved 20 percent this week, would you do anything differently? If not, leave it out.
A typical small service business might land on something like this:
- Leads: new leads this week, by source
- Speed: average time to first response
- Sales: quotes sent, quotes won, close rate
- Revenue: invoiced this week, collected this week
- Money owed: total past due, and how much is over 30 days
- Marketing: ad spend and cost per lead
- Phone: calls answered vs. missed
- Reputation: new reviews and average rating
That's ten. Your list will look different, and that's the point. A retail shop cares about average sale and foot traffic. A dental office cares about new-patient appointments booked. Pick yours, write them down, and resist adding more until the first set is working.
Connect your sources
Most of the numbers above already exist in software you pay for. The job is getting them out automatically, every week, without someone copying them into a spreadsheet.
| Source | Typical numbers | How it usually gets pulled |
|---|---|---|
| Ad accounts | Spend, clicks, leads, cost per lead | Built-in connections or reporting tools |
| Website analytics | Visits, form fills, calls from the site | Direct connection to the analytics account |
| CRM | New leads, stage changes, close rate | Direct connection, or a scheduled export |
| Accounting software | Invoiced, collected, past due | Direct connection, read-only login |
| Phone system | Calls, missed calls, call length | Direct connection or weekly export |
| Reviews | New reviews, rating | Review platform connection or a quick check |
When I set this up as a fractional AI manager, I build these pulls with workflow tools like n8n and store the results in a simple database, so each week's numbers are saved and comparable. Some older software has no good way to connect. In that case, a two-minute weekly export dropped into a shared folder works fine. Reporting doesn't have to be fully automatic to be worth it.
Use read-only access everywhere. A reporting system never needs to change anything in your accounts.
Let AI write the summary, with "watch" and "win" callouts
This is where AI earns its keep. Raw numbers don't tell an owner much on a Friday afternoon. A short paragraph does.
The trick is the order of operations. The numbers get calculated first, by code or a spreadsheet, not by the AI. Then the AI gets the finished table plus last week's and last month's numbers, and writes a summary. A made-up example:
WIN: 14 new leads this week, up from 9. Most of the jump came
from Google Business Profile calls after the new photos went up.
WATCH: Past-due invoices over 30 days rose to $6,200. Three
customers account for most of it. Worth a personal call.
WATCH: Average first response time slipped to 4 hours on
Wednesday and Thursday, when the office was short-staffed.
Two or three wins, two or three things to watch, and a line about what to do. That's it. If you want to see the full format, here's a sample Friday report.
See the sample report
Dashboard or weekly one-page report?
People ask for dashboards because that's what reporting looks like in their head. But a dashboard only helps if someone looks at it. A report that lands in your inbox every Friday gets read because it comes to you.
| Weekly one-page report | Live dashboard | |
|---|---|---|
| Best for | Owners who want the story | Managers who watch numbers daily |
| How you use it | Read it Friday, act Monday | Check it whenever you want |
| Explains what changed | Yes, in plain English | Not unless you build that in |
| Effort to keep useful | Low | Higher; dashboards drift |
| Good first step? | Yes | Usually second |
My suggestion for most small businesses: start with the weekly report. Once you find yourself wanting to check a number mid-week, add a simple dashboard for those few numbers. Plenty of businesses never need the dashboard at all. Both are part of my AI reporting and dashboards work.
Pitfalls to watch for
Bad source data. If your CRM has duplicate contacts, leads with no source, or deals that never get closed out, the report will faithfully report garbage. Expect to spend the first few weeks fixing data habits, not building charts.
Fuzzy definitions. What counts as a lead? Is a spam form fill a lead? Is revenue what you invoiced or what you collected? Your ad platform and your website analytics will count conversions differently and never match exactly. Write a one-line definition for every number and keep it at the bottom of the report.
AI misreading numbers. Language models can swap columns, confuse a percent with a percentage point, use the wrong date range, or simply do arithmetic wrong. That's why the math happens in code and the AI only writes words about finished numbers. Even then, check the totals against the source for the first month, then spot-check a few numbers every week. I cover this more in what happens when AI makes a mistake.
Small numbers swing. If you get eight leads a week, going to six isn't a trend. Tell the AI to compare against a four-week average, not just last week.
What this looks like in a real small business
Example: picture a ten-person remodeling contractor. The owner checks five different apps when he remembers to, and mostly runs on gut feel. He doesn't know which marketing actually brings jobs.
He picks nine numbers: leads by source, estimates booked, estimates sent, jobs won, revenue invoiced, revenue collected, past due over 30 days, ad spend and missed calls. He writes one line defining each. A lead is a real person asking about a real project, not spam or a vendor.
Read-only connections pull from the ad accounts, website analytics, CRM, accounting software and phone system every Friday morning. Code calculates the numbers and compares them to the four-week average. AI writes the summary with wins and watches. Before it's sent, a check compares totals against each source and flags anything that doesn't match.
For the first month, the office manager spot-checks every number. After that, a few per week. The owner reads it Friday over lunch and makes two or three decisions Monday morning.
The honest part
If your data lives mostly in your head, on paper, or in a spreadsheet nobody updates, AI reporting won't help yet. Fix the data entry habit first.
If you won't read the report, don't build it. The value comes from the decisions it prompts, not from having it.
AI summaries also carry a quiet risk: they make bad data sound authoritative. A well-written paragraph about wrong numbers is worse than a messy spreadsheet, because people believe it. That's why the source checks matter more than the writing.
Very small businesses may not have enough volume for weekly trends to mean anything. A monthly report might fit better.
And AI can't tell you why customers chose you. It can show patterns. You still have to ask people.
Quick questions
What tools do I need for AI reporting?
Usually nothing new. The numbers already live in your ad accounts, analytics, CRM, accounting software and phone system. You need a way to pull them on a schedule (a workflow tool or built-in connectors), a place to store each week's results, and an AI model to write the summary.
Can AI predict next month's sales?
It can project a trend from your history, but treat that as a rough guide, not a forecast you'd bet payroll on. Small businesses have too few data points and too many one-off events for reliable predictions. Use it to start a conversation, and pair it with what you know about upcoming work.
How often should the report run?
Weekly works for most owners, sent Friday so you can act Monday. Daily reports tend to create noise and get ignored. Monthly works if your volume is low. Whatever the schedule, keep the same numbers and definitions so each report compares cleanly with the last.
Related answers
