Example dashboard · invented company

A weekly KPI dashboard, with the checks first

AI agents build the dashboard from your billing, CRM and product data, and it arrives each week as a simple web page. It opens with what was loaded and which checks fired, then the week’s numbers against last week and the 8 weeks before, then the detail behind them. Nothing is installed on your systems. Building it is a sprint, from €12k, fixed before work starts; the weekly refresh is repeat work, on an async retainer from €8k a month. Below is one week’s dashboard for an invented subscription company.

The example · invented company

Dashboard, week to Sunday 27 September 2026

The agents ran it on Monday 28 September at 07:00. The company is the one in the revenue forecast demo and the monthly revenue report, in the same simulated run, so at each month-end its recurring revenue is the report’s to the euro. Every figure is computed when this page is built. To pick another week or switch the planted faults off, run the weekly dashboard demo.

What we pulled

Three of four sources hold data to Sunday 27 September. Product events stop at Thursday 24 September: the last load ran on Friday 25 September at 06:00, and nothing has loaded since. Each source loads at 06:00 with the day before.

Sources, week to 27 September (invented company)
Source	State	Data to	Feeds	Rows this week	What a row is
Billing	Fresh	Sun 27 Sep	Recurring revenue, net new, both churn rates, plans	103	subscription changes
CRM	Fresh	Sun 27 Sep	Trial conversion	132	trials that ended
Product events	Stale	Thu 24 Sep	Sign-ups	11,607	sign-ups and logins
Web analytics	Fresh	Sun 27 Sep	The sign-up cross-check	7,841	sessions

Sign-ups, the one figure that reads product events, count Monday to Thursday this week, and their tile compares the same four days of every earlier week. The other five figures read billing and the CRM, both loaded to Sunday. Web analytics holds the whole week, but it misses the visitors who block tracking, so it cross-checks the sign-up count and does not replace it.

What is broken

One of four checks fired. Scripts run the checks before any figure is printed.

Checks, week to 27 September (invented company)
Check	Result	This week	What it tests
Totals reconcile	Passed	30 new customers in billing and 30 paid trials in the CRM; 86 sign-ups in product events and 79 in web analytics	Billing's new customers, after the duplicate match, are the trials the CRM marks as paid. Product events and web analytics count sign-ups within 10% of each other on the days both hold.
No duplicate customers	Fired	2 of the 6 returning customers came back under a new customer ID	Each customer has one customer ID in billing, matched by email.
The bridge adds up	Passed	€342,498 + €2,024 − €2,591 = €341,931	Last Sunday's recurring revenue, plus what new and returning customers and upgrades added, less what cancellations and downgrades took, is this Sunday's.
No plan with a negative count	Passed	Lowest at any day's end: Scale, 468	Every plan holds zero customers or more at the end of every day.

What it means. Billing gave two of this week’s six returning customers a new customer ID, so billing on its own reads them as new customers. The check matches each new ID to an old one by email, and on this dashboard they count as returning. Unmatched, the week would show 32 new customers instead of 30, and 4 returning instead of 6. Recurring revenue is the same either way: new and returning customers both join Starter at €29. Merging each pair of IDs in billing stops the check firing for them.

The numbers that matter

Recurring revenue

€341,931

At Sunday's close, down €567 on last week. Previous 8 weeks: €339,007 to €342,498. Invented company.

Net new recurring revenue

−€567

€2,024 added, €2,591 lost. Last week +€29. Previous 8 weeks: −€79 to +€1,346. Below that range. Invented company.

Customer churn

1.1%

49 of the 4,492 customers paying on Monday left. Last week 0.7%. Previous 8 weeks: 0.7% to 0.9%. Above that range. Invented company.

Revenue churn

0.8%

€2,591 lost to cancellations and downgrades, of €342,498 at the week's start. Last week 0.7%. Previous 8 weeks: 0.5% to 0.8%. Invented company.

Sign-ups, Monday to Thursday

Product events stop at Thursday, so every week counts the same four days. Last week 91. Previous 8 weeks: 77 to 100. Invented company.

Trial conversion

22.7%

30 of the 132 trials that ended paid. Last week 23.5%. Previous 8 weeks: 19.9% to 25.8%. Invented company.

The detail behind it

Recurring revenue at each Sunday’s close, 5 July to 27 September 2026

To last SundayThis weekShade: the range of the 8 weeks before this one

€336k€338k€340k€342k€344k

5 Jul2 Aug30 Aug

Invented company, one simulated run, in whole euros.

The 13 weeks as a table, with every KPI

KPIs week by week, to 27 September (invented company)
Week to	Recurring revenue	Net new	Customer churn	Revenue churn	Sign-ups, Monday to Thursday	Trial conversion
Sun 5 Jul	€336,163	−€127	0.9%	0.8%	90	28.3%
Sun 12 Jul	€337,713	+€1,550	0.5%	0.3%	71	29.2%
Sun 19 Jul	€338,403	+€690	0.8%	0.5%	80	28.1%
Sun 26 Jul	€339,086	+€683	0.7%	0.5%	87	23.5%
Sun 2 Aug	€339,007	−€79	0.7%	0.6%	100	25.8%
Sun 9 Aug	€339,566	+€559	0.9%	0.7%	79	22.6%
Sun 16 Aug	€340,126	+€560	0.8%	0.8%	90	25.5%
Sun 23 Aug	€340,328	+€202	0.8%	0.6%	86	21.8%
Sun 30 Aug	€340,489	+€161	0.9%	0.7%	84	19.9%
Sun 6 Sep	€341,123	+€634	0.8%	0.6%	77	22.4%
Sun 13 Sep	€342,469	+€1,346	0.8%	0.5%	82	21.7%
Sun 20 Sep	€342,498	+€29	0.7%	0.7%	91	23.5%
Sun 27 Sep	€341,931	−€567	1.1%	0.8%	86	22.7%

Paying customers by plan, week to 27 September (invented company)
Plan	Sun 20 Sep	Joined	Left	Moved, net	Sun 27 Sep	Recurring revenue
Starter (€29)	2,803	36	41	−11	2,787	€80,823
Growth (€99)	1,219	0	7	+12	1,224	€121,176
Scale (€299)	470	0	1	−1	468	€139,932
All plans	4,492	36	49	0	4,479	€341,931

What changed.

Recurring revenue fell €567 to €341,931. Cancellations took €2,181 and downgrades €410, against €2,024 from 30 new customers, 6 returning and 14 upgrades.
49 customers left, 1.1% of those paying on Monday, above the range of the previous 8 weeks. 41 were on Starter, 7 on Growth and 1 on Scale; the 8 on Growth and Scale took €992 a month of the €2,181 lost to cancellations.
Trial conversion fell to 22.7%, from 23.5% last week: 30 of the 132 trials that ended paid. It stays inside its 8-week range, whose low is 19.9%.
Sign-ups were 86 from Monday to Thursday, fewer than the 91 on the same days last week and inside their range. The days after Thursday are missing until product events load again.

Checks and sign-off

Sign-off. This dashboard is a routine refresh: the same page every week, on the week’s data. It carries its checks and no signature. What the analytics director signs is what you act on, such as a decision memo drawn from it.

Method note. Which checks ran on this dashboard:

The company and its customers are invented: the revenue forecast demo's model in the same simulated run as the monthly revenue report (run 1,553 of seed 20261013), its months spread over their days with a second seed (20261086). No client's data.
At every month-end from September 2025 to September 2026, the dashboard's recurring revenue is the monthly report's to the euro, and September's revenue bridge, summed over its days, is the report's, line by line.
In each of the 13 weeks, last Sunday's recurring revenue plus the week's bridge lines is this Sunday's; customers at the start, plus joiners, less leavers, are the customers at the end, plan by plan; and the plans add up to the totals.
No plan falls below zero customers on any day from 1 September 2025 to the end of September 2026.
Customer churn is customers who left over customers paying at the start of the week. Revenue churn is revenue lost to cancellations and downgrades over revenue at the start of the week. Trial conversion is trials that paid over trials that ended in the week.
Planted, to show what the dashboard does with them: the load times, the stale product-events load and the two returning customers under a new ID. Product events' and web analytics' row counts come from invented rates, and web analytics misses 6% of sign-ups; the cross-check accepts a gap of up to 10%.
Each range flag on the tiles and each direction word under "What changed" is checked against the numbers it describes.
These checks run every time the page is built, and a failure stops the build. Figures are whole euros; rates are rounded to one decimal place.

On your data

What an agent-made dashboard is

A recurring report, or a simple hosted page like this one, that AI agents build from your warehouse or your exported files. It is not a Looker or Power BI build, and nothing is installed on your systems.

What it reads. Billing history per customer, with plan changes and cancellations; trials from your CRM; sign-ups from your product’s events. From exported files or your warehouse. Where the data goes is set out on data handling.

How it is built. In the first sprint, agents write the dashboard’s queries and its checks: the loads, the duplicates, the totals and the bridge. The checks are scripts, not AI, and a review agent reviews the work before it reaches you (on what you act on, a panel of three reviewer agents and a judge, then the analytics director’s sign-off). Each week after that, the dashboard runs again on the week’s data, with the same checks and its method note. How the checks and the review fit together is on how it works.

How it arrives. As a simple web page like this one. The month’s view, with a forecast, is the monthly revenue report. Which KPIs a young subscription company tracks, and how each is defined, is in what KPIs a Series A startup should track.

Not covered. Building a data warehouse, or dashboards in Looker or Power BI. The dashboard reads the data where it already is.

Start here

Want this dashboard on your own numbers?

Up to thirty minutes on your billing data, then a fixed proposal.

Book a free call