You ask the questions. The agents do the digging.
Pulling the numbers, reconciling the sources, working out what changed and why - the agents do all of it. You ask, and you decide what the numbers are allowed to mean.
The same questions. A fraction of your team's time.
Same questions, same answers expected. What changes is how much of the digging your team has to do.
It knows what your numbers actually mean.
Agreed with you over the first weeks, in plain language:
Your definitions
What counts here as a lead, a customer, a churn, a month. Agreed once, so nobody argues about whose figure is right.
Which source wins
When two systems disagree on the same number, which one is the truth here - and what each quietly gets wrong: sampling, lag, the usual traps.
What normal looks like
Your seasonality, your usual range, and the point at which something is worth interrupting someone about.
What you keep asking
The questions that come round every week and every quarter, so the answer is already worked out.
It runs on what you already have.
The agents read your accounts directly. No new warehouse, no migration, no rebuilding your tracking.
From the five places your numbers live to the answer you asked for.
Every source pulled and reconciled, into whatever cut you asked for:
Every number arrives with its source and the moment it was pulled.
You run it from wherever your team already works.
Same agents, same definitions. Only the surface differs.
In a terminal - Claude Code, Codex and the like
›why did signups drop in july
·sessions flat; organic conversions -38%, all on /pricing
·starts 3 July - the day the pricing table changed
›is it seasonal?
·last two Julys: no dip
·paid conversions on the same page fell the same amount
›what would you change first
·old table on half the traffic for two weeks
·cheapest test; write-up for the team is ready
›build me a dashboard for the monday call
·6 tiles, week on week and month on month, already computed
For the people who already live in a terminal.
In chat - Slack, Teams, WhatsApp
Your team
Put ad spend next to closed deals, last 12 months.
Agent
In the thread. Paid costs €610 a customer, organic €95. The organic figure leaves out what you spend on content - say the word and I will fold it in.
Your team
Is anything broken this week?
Agent
Yes. The checkout page has fired no events since Tuesday. The numbers from then on are too low, not the sales.
Your team
And how did the new landing page do?
Agent
Better, but not significantly - 340 sessions is too few to call it. I will know in two weeks.
For everyone else - no new program, no training.
You can trust it.
It does not change anything.
The agents read. Tracking, budgets and live settings are not theirs to touch.
An empty report says why it is empty.
You never see a zero that is really a connection that stopped working three weeks ago.
Every number carries its source.
And the moment it was pulled, so you can check it rather than trust it.
EU compliant.
Set up around your GDPR obligations: your data stays in your own accounts, and nothing is copied into a product of ours.
EU hosting
European regions, or your own servers.
Local models
For when nothing may leave your network.
A full audit trail
What was used, when and why.
It gets sharper the longer it runs.
Every correction you make is written into it the same day and holds from then on.
Week one
It knows your definitions and where every number comes from.
Month three
It knows what normal looks like for you, which is why it can tell you when something is not.
Month nine
It carries the history of what you changed and what happened after. That is the part that is yours.
See what this would look like for your numbers.
Bring the questions your team keeps failing to answer. We go through what your data can already tell you, and what it would take to get the rest.