02 / 07 · Deep analytics
One joined view, audited before it is trusted.
Every other agent depends on this one. The analytics agent joins your web analytics, ad accounts, CRM and product events into a single view of what actually produced revenue, and it audits the tracking first, because a decision made on a broken event is worse than no decision. It is the reason a result in one channel counts as evidence for the next rather than a separate report.
What it reads
What it analyses
- Does each event fire once, at the right moment, with the attribution it should carry? Duplicate and missing events are found before anything is optimised against them.
- Where platform-reported conversions and your own records disagree, and by how much.
- Which source, page and campaign combinations produce qualified pipeline rather than just form fills.
- What the funnel looks like end to end, stage by stage, so a change anywhere can be read at the revenue end.
How it decides
The agent does not optimise anything itself. It decides what the other agents are allowed to trust. A metric is marked reliable only once it reconciles against a second source, and an experiment elsewhere in the stack cannot be verdicted on a metric this agent has flagged. When platforms disagree with your own data, your data wins and the gap is reported rather than averaged away.
Example experiments
The shape of a test, written before it ships.
Illustrative, not a promise. Every experiment on your account gets this record before anything changes: the hypothesis, what the verdict is read from, what stays fixed, how long until a verdict, and where a person decides.
Hypothesis 01
Platform-reported conversions overstate what the CRM later confirms, and the gap is concentrated in a small number of campaigns.
- Measured on
- Reported conversions against CRM-confirmed leads, by campaign, over the same window.
- Held constant
- Campaign settings and budgets while the reconciliation runs.
- Verdict after
- One month of matched data.
- Human gate
- You confirm which CRM stage counts as a real lead.
Hypothesis 02
A material share of enquiries carry no source at all, and fixing attribution on the form changes which channel looks best.
- Measured on
- Share of leads with a resolved source before and after the fix, and the channel ranking under each.
- Held constant
- Marketing activity itself; only the measurement changes.
- Verdict after
- Two to four weeks after the fix ships.
- Human gate
- Tracking changes reviewed before they go live.
Where people decide
You define what counts: a lead, a qualified lead, a customer.
Tracking changes are proposed as code and reviewed before they ship.
Access is read-only by default and scoped per source.
What you get
- 01A tracking audit with every broken or duplicated event named.
- 02One joined funnel view, from first touch to revenue, that every other agent reads from.
- 03A reconciliation each month between what platforms claim and what your records show.
- 04The baselines every experiment on every other page is measured against.