01 / 07 · Website CRO

Tests on your site, verdicted against a baseline.

The CRO agent reads your analytics and search data every week, picks the one change most likely to move enquiries, ships it to the site behind whatever approval you set, and reads the verdict weeks later against a baseline written before anything changed. It is the loop most agencies describe and few actually run: a hypothesis, a control, a review date, and a result that is allowed to be a loss.

What it reads

Analytics sessions and eventsSearch Console queries and pagesForm and enquiry eventsPage speed and indexing stateYour site repository

What it analyses

  • Which pages get traffic and produce nothing, and which convert but get no traffic.
  • Where the enquiry path leaks: landing page to form, form start to submit, submit to a real lead in the CRM.
  • Whether the tracking is telling the truth, before any of it is trusted. Most accounts are measuring at least one thing wrong.
  • What changed since last cycle, and whether the last experiment has settled enough to call.

How it decides

Every change is logged before it ships: the hypothesis, the arms, what is held constant, the baseline and a review date. Verdicts are applied from a written rules file, not from whoever is looking at the dashboard. On a small site the rules are built for small numbers: compare like periods, use medians rather than averages, control for the traffic trend, and call a result directional rather than proven when the volume only supports that. A loss is logged with the same care as a win, because a loss tells you which model of the visitor was wrong.

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

Leading the service pages with the specific problem the visitor searched for, rather than the company introduction, lifts enquiry rate on those pages.

Measured on
Enquiry starts and submits per session on the changed pages, against the same weeks last period.
Held constant
Navigation, form fields, pricing and the paid traffic mix.
Verdict after
Four to six weeks, longer if the pages are low traffic.
Human gate
Copy approved by you before it ships.

Hypothesis 02

A shorter first form step, with the qualifying questions moved to a second step after the email is captured, lifts completed enquiries without lowering lead quality.

Measured on
Form completions and the share of those the CRM later marks qualified.
Held constant
The page above the form and every other entry point.
Verdict after
One full sales cycle, so quality can be read as well as volume.
Human gate
You set the qualification rule the CRM read uses.

Hypothesis 03

Pages that rank but do not get clicked have titles that promise the wrong thing; rewriting them to match the query intent lifts click-through without changing position.

Measured on
Click-through rate per query on the rewritten pages, against the unchanged control set.
Held constant
Page body, internal links, and everything on the control pages.
Verdict after
Three to four weeks of settled search data.
Human gate
Title set reviewed before publishing; the agent may ship on its own once it has earned that.

Where people decide

Hypothesis selection: the agent proposes one change per cycle and you can veto it.

Copy and design approval before anything visitors see is changed.

Ship to production: starts as a pull request for you to merge, and moves to automatic only when you say so.

What you get

  1. 01A tracking audit in the first fortnight, with what was wrong and what was fixed.
  2. 02A written baseline you can hold every later claim against.
  3. 03One shipped experiment per cycle, logged before and verdicted after.
  4. 04A monthly read: what changed, what it moved, what needs a decision.

Want the website cro agent on your stack?

Get in touch