Conversion Optimizer . Agency OS

What it does

Turns a live page or funnel plus its numbers into a diagnosis and a plan. It scores a page on a weighted scorecard covering value proposition, relevance, clarity, social proof, CTA effectiveness, anxiety and distraction, maps a funnel step by step to find where the biggest drop-off meets the most traffic, designs A/B tests with a stated hypothesis and a single deciding metric, strips friction out of forms, and can rewrite a full landing page structure when the diagnosis calls for it. Every finding ties to something visible on the page or present in your data, and every fix is ranked by expected revenue impact against implementation effort. It thinks in visitor psychology using the LIFT Model, Cialdini's principles, the Fogg Behavior Model and AIDA, not in vanity metrics.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

> why is this page not converting?

> audit this landing page

> diagnose my funnel, where am I losing people?

> plan an a/b test for my headline

> quick wins I can ship on this page today

When to reach for it

a live page is getting traffic but sign-ups or sales are thin
you have funnel numbers and need to know which step is leaking the most revenue
ads are getting clicked but the landing page is not holding anyone
a form is where prospects go to die and you suspect the field count
you want to test something but need a plan before touching the page
you need changes you can implement today without a designer or developer

When NOT to use it

If you actually want	Use this instead
a funnel or page set that does not exist yet: this skill improves pages that exist, it does not build new ones	`funnel-architect`
a judgement call on a draft with no traffic behind it, where the question is message, voice or whether it lands with the reader	`critique`
fresh persuasive copy written from master frameworks once the diagnosis has named the problem	`legendary-copywriters`
the recommended fixes actually implemented on the site	`site-builder`
email or SMS sequences to patch the drop-offs the funnel analysis finds	`email-sequence`

Before you start

What you need	Why
The page itself: a URL it can read or the copy pasted in	every diagnosis must map to a quoted element on the page, so there is nothing to diagnose without the page in front of it	Required
Your raw numbers with a date range: sessions and conversions per step, not a pre-rounded rate	rates get computed from your figures with the formula shown, and metrics that arrive without a date range are flagged before any trend claim is allowed	Required
Brand context: `positioning.md` and `audience.md`, or offer, price point and audience supplied by hand	fixes are matched to who actually visits and what they are buying; without this the recommendations stay structurally sound but generic	Optional
`ica-output.md`	deepens the psychology behind relevance and anxiety findings; message match is judged against the real visitor, not a guess	Optional
Analytics, heatmap or session recording data if you have it	used only when supplied; behavior data you did not provide will never be cited as if it existed	Optional

How it runs

Load contextReads brand memory for the active workspace under the freshness rules, plus any learnings.md patterns tagged conversion-optimizer from earlier runs, so past test results and corrections shape this one.
Pick the operating modeRoutes your request to the right job: a full funnel diagnostic, a scored page audit, a test plan, a form fix, a quick-wins pass, or a complete page rebuild. You do not have to name the mode, the request itself decides it.
Work only from what you gave itComputes conversion rates from your raw numbers and shows the working, for example 38 opt-ins from 1,200 sessions is 3.2 percent, stated to no more precision than the data supports. Missing date ranges are flagged here, before any interpretation begins.
Diagnose against the frameworksEvaluates the page with the LIFT Model, Cialdini, Fogg and AIDA. Each finding names the specific element it comes from: the headline, the CTA label, the field count, the absent proof section. Nothing is attributed to data that was not provided.
Score and prioritiseProduces the weighted scorecard, then ranks fixes by revenue impact: traffic at the step, times the plausible improvement, times value per conversion. High traffic meeting above-average drop-off rises to the top. Each fix states what is broken, why, the exact change and the effort.
Plan the testsEvery proposed test carries a hypothesis in the form "we believe this change will move this metric because of this reasoning", a described control and variant, one primary metric that decides the winner, a minimum sample size and duration, and kill criteria. Ideas are prioritised by impact, confidence and ease.
Deliver and logOutputs the structured cro-audit-v1 record that downstream skills read, logs the deliverable to assets.md, and asks how the output landed so the learning loop has something to learn from.

What you get

A scored page audit: the weighted score, every dimension below par with the specific problem, the conversion impact, the ready-to-implement fix and the framework behind it
A funnel map with the conversion rate between each step, the drop-offs quantified, and the single highest-leverage fix named
A fix priority stack ordered by revenue impact against effort
A/B test plans, each with hypothesis, control, variant, primary metric, sample size, duration and kill criteria
A form audit with the reduced field set and the reasoning per removed field
A quick-wins list of changes shippable inside a day without design or dev resources
Full landing page copy and structure when you ask for a rebuild, from hero to thank-you page
Revenue framing on the priority fixes, always as an illustration from your own figures, for example "lifting this step's rate by two points at your traffic and order value is worth roughly an extra few thousand a month"
A cro-audit-v1 record that critique can validate, site-builder can implement, legendary-copywriters can rewrite from and email-sequence can patch against
An assets.md entry tagged conversion-optimizer

Honest limits

Read this before you rely on it

It diagnoses only from the page and the numbers you give it. If you did not supply heatmaps, recordings or user quotes, none will be cited. A recommendation will never lean on research that does not exist.
No invented industry benchmarks. It will not tell you what "the industry average" is unless a source is on the table. Comparisons run against your own baseline or they do not run.
Metrics without a date range are flagged before any trend claim. "Conversion is dropping" is not a sentence it can produce from undated numbers.
Every test comes with a hypothesis and a metric. A change that cannot state what it should move and why does not make the plan, which also means it will refuse the vague "just try stuff" brief.
It recommends, it does not implement. The actual page edits, tracking setup and test tooling sit with site-builder, your developer or your page platform.
Revenue impact figures are estimates built from the numbers you provided. They are for ranking fixes against each other, not for forecasting to a client.

Where people go wrong

The mistake	Do this instead
Bringing it a draft nobody has visited yet	That is a critique job. This skill earns its keep on something live, where real behavior and real numbers exist to diagnose.
Pasting a pre-rounded conversion rate instead of the raw counts	Give it sessions and conversions with dates. The raw figures let it show the working and size the samples a test needs; a bare percentage hides both.
Starting with button colors and small visual tweaks	Test in priority order: the headline and value proposition carry the most weight, then CTA copy and placement, then proof, then structure, then everything else.
Running a change with no hypothesis and calling it a test	Write the sentence first: we believe this change will move this metric because of this reasoning. If you cannot finish the sentence, you are redecorating, not testing.
Calling a winner after one strong weekend	Let the test run its full planned duration through complete weekly cycles at the stated confidence level. Early peeking crowns false winners.
Auditing the desktop page and shipping the fixes	Most visitors are on a phone. Check the sticky CTA, the field count, the font size, the tap targets and the load time before anything ships.
Quoting an industry benchmark in a client report	Use the client's own baseline, or say plainly that the number does not exist yet and go establish it.

Worth knowing

The fastest route to value is usually the funnel diagnostic, not the page audit. One highest-leverage fix, chosen where heavy traffic meets a heavy drop-off, tends to beat a long list of page polish. Run the diagnostic first, then audit the one page it points at.