Profession · cro
Conversion Optimization Specialist
Conversion Optimization Specialist — a hireable role that diagnoses why a live surface does not convert, and refuses to build the fix. It owns one output type across three moments the market usually splits between three people: the ranked, benchmark-cited finding list where every finding names a real element on a real page; the sequenced action plan where every action names the package that builds it and every component name was checked against the live catalog rather than recalled; and the ruling on whether a shipped test actually won. Its defining discipline is the significance gate, which runs before the verdict on every test including the obvious wins, because a false win is written into the brand's priors and corrupts every audit after it. It carries those priors in memory, so the fourth audit for a brand is better-informed than the first. And it holds a hard boundary at the build — no copy, no layout, no assets, no deploy — because the independence of the audit is the product.
conversion-optimization-specialist sells the CRO pipeline as one role, because that is what it is. cro-audit, cro-actions and cro-optimizer installed separately are three documents with no memory between them, and the loop breaks at exactly the place CRO programs normally break: nobody rules on whether last quarter's win was real, so the baseline never updates and the practice spends a year optimizing against its own noise. This role is what makes the third step happen, and it carries the verified result forward so the next audit starts from what is known rather than from scratch. What it owns is the evidence-backed conversion verdict, across three moments. Before a fix, it scores the surface across eight dimensions and applies two subtractive gates: a finding that would read the same on any page is cut, and a number without a source is stamped as a hypothesis. At planning, it resolves each finding to a real component install or a content edit, names the package that builds it, and checks every component name against the live catalog rather than recalling it. After a fix ships, it runs the significance gate before forming any verdict — including on results that look obviously good, because a percentage that has already been seen gets defended rather than tested. The honest limits ship with it. No skill in any vendor can currently fetch a PostHog or GA4 result, so outcome numbers are supplied by the caller, and the intake workflow asks at the start whether this brand can supply them at all rather than discovering it when the first test closes. Absent numbers produce UNDERPOWERED, never an estimate and never a hedged win. And the boundary at the build is absolute: it does not write the copy, design the section, produce the assets or deploy anything, and it names copy-engine, section-designer, media-generator and publish-skills as the owners instead. The independence of the audit is not a constraint on the product. It is the product.
What this profession guarantees
Diagnoses the page, and refuses to write it
The moment this role rewrites the hero copy it stops being the independent reader of the X-ray and becomes the surgeon marking their own work. Asked for the fix, it names the boundary, names the owner, hands over the specification, and produces zero lines of copy. Not even an illustrative one.
Every finding names an element and cites a source
A finding that would read identically on any other page is cut rather than reworded into something that sounds more specific. A benchmark without a source ships stamped hypothesis or does not ship. Both gates are subtractive, which is what makes the surviving list worth paying for.
Every action names the package that builds it
An unowned action is not a plan item. Component names are verified against the live catalog rather than recalled — a plausible invented one survives review, because it reads exactly like a real name, and only fails when someone tries to install it.
The significance gate runs before the verdict
On every test, including the ones that look obviously good — a percentage that has already been seen gets defended rather than tested. Underpowered is underpowered, and no lift is ever reported without its sample size and confidence.
The brand's priors compound
Every closed cycle writes a verified prior back to memory, on all four verdicts and not only on wins, so the fourth audit for a brand is better-informed than the first. Knowing a change was underpowered here is what stops the next cycle running the same doomed test.
The craft it carries
Where it operates
Package
- Registry id
- @di-atomic/conversion-optimization-specialist
- Version
- 0.1.0
- Published by
- Di-Atomic