conversion-optimization

Imported from wondelai/skills (conversion-optimization/SKILL.md). Install upstream with npx skills add wondelai/skills --skill conversion-optimization. Copyright stays with the author (MIT).

Conversion Optimization

Turn one leaking conversion flow — a landing page, a signup, a checkout, an in-app onboarding — into a measured, tested funnel. This is an interactive, resumable journey of seven phases: the agent asks before every decision and records the outcome in your project's docs/ folder, so you can stop after any phase and pick up later. It works on websites and inside products alike; the unit of work is the flow and its ONE action, not the whole site.

Core Principle

Find the leak with numbers, learn the reason from customers, fix message → offer → proof → friction in that order, and prove every fix with a pre-committed test. The order is causal: a number tells you where, only research tells you why, motivation must be raised before friction-cutting pays, and an untested fix is a guess that compounds. This skill sequences the phases, asks the decision questions, and records every choice in docs/. The constituent skills carry the method — invoke them rather than improvising their frameworks.

Journey Map

Operating Rules

1. Resume first. Before anything else, read docs/CONVERSION-OPTIMIZATION-PLAN.md and every artifact in the Journey Map. If the tracker exists, summarize the journey state in 3-5 lines and ask which phase to enter. Done when the user has confirmed an entry point. A journey with a tracker is resumed, never restarted.
2. Intake on first run only. No tracker: run the Intake below, then create docs/CONVERSION-OPTIMIZATION-PLAN.md with every phase statused pending | in-progress | awaiting-evidence | done | deferred: reason | skipped: reason. Done when the tracker exists and the user has confirmed the phase plan.
3. Phase entry. Announce: what the phase does, the decision it forces, the artifact it produces, rough effort. Offer proceed / skip / defer — phases marked GATE may be deferred, never skipped. Mark the phase in-progress on proceed. Done when the user chose.
4. Skill invocation and fallback. Load the phase's skill and use it: each phase's Invoke line names the skill by slug — use that skill to run the phase. If it is not available, offer: npx skills add wondelai/skills/<slug> --global. If the user declines, run the phase from its Brief — the minimum viable method. State which mode you are in.
5. In-phase decisions. Ask every question under "Decide with the user" — with concrete options and your recommendation. Record the choice in the tracker's Key Decisions. A decision made silently is a defect.
6. Phase exit. Present the draft artifact content for sign-off before writing. On approval: write or extend the docs/ files, update the tracker (status, Key Decisions, Next Actions). Done when the files are written and the phase row shows done.
7. Artifact discipline. Read before writing; create a file only if missing, otherwise extend — add or update your sections, preserve everyone else's. Files are UPPERCASE in docs/. Every recommendation lands as a checkbox or a table row with owner and priority. See references/artifact-templates.md when creating a docs/ file for the first time — create it from the full skeleton (all section headings), then fill the sections your phase names.
8. Measure, research, then change — and every change is a test. No fix ships without a Phase 1 leak and a Phase 2 researched reason behind it, and every shipped change lands in docs/EXPERIMENTS.md with a pre-committed primary metric and a guardrail. A fix with no research behind it goes back to Phase 2; a bold change with no test attached stays in the backlog until it has one.

Intake

Ask these before creating the tracker:

1. Which flow are we optimizing, and what is the ONE action at its end? (Scopes every phase — a flow with three competing CTAs has no goal.)
2. Where do the numbers say people drop — analytics, funnel steps, cohort data? Paste what you have. (Feeds the Phase 1 leak diagnosis; no instrumentation means Phase 1 starts by adding it.)
3. Roughly how much traffic or volume enters the flow per week? (Gates Phase 7 — decides whether A/B tests can reach significance or the journey leans on qualitative evidence and before/after windows.)
4. What voice-of-customer sources exist or can be gotten — exit surveys, session recordings, support tickets, sales calls, reviews — and can you paste or export the raw text of the best one? (Phase 2 objections must quote the customer's own words, so the phase needs the content, not just the source name.)
5. What is the current offer at the conversion point — price, guarantee, bonuses — and can it change? (Gates Phase 4; a contractually fixed offer narrows it to presentation.)
6. Do docs/POSITIONING.md or docs/OFFER.md already exist from another journey? (Phases 3-4 build on them rather than restarting.)
7. How much of the journey do you want now? (Phases 1-2 are the mandatory diagnosis; 3-6 are the fix passes aimed by it; 7 turns fixes into proof.)

Skip heuristics: skip Phase 3 when messaging was already validated (e.g. an improve-website journey completed its message phases); skip Phase 4 when the offer is fixed by contract — record skipped: reason. Phase 7 may be deferred: reason at very low traffic in favor of before/after evidence with an explicit revert trigger, never silently skipped. Never skip Phases 1-2 — an unfound leak and an unresearched reason turn every later phase into guessing.

Then create docs/CONVERSION-OPTIMIZATION-PLAN.md from the template and confirm the plan. Done when the tracker exists with every phase statused and the user has confirmed the plan.

Phases

Phases run in the listed order — each assumes the previous phase's artifact exists. Any phase can be entered, skipped, or deferred per the Operating Rules, but Phases 1-2 gate them all: nothing downstream fixes a problem that isn't a measured leak with a researched reason. When running any phase from its Brief (constituent skill not installed), read references/methods.md first — it carries each phase's full method, checklists, formulas, and benchmarks; the Brief is only the summary.

Phase 1 — Find the leak (lean-analytics) — GATE

Purpose: Locate where the flow actually loses people and pick the one metric this journey moves — before any opinion about why.

Brief (fallback): A good metric is a comparative ratio that changes what you do next; totals and cumulative charts are vanity. Express each step of the flow as a conversion rate, compare against your own history and published benchmarks (e-commerce converts ~1-3% of visitors; landing pages on paid traffic low single digits), and find the biggest absolute drop on the highest-value path. Pick the One Metric That Matters for this journey, pair it with a counter-metric so it can't be gamed (signup rate × 30-day retention), and draw a line in the sand: target, date, pre-committed miss response. Cohort and segment (channel, device, plan) — one collapsing segment hides inside a flat average.

Invoke: Use the lean-analytics skill with the flow steps and analytics from intake. Ask for a step-by-step funnel table with baselines and benchmarks, the OMTM plus counter-metric for this journey, and the biggest leak ranked by absolute lost value.

Decide with the user: (1) Confirm the OMTM and its counter-metric. (2) Which leak to attack first — biggest absolute loss on the money path, not the easiest percentage. (3) If instrumentation is missing, which minimal events to add first — the phase stays awaiting-evidence until the numbers exist.

Artifact: Extend docs/METRICS.md ## Funnel (stage | conversion | benchmark | bottleneck?), ## Stage & One Metric That Matters, and ## Baselines & Targets. Update the tracker.

Done when: the funnel is measured at the coarsest granularity that still localizes the leak to a single step, the OMTM and counter-metric are recorded with a line in the sand, and the leak is named — only then are Phases 2-7 unlocked. Finer sub-steps awaiting instrumentation stay awaiting-evidence in Next Actions and do not block the journey, provided the named leak does not depend on them.

Phase 2 — Research why they leave (cro-methodology) — GATE

Purpose: Replace guesses about the leak with evidence from real visitors. Phases 3-6 may only fix problems traceable to a finding here.

Brief (fallback): Don't guess — discover. Mine primary sources (a one-question exit survey: "What's preventing you from [action] today?"; post-conversion: "What almost stopped you?"; chat logs, tickets, sales calls) and secondary sources (reviews, competitors) for the customer's own words. Sort objections into the Big 5 — Trust, Price, Fit, Timing, Effort — and build the O/CO table: every objection gets an evidence-backed counter placed at the exact step the doubt arises, never in an FAQ. Diagnose each step with the LIFT lenses (value proposition ± clarity, relevance, urgency, minus anxiety and distraction — Goward) and the MECLABS heuristic (conversion rises with motivation and value clarity, falls with friction and anxiety). Rank fix hypotheses by ICE and apply the 10x screen: if a change couldn't plausibly double the step, don't queue it.

Invoke: Use the cro-methodology skill with the Phase 1 leak, the flow, and the voice-of-customer sources from intake. Ask for the researched objection list in customer words, the O/CO table with placements, missing persuasion assets, and an ICE-ranked hypothesis backlog.

Decide with the user: (1) Which researched objection is the primary leak driver. (2) Low traffic: accept qualitative plus heuristic evidence — explicitly. (3) Which implicit objections need CO-Only counters (countered without being stated).

Artifact: Create docs/FUNNEL.md with ## Flow Map & ONE Action, ## Leak Diagnosis, and ## Objections & Counters (O/CO); extend docs/EXPERIMENTS.md ## Experiment Backlog (ICE-ranked). METRICS.md ## Funnel stays canonical for the conversion numbers — Leak Diagnosis cites them and adds the researched reason and severity. Update the tracker.

Done when: the flow map names the ONE action per step, every researched objection has an evidence-backed counter and a placement, and the backlog is ICE-ranked.

Phase 3 — Fix the message at the leak (storybrand-messaging)

Purpose: Make the leaking step say what the visitor gets, in their words, in five seconds — clarity converts before persuasion can.

Brief (fallback): The customer is the hero; you are the guide. Run SB7 on the leaking step: a Character who wants one thing, their Problem at three levels (external, internal — the frustration the copy must name, philosophical), you as the Guide (empathy + authority), a 3-step Plan that makes acting feel safe, one Direct plus one Transitional CTA, and named failure/success stakes. Rewrite the step's headline in customer language pulled straight from Phase 2 — customer words outperform copywriter words. Then make it stick (Made to Stick): concrete beats abstract ("save 16 hours a month," not "boost productivity"), and pick the single Commander's Intent message the visitor must still remember tomorrow.

Invoke: Use the storybrand-messaging skill with the step's current copy, the Phase 2 objection evidence, and POSITIONING.md if it exists. Ask for above-the-fold rewrites that name the internal problem, a one-liner, and one Direct plus one Transitional CTA.

Decide with the user: Which rewrite passes the 5-second test (a stranger can say what's offered and why it matters); which internal problem the copy names; whether the step keeps a transitional CTA or goes single-CTA.

Artifact: Extend docs/POSITIONING.md ## Brand Script (StoryBrand), ## One-Liner, and ## Key Messages (surface | message | status); record the rewrite in the Proposed message/CTA column of docs/FUNNEL.md ## Flow Map & ONE Action, leaving Current message/CTA intact as the before-state Phase 7 measures against; append copy tests to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.

Done when: the leaking step has a rewritten message that names the internal problem, one primary CTA, and a logged test hypothesis.

Phase 4 — Make the offer worth acting on (hundred-million-offers)

Purpose: Strengthen what is actually exchanged at the conversion point — the best flow cannot sell a weak offer, and the offer is the biggest single lever.

Brief (fallback): Value = (Dream Outcome × Perceived Likelihood) ÷ (Time Delay × Effort & Sacrifice). Raise the numerator with outcome language and proof; crush the denominator with speed ("first result in 5 minutes") and done-for-you framing. Reverse the risk with a guarantee aimed at the top Phase 2 objection — it raises perceived likelihood and lowers anxiety at once. Stack named, honestly-valued bonuses that each kill one objection; present price after value, anchored against the stack. Scarcity and urgency only when real — fake deadlines convert once and churn forever.

Invoke: Use the hundred-million-offers skill with the current offer from intake and the Phase 2 Price/Timing/Effort objections. Ask for a Value Equation score per lever, a guarantee design matched to the top objection, and a trim-and-stack pass on the offer components.

Decide with the user: Which guarantee the business can actually honor; which bonuses are real and sustainable; whether price presentation changes (anchoring, payment plans) — pricing itself may be out of scope; record that.

Artifact: Extend docs/OFFER.md ## Offer Stack (element | description | honest value | objection it kills), ## Price Metric when the flow raises what the price is charged per, plus ## Willingness-to-Pay Evidence if new evidence surfaced; append offer tests to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.

Done when: each Value Equation lever has a concrete change or a reason it stays, the guarantee targets the top researched objection, and every offer change carries a test hypothesis.

Phase 5 — Put proof at every doubt (influence-psychology)

Purpose: Answer each remaining objection with honest evidence placed at the exact step the doubt arises.

Brief (fallback): Under uncertainty people use shortcuts: social proof (specific numbers — "2,347 teams" — and similar-others beat generic praise), authority (credentials, certifications), commitment (a small first yes makes the big yes consistent — ask for the card after the value moment, not before), reciprocity (give the useful thing first), and scarcity (real only). Audit the flow for proof gaps: every O/CO row needs its counter actually rendered — testimonial, data point, logo bar, guarantee seal — at its placement. Proof hierarchy: specific results with context > named testimonials with faces > case studies > statistics > logos. The transparency test gates everything: if knowing the technique would make the visitor feel tricked, it fails. Fabricated proof and hidden costs are defects, not tactics.

Invoke: Use the influence-psychology skill with the FUNNEL.md O/CO table and the flow's current proof. Ask for a per-step proof audit, which principle answers each open objection, and copy for the two highest-impact placements.

Decide with the user: Which proof assets exist versus must be acquired (testimonial requests, case studies — log acquisition as Next Actions); where scarcity or urgency claims are genuinely true; which steps get a commitment micro-ask.

Artifact: Extend docs/FUNNEL.md ## Proof Inventory (asset | type | placement | status) and complete the ## Objections & Counters (O/CO) placements; append proof tests to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.

Done when: every open objection row has a real proof asset placed or an acquisition task with an owner, and no claim in the flow is unverifiable.

Phase 6 — Remove the friction (design-everyday-things)

Purpose: Protect the visitors who decided to act. B=MAP: behavior happens when Motivation, Ability, and a Prompt converge — Phases 3-5 raised motivation; this phase raises ability and sharpens prompts.

Brief (fallback): Bridge Norman's two gulfs. Execution: clear signifiers (buttons look pressable, fields look editable) and constraints that make errors impossible (date picker over free text, submit disabled until valid). Evaluation: feedback within 0.1s of every action, progress indication on multi-step flows. Forms (Baymard): every field costs conversions — cut to the minimum, ask for payment as late as possible, show all costs before the final step (surprise shipping and taxes is the #1 checkout killer), offer guest checkout, validate inline with messages that say how to fix. Error messages state what went wrong and how to fix it, without blame; slips get undo, not are-you-sure dialogs.

Invoke: Use the design-everyday-things skill with the conversion-critical steps (form, payment, confirmation). Ask for weak signifiers, where constraints replace error messages, feedback gaps, field-by-field form cuts, and message rewrites.

Decide with the user: Which fields are truly required now versus collectable later; where a constraint replaces a warning; whether the flow shows total cost earlier; which severity-4 friction items ship immediately as a logged before/after card versus wait for a full test.

Artifact: Extend docs/FUNNEL.md ## Flow Friction Audit (step | issue | severity 0-4 | fix | status); extend docs/DESIGN.md ## UX Audit Findings for reusable component fixes, naming the Norman gulf (execution or evaluation) in the Heuristic column; append fixes to docs/EXPERIMENTS.md ## Experiment Backlog. Update the tracker.

Done when: every step has a friction audit row, forms are cut to minimum fields, severity-4 items have owners, and error messages meet the checklist. Where money changes hands in this flow, all costs are visible before the final step and guest checkout is offered; in a flow that takes no payment, mark those two rows n/a.

Phase 7 — Prove it (cro-methodology)

Purpose: Turn the high-ICE fixes into trustworthy experiments — without rigor you cannot tell a real lift from noise, and a false winner compounds forever.

Brief (fallback): Pre-commit everything (Kohavi): primary metric, guardrail metrics (the Phase 1 counter-metric, revenue per visitor, support volume), sample size computed from baseline rate and minimum detectable effect at 95% significance / 80% power, and a duration of at least one full business cycle covering weekdays and weekends. Never peek and stop early — it manufactures false positives; never rerun until you like the answer. Test bold changes — meek tweaks rarely reach significance. Low traffic: run a before/after window with qualitative confirmation, or ship as a reversible bet with an explicit revert trigger — and say which you are doing. Practical significance gates rollout: a statistically significant 0.1% lift may not pay for its complexity. Winners become the new control; losers become learnings written down.

Invoke: Use the cro-methodology skill with the top ICE hypotheses from the backlog. Ask for full experiment designs: hypothesis in "If [change], then [metric] because [research]" form, pre-committed sample size and duration, and the decision rule.

Decide with the user: Which 1-3 hypotheses run first (highest ICE on the money path); the guardrail metrics; what happens on a flat result — iterate the fix, or return to Phase 2 for better research.

Artifact: Promote backlog rows to docs/EXPERIMENTS.md ## Experiment Cards, adding the pre-committed sample size per arm, planned duration, and the minimum lift worth keeping as extra bullets on the card (the skeleton's bullets carry hypothesis, metrics, and decision rule; extra bullets are additive and allowed); record results and verdicts as they land; extend docs/METRICS.md ## Baselines & Targets with a new dated row per tested metric, leaving the pre-test baseline row intact. Update the tracker.

Done when: each shipped fix is a card with a pre-committed metric, sample size, and decision rule; results carry verdicts; and the OMTM's new baseline is written down.

Optional Phases

Optional phases follow the same operating rules — load and use each listed skill exactly as a core phase would; insert where the Add-when condition first becomes true. They carry no inline Brief: standalone, run ux-heuristics as a severity-rated pass over Nielsen's 10 heuristics, microinteractions as a trigger/rules/feedback/loops inventory, scorecard-marketing as a quiz-funnel design ending in a personalized result, and hooked-ux as a trigger → action → variable reward → investment loop audit — or install the named skill for its full framework.

Common Mistakes

Completing the Journey

A funnel always has a next-biggest leak: when the Phase 7 verdicts land, re-enter Phase 1 with fresh numbers rather than declaring victory — the journey is a loop with an exit condition, and the exit condition is the line in the sand from Phase 1, not exhaustion.

Exit checklist — every box tied to an artifact:

• Every phase in docs/CONVERSION-OPTIMIZATION-PLAN.md is done, deferred: reason, or skipped: reason.
• The OMTM, counter-metric, and line in the sand are recorded with pre- and post-test baselines (METRICS.md).
• Every researched objection has a placed, verifiable counter (FUNNEL.md O/CO and Proof Inventory — no open rows on the money path).
• The flow's forms and steps carry no severity-4 friction without an owner (FUNNEL.md Flow Friction Audit).
• Each shipped change is an Experiment Card with a pre-committed metric and a recorded verdict (EXPERIMENTS.md).

Close the tracker: remaining Next Actions carried into FUNNEL.md and EXPERIMENTS.md so nothing is lost. Then route forward: when the flow converts and needs more qualified traffic, continue with the grow-website skill; when the leak has moved past conversion into engagement and retention, continue with the improve-app skill; when the whole site — look, typography, speed, message — needs the broader pass, continue with the improve-website skill.