No operator in the installable-skill ecosystem publishes free-to-paid conversion data — and the cannibalization framing is the wrong worry
The question
"Does a free top-of-funnel skill measurably convert to paid in the installable-agent ecosystem, or does free-catalog abundance cannibalize the paid tier?" Surfaced as open follow-up #5 from [[2026-07-23-executable-agentic-info-product-pricing-scan]], which found every live operator runs a large free catalog as top-of-funnel while MAC has no free surface at all.
What we already know (from the vault)
- The parent scan established the funnel shape but measured none of it. [[2026-07-23-executable-agentic-info-product-pricing-scan]] verified that CLSkills gives away 1,545 community skills, 20 agents, and a guide before charging $15, that tonsofskills carries 2,810 free skills, and that Anthropic's own publishing surface is free. It found zero conversion, install-count, or refund figures for any of them. The "free is the real competitor" conclusion was structural inference, not measurement.
- RDCO's planned free surface is a lead magnet, not an installable skill, and it has not shipped. [[2026-05-18-mac-matrix-downloadable-template-status]] identifies the MAC testing matrix as the Phase 1 lead magnet, living only as a 19KB markdown file in the private vault. Its Notion/Sheet duplicable version was scoped but no shipped artifact exists; the only trace in the landing-page build is a cross-reference in [[2026-05-04-mac-product-shape-decisions]]'s sibling build spec. So MAC's free tier is currently an email-gated document in a channel RDCO controls, not a free install in the channel where the buyers already are.
- The vault's own lead-magnet conversion band is opt-in rate, not purchase rate, and people routinely conflate the two. [[2026-05-18-mac-matrix-downloadable-template-status]] logs niche checklists and self-assessments at 25-40% opt-in versus 5-10% for a generic PDF. That is stage one of a two-stage funnel. Stage two, list-to-purchase, sits at 1-5% for a first course launch and 5-8% with heavy warm-up per [[2026-08-04-sanity-check-list-size-mac-revenue-bar]], with beehiiv's platform-wide median at 0.62%.
- Freemium's theoretical case in the vault is a risk-reduction argument, not a volume argument. [[2026-04-03-reforge-hidden-freemium-advantage]] frames the purchase decision as Price < Potential Value x Probability of Realizing It, and argues free usage raises the probability term, which raises willingness to pay and post-conversion retention. It also names the exact failure mode this question is asking about, unresolved: "What's the freemium anti-pattern where the free tier is so good nobody upgrades?"
- MAC has zero sales, so RDCO has no baseline to cannibalize. [[2026-06-20-bookstores-for-agents-2026-live-comps]] catalogues a marketplace layer racing take-rates toward zero before volume exists, and the MAC ledger records 0 sales / $0 revenue at pre-launch. Cannibalization is a risk you can only run once you have paid demand to divert.
What the web says
- Nobody in this ecosystem publishes the number. Three targeted searches and two fetches surfaced no free-to-paid conversion rate, install-count ratio, or funnel metric for any paid Claude skill, skill pack, MCP server, or agent marketplace. This repeats the parent scan's negative result on refund rates. The category has no published funnel data at all.
- Agensi states free and paid skills are functionally identical, which is the clearest cannibalization setup on record. Its free-skills page says "All free skills on Agensi are the same as paid skills, security-scanned, reviewed, and maintained by creators. The only difference is the price tag" (agensi.io). No install counts, no catalog-size comparison, no stated rationale for the dual tier. A free/paid boundary drawn on price alone rather than on capability supplies no upgrade trigger.
- General freemium benchmarks are the nearest quantified analogue, and they are bimodal. A January 2026 survey of 200 B2B software products ($1-10M ARR, $50-249 ARPU, 6-month conversion window) reports an overall median of 8%, with freemium products splitting 25% below 2.5%, 29% in the 2.5-7.5% band, and 25% reaching 10-15% (Growth Unhinged). The section comparing self-serve to sales-assisted is paywalled and was not retried.
- In the closest structural analogue, plugin ecosystems, feature-gated freemium converts at roughly 2.1%, rising to 7.4% with active product-led tactics. Freemius cites case-level lifts: Panto AI 0.9% to 3.6% over nine months, Suped +46% over six months with median upgrade time falling from 41 to 26 days, and DataNumen lifting upgrades 19% by moving free users to email-only support (Freemius).
- No source found produces evidence that a generous free tier cannibalizes paid revenue. Freemius explicitly attributes freemium failure to weak upgrade paths and badly timed prompts rather than to the free tier undercutting demand, and prescribes gating "after a user has experienced clear value." Daydream, covering the same ground, reports 1-5% conversion for broad-market tools and 5-15% for tightly targeted high-intent apps and offers no cannibalization data either (daydream). The absence is consistent across every source, which is weak evidence of absence but not none.
- The one adjacent operator behavior the vault already captured points the same way. Ship30for30 ran a free lead-magnet-idea-generator skill in the PS of a sales email whose actual ask was a paid bootcamp waitlist at 700+ signups ([[2026-05-05-ship30for30-claude-code-marketing]]). The free skill is a capture device attached to a paid motion, with no published conversion figure. That is the modal pattern.
Convergences and contradictions
- Convergence: every quantified source blames the boundary, not the abundance. Freemius, daydream, and the vault's own [[2026-04-03-reforge-hidden-freemium-advantage]] independently locate the failure at the gate design and its timing rather than at free-tier size. The practical translation is that a free skill kills the paid tier only when it delivers the same job at the same completeness, which is precisely what Agensi's price-only distinction does and what CLSkills' unverified-community versus tested-and-supported split does not.
- Contradiction with the question's own premise: this is not a freemium market, it is a commons. Freemium cannibalization requires the same vendor's free good substituting for its paid good. In this ecosystem the 2,810-skill free catalog and Anthropic's zero-cost publishing surface belong to nobody RDCO competes with commercially. That is exogenous abundance, not a self-inflicted free tier, and the freemium benchmarks above do not measure it. The right question for MAC is not whether to risk a free tier, it is whether a free skill can win attention inside a commons that is already saturated and free.
- Contradiction of measurement kind: the benchmark band and MAC's product shape do not match. Every conversion figure found is subscription SaaS measured over a 6-month window with recurring upgrade triggers. MAC is a one-time $350 purchase per [[2026-05-14-mac-pricing-intent]]. Applying a 2-5% freemium rate to it silently imports an upgrade mechanic MAC does not have, the same category mismatch [[2026-08-04-sanity-check-list-size-mac-revenue-bar]] flagged when 1,000-True-Fans arithmetic was applied to a one-time sale.
Synthesis for RDCO
The honest answer is that no measurable public evidence exists, and RDCO should stop waiting for it. Two independent scans, seven months of the category's life, and every operator page fetched have produced exactly zero published free-to-paid conversion numbers for installable agent skills. Agensi publishes seller earnings but not funnel metrics; CLSkills publishes prices but not installs; explainx publishes heuristics but not observations. This is the same shape as the refund-rate finding in the parent scan. The category is too young and too small for anyone to have a denominator worth publishing, and the operators with one have a commercial reason to keep it. Any number RDCO puts in a MAC forecast for free-skill conversion is borrowed from subscription SaaS and should be labelled unverified wherever it appears.
The cannibalization half of the question resolves NO on the available evidence, but for a reason that reframes the decision. Nothing found supports free abundance suppressing paid demand; the consistent finding is that free tiers fail by never producing an upgrade trigger. Agensi's own words are the cleanest illustration on record: when the only difference between free and paid is the price tag, there is no reason to cross the gate. That is a boundary-design failure that happens to look like cannibalization. MAC's boundary is already drawn in a defensible place per [[2026-07-23-executable-agentic-info-product-pricing-scan]]'s recommendation, a scoped /dq audit free, with the matrix methodology, test-generation grammar, and triage runbook paid. The risk to manage is not that the free audit is too good; it is that it is too disconnected, delivering a finding the buyer cannot act on without a next step that is visible and priced.
The sharper finding this scan adds is a channel mismatch the parent brief missed. RDCO is not free-surface-less in intent, it is free-surface-less in the ecosystem. [[2026-05-18-mac-matrix-downloadable-template-status]] shows the planned free artifact is an email-gated matrix template, a document in a channel RDCO owns and must drive traffic to. Every operator in the parent scan instead puts free artifacts where the buyer already searches and installs: the marketplace, the curl one-liner, the plugin index. Those two motions are not substitutes. The document captures an email from someone RDCO already reached; the free install borrows the commons' distribution and yields a qualified user who has run RDCO's code against their own warehouse. Given [[2026-08-04-sanity-check-list-size-mac-revenue-bar]]'s finding that the Sanity Check list is uninstrumented and small, the owned-channel lead magnet is the weaker of the two, because its stage-one 25-40% opt-in rate applies to a denominator RDCO does not have.
The decision this makes actionable: ship the free /dq skill as the instrument, not as the funnel. With MAC at 0 sales, its first hundred installs and first hundred sales are the category's first funnel dataset in either direction, exactly as the parent scan argued for refunds. That reframes the free skill from a marketing bet needing external justification into a measurement device: publish it to the public marketplace, instrument install count, activation (audit actually run against a real project), and click-through to the paid page, and RDCO learns its own conversion rate in weeks instead of waiting for an operator who will never publish theirs. The one design constraint the evidence does support is Freemius's timing rule, gate after value is realized. In MAC's shape that means the free audit should surface real findings against the user's own warehouse and then stop at the point where the fix requires the matrix, rather than being a trimmed demo that produces nothing worth acting on.
Why this is in the vault
This closes open follow-up #5 of [[2026-07-23-executable-agentic-info-product-pricing-scan]] with a negative result, which directly unblocks the pre-launch decision of whether MAC ships a free marketplace skill before the $350 paywall: the answer cannot be evidence-led because no evidence exists, so the free skill should be scoped and instrumented as MAC's own first conversion experiment rather than debated against benchmarks borrowed from subscription SaaS.
Open follow-ups
- What are the actual install counts on any free skill in the public Claude marketplace, and is install-count telemetry even exposed to a publisher? If not, MAC's free skill needs its own callback instrumentation designed in before publish, not after.
- Does the Anthropic plugin marketplace permit a free skill whose output links to a paid off-platform product, and are there listing or policy constraints on commercial CTAs inside a free skill?
- What is the observed activation rate of a free installed skill, meaning installs that actually get run at least once against real data? Install-to-activation may be the binding leak, and no source in this scan measured it.
- Has any operator (CLSkills, Agensi, tonsofskills) published or privately shared a free-to-paid figure since July 2026, or would direct outreach to a smaller operator plausibly get one?
- Does a one-time-purchase product behave differently from a subscription on free-to-paid conversion at the same funnel stage? Every benchmark found assumes a recurring upgrade trigger MAC does not have.
- If the free
/dqaudit surfaces real findings, what fraction of users can act on those findings without buying MAC? That self-sufficiency rate, not catalog abundance, is the only cannibalization mechanism this scan found credible.
Related
- [[2026-07-23-executable-agentic-info-product-pricing-scan]]
- [[2026-05-18-mac-matrix-downloadable-template-status]]
- [[2026-08-04-sanity-check-list-size-mac-revenue-bar]]
- [[2026-04-03-reforge-hidden-freemium-advantage]]
- [[2026-05-14-mac-pricing-intent]]
- [[2026-06-20-bookstores-for-agents-2026-live-comps]]
- [[2026-05-18-low-ticket-dev-tool-launch-friction]]
- [[2026-05-04-mac-product-shape-decisions]]
- [[2026-05-05-ship30for30-claude-code-marketing]]
- [[2026-06-16-mac-challenge-vs-drip-completion-format]]
Sources
- Vault: [[2026-07-23-executable-agentic-info-product-pricing-scan]] (
~/rdco-vault/06-reference/research/2026-07-23-executable-agentic-info-product-pricing-scan.md) - parent brief, source of the question - Vault: [[2026-05-18-mac-matrix-downloadable-template-status]] (
~/rdco-vault/06-reference/research/2026-05-18-mac-matrix-downloadable-template-status.md) - lead-magnet opt-in bands, unshipped status of the free artifact - Vault: [[2026-08-04-sanity-check-list-size-mac-revenue-bar]] (
~/rdco-vault/06-reference/research/2026-08-04-sanity-check-list-size-mac-revenue-bar.md) - list-to-purchase conversion bands, instrumentation gap - Vault: [[2026-04-03-reforge-hidden-freemium-advantage]] (
~/rdco-vault/06-reference/2026-04-03-reforge-hidden-freemium-advantage.md) - purchase-decision formula, unresolved anti-pattern question - Vault: [[2026-05-14-mac-pricing-intent]] (
~/rdco-vault/01-projects/mac/2026-05-14-mac-pricing-intent.md) - $350 one-time, founder-confirmed - Vault: [[2026-06-20-bookstores-for-agents-2026-live-comps]] (
~/rdco-vault/06-reference/research/2026-06-20-bookstores-for-agents-2026-live-comps.md) - marketplace layer, take-rate race - Vault: [[2026-05-18-low-ticket-dev-tool-launch-friction]] (
~/rdco-vault/06-reference/research/2026-05-18-low-ticket-dev-tool-launch-friction.md) - format-over-price finding - Vault: [[2026-05-04-mac-product-shape-decisions]] (
~/rdco-vault/01-projects/mac/2026-05-04-mac-product-shape-decisions.md) - install flow, category framing - Vault: [[2026-05-05-ship30for30-claude-code-marketing]] (
~/rdco-vault/06-reference/2026-05-05-ship30for30-claude-code-marketing.md) - free skill as lead magnet attached to a paid motion - Vault: [[2026-06-16-mac-challenge-vs-drip-completion-format]] (
~/rdco-vault/06-reference/research/2026-06-16-mac-challenge-vs-drip-completion-format.md) - executable-format completion argument - Vault:
~/rdco-vault/09-bet-stacks/mac-pnl-ledger.md- 0 sales / $0 revenue, pre-launch snapshot 2026-05-18 - Agensi - Best Free Claude Code Skills in 2026 - fetched 2026-08-18; free/paid parity statement, no funnel data (negative result)
- Growth Unhinged - The 2026 free-to-paid conversion report - fetched 2026-08-18; 200 B2B products, Jan 2026, median 8%, freemium distribution bands. Self-serve vs sales-assisted section PAYWALLED, not retried
- Freemius - How to build a freemium pricing strategy that converts - fetched 2026-08-18; 2.1% feature-gated baseline to 7.4% with PLG tactics, named case lifts, explicit no-cannibalization framing
- daydream - Freemium Conversion Rate Benchmarks - fetched 2026-08-18; 1-5% broad-market, 5-15% high-intent targeted, no cannibalization data