Padel AI platform — first 90 days operating plan

A first-quarter operating plan for a padel coaching AI is structured around twelve testable hypotheses, a metric tree from the headline number down to weekly leading indicators, and a risk register with named exit criteria. Days 1 to 30 focus on capture-quality validation in a single club; days 31 to 60 add a coach panel and the first paid cohort; days 61 to 90 measure ninety-day paid retention before any second-club expansion.

A 90-day plan earns its title by naming what would stop it on day 30, day 60, and day 90. This document does that for the Padel AI Platform: three phases, one stop memo, one pivot route, twelve hypotheses on the clock.

Voice: third-person professional, capability-conditional. No first-person pronouns. No promised headcount, budget, or timelines beyond what existing evidence supports.

>

Audience: hiring-panel screening interview.

>

Evidence base: the published evidence. Every numeric threshold below cites either a verified URL or carries an "anchor — to be validated" flag.

>

profile assumed: deep product-strategy and research expertise, zero playing-level expertise in padel itself. The plan treats the domain gap as a primary risk and routes mitigations through an advisory coach pool (R-05).

Executive frame

The role would inherit a pre-validated research package that has already done two things the plan must respect.

First, the adversarial review pass (the published evidence) has already pruned the segment list. (plateau-stuck regular), (newly-ranked competitor), and (club coach, business-to-business-to-consumer) carry PASS or PASS-with-caveats verdicts. (club operator) and (injury-aware returner) were KILLED. (travelling enthusiast) was MERGED into . The plan does not re-litigate these calls; it executes against them.

Second, the jobs graph (the published evidence) names the rating-clarity backbone as the critical chain: smartphone match capture → auto-tag → rating delta → defensible rating. Every other Core Job — drill prescription, coach co-pilot, post-match argument settler — is gated on working at smartphone-only fidelity. The plan invests there first.

Third, monetisation (the published evidence) sets a hard exit threshold for the business-to-consumer path: below 4% paid conversion at EUR 7.99/month triggers a pivot to the business-to-business coach-SaaS hedge. The plan sequences a smoke test in Spanish and Russian against this threshold inside phase B.

A roadmap that does not name what would stop it is not a roadmap, it is a wishlist.

Phase A — Days 0–30: Discovery and anchor

A.1 Objectives

• The role would establish a documented mental model of (rating-clarity backbone) and (coach-renewal chain) before touching scope.
• The role would validate or stop the inherited top-10 hypotheses without re-running existing research.
• The role would stand up the jobs-to-be-done interview cohort across the three PASS segments.
• The role would scope a single-club manual MVP loop without launching it.

A.2 Stakeholder map

The role would coordinate with engineering and design as currently staffed. No headcount asks during the first 30 days. The advisory coach pool is the mitigation for the domain gap (R-05) and is treated as a paid retainer secured in week 1.

A.3 jobs-to-be-done interview cohort

Total: 20 interviews, all conducted in the first 28 days, drawn exclusively from PASS segments per the published evidence.

Explicit exclusions (per Phase 06 adversarial review):

• (club operator, business-to-business): killed — venue-size descriptor in disguise.
• (travelling enthusiast): merged into .
• (injury-aware returner): killed — life-situation label without verified switch trigger.

A.4 Evidence audit

The role would review the inherited research package (peer cards, jobs graph, adversarial review, monetisation, go-to-market, capability map, market sizing) and produce an annotated hypothesis backlog. Hypotheses are tagged keep / stop / promote-to-experiment. Killed hypotheses move to a graveyard log; reactivation requires new evidence, per the project operating contract.

A.5 Single-club pilot scoping (not yet running)

Scoping deliverables only — no execution in phase A:

• Pilot brief: target club profile (Madrid or Barcelona, 4–8 courts, 1 coach champion).
• Manual loop diagram: capture → tag → recap → drill → next session.
• Reliability thresholds for graduating each step out of manual ops (see B.2).
• Privacy / consent flow draft (GDPR baseline; 152-FZ flagged for Russian-language path).

A.6 Output artifacts (end of phase A)

1. Validated AJTBD canvas per PASS segment.
2. Prioritised hypothesis backlog (this document, section D).
3. PRD outline for the manual MVP loop.
4. Instrumentation spec covering match-upload, share, partner-invite, and recap-acknowledgement events.
5. Stakeholder map and cadence document.
6. Risk register v1 (this document, section F).

Phase B — Days 31–60: Manual MVP and pricing test

B.1 Objectives

• The role would run the manual measurement → insight → training loop in one club end-to-end without productising any step.
• The role would run the WTP smoke test in Spanish and Russian against the EUR 7.99 anchor.
• The role would begin the first 4-week retention cohort (week 5–8 of pilot).
• The role would produce the first stop/keep call against segment-level evidence.

B.2 Manual MVP loop

Pilot clubs film matches on a tripod-mounted phone. The OSS baseline (Joao-M-Silva/padel_analytics) auto-tags shots. The plus advisory coach review three matches per pilot user. A recap card and a two-drill prescription get sent to the player and copied to the coach before the next booking.

Instrumentation events to land in the spec:

• match_uploaded, match_tagged, recap_sent, recap_opened, drill_acknowledged, coach_outcome_tag.

Out of scope for the 90-day window: pair-share and partner-invite mechanics. The manual MVP loop in 03_mvp_loop_design.md is single-player → coach. Any pair-viral surface (share_clicked, partner_invited, partner_activated) requires a Stage 2.5 build that is not part of this pilot. Pair-share viral lift is logged as a post-pilot exploration item, not a 90-day hypothesis.

Reliability gates to graduate manual steps into product (used in phase C):

B.3 WTP smoke test (Spanish + Russian)

Two landing pages — Spanish and Russian — each with a fake-door "request a padel rating" CTA priced at the geo-localised tier per the published evidence:

Anchor threshold (stop / keep). Internal exit threshold synthesised from the Foundry CRO 2026 5–9% free-to-paid range (verbatim quote sits in the published evidence at the F4 funnel row): at EUR 7.99/month, ≥8% paid conversion within three matches is keep; below 4% triggers the hedge model (business-to-business coach SaaS). Source for the underlying range: foundrycro.com/blog/cac-benchmarks-2026 — the EUR 7.99 anchor and the three-match window are internal product policy, not a Foundry quote.

Secondary metrics:

• CTR on padel community channels: ≥ 6% (per Phase 06 stop test).
• Direct-traffic share of beta sign-ups: ≥ 40% within 8 weeks (per the published evidence#).

stop decision logic. If WTP < 4% AND direct-traffic share < 25% by week 8, the hedge model (business-to-business coach SaaS at EUR 19–29 per coach per month, anchored on CoachLogic and Hudl Assist tiers) becomes primary. The role would draft the pivot memo for founder signoff.

B.4 First retention cohort

Starts week 5; runs 4 weeks. Mixed + cohort. Primary metric: weekly active uploaders with ≥ 1 acknowledged insight per week. Leading indicator: recap-acknowledgement rate by week 2 (anchor — to be validated, see metric tree).

B.5 Output artifacts (end of phase B)

1. Pilot weekly review packet (one-pager + recap exemplars).
2. Conversion-curve dashboard (free → activated → paid) updated weekly.
3. First stop/keep memo on segments and pricing tier.
4. Retention cohort dashboard (engineering-staffed; no new tooling spend assumed).

Phase C — Days 61–90: Productisation and channel

C.1 Objectives

• The role would pull manual steps that hit reliability thresholds into product surface.
• The role would open a second club pilot to test transferability of the recipe.
• The role would run the first paid-acquisition test against a capped budget.
• The role would deliver a decision memo on the rating-platform path.

C.2 Productisation targets

Steps that fall short of their gate stay manual into the next 90-day window.

Post-pilot exploration (not a 90-day target). A pair-level surface that drives ≥1.5x sharing rate vs the single-user baseline (the published evidence#) sits in the post-pilot exploration backlog. The manual MVP in 03 does not generate the data needed to evaluate this gate inside 90 days; gating it inside this plan would be a sequencing error.

C.3 Second club pilot (Italy)

Target: Milan or Rome, to stress-test localisation. stop metric: if second-club week-4 cohort retention < 60% of first-club benchmark, the manual loop is over-fit to the original club; the role would pause Italy expansion and document the gap before any Italian paid-acquisition spend.

C.4 First paid-acquisition test

Two channels, capped budget. The role would propose the cap at the lower bound of CAC × intended cohort size; exact figure pending founder signoff. No fixed budget is committed in this plan.

stop metric: paid CAC > 3x organic CAC after 4 weeks (per the published evidence#).

C.5 Rating-platform decision memo

Two paths come into the memo:

• Option A — FIP-affiliated organiser integration (the published evidence#). Partner-dependent, two LOIs targeted in 6 weeks.
• Option B — Derived-only rating. Ship the rating model on smartphone footage; expose a public profile that links out from Playtomic / MATCHi.

Decision criteria:

• ≥ 2 LOIs by week 12 → pursue Option A in parallel with B.
• 0 LOIs by week 12 → default to Option B; revisit federation integration after the rating reaches ≥ 5,000 active rated players in ≥ 3 cities (expansion threshold per the published evidence).

C.6 Output artifacts (end of phase C)

1. Roadmap for days 91–180.
2. go-to-market dashboard (channel mix, CAC, payback period, direct-traffic share).
3. Decision memo on rating-platform path.
4. Productisation log: which manual steps graduated, which stayed manual, why.

D. Hypothesis backlog

Format: H-XX | hypothesis | metric | threshold | timeline | source/data gap.

E. Metric tree

North-Star metric. Rated players completing 90-day retention with ≥ 1 weekly insight. This combines retention (subscription health) and product engagement ( insight delivery). No public peer benchmark available; it is an anchor — to be validated in week 13 cohort.

Input metrics

Output metrics

Guardrail metrics

F. Risk register

Five risks ranked by probability × impact. Each carries a mitigation, an owner pattern, and a trigger to escalate.

R-01 — OSS capture pipeline cannot reach acceptable accuracy (rank 1)

• Probability × impact: medium × critical → HIGH.
• Description: Joao-M-Silva/padel_analytics cannot reach acceptable accuracy on representative club video. Per the published evidence#, the stop signal is "OSS pipeline cannot reach acceptable accuracy on a representative club video sample."
• Trigger to escalate: ≥ 25% of pilot matches require manual re-tagging at week 4; OR shot-classification accuracy < 70% on a 50-match validation set.
• Mitigation: The role would run accuracy validation in week 1 of phase A on existing club footage. If below threshold, productisation work pauses and the path routes to engineering for OSS-fork or model-swap before pilot launch.
• Owner pattern: PM coordinates; engineering owns accuracy work; advisory coach owns ground-truth labelling.

R-02 — WTP and direct-channel both miss thresholds (rank 2)

• Probability × impact: medium × critical → HIGH.
• Description: WTP test fails the exit threshold (< 4% paid conversion) and the direct-channel audience does not materialise (< 25% direct-traffic share). Either alone is a setback; both together stop the business-to-consumer primary thesis.
• Trigger to escalate: WTP < 4% AND direct-traffic share < 25% by week 8 of phase B.
• Mitigation: A pre-drafted pivot memo to the hedge model (business-to-business coach SaaS at EUR 19–29 per coach per month) would be ready for founder signoff before week 8. Segment recruitment expanded to in parallel.
• Owner pattern: PM owns memo; founder owns decision; ops owns coach recruitment.

R-03 — Coach-renewal chain collapses on tool-refusal (rank 3)

• Probability × impact: medium × high → HIGH.
• Description: Coaches refuse any tool the student also touches. Per the published evidence#: "Coaches sometimes refuse any tool the student also touches; if that refusal dominates, the segment splits into coach-only vs business-to-business-to-consumer and the latter shrinks."
• Trigger to escalate: ≥ 5 of 8 coaches in pilot decline shared-tool model; coach-driven handoff rate < 5% by week 8 (per the published evidence#).
• Mitigation: Split the product surface — coach-only dashboard becomes a separate SKU, narrows to coach-only, and the business-to-business-to-consumer distribution claim retires with a adversarial review-style memo.
• Owner pattern: PM owns SKU split; advisory coach pool owns the refusal-pattern interview; design owns coach-only surface.

R-04 — Privacy / regulatory drag on the Russian-language path (rank 4)

• Probability × impact: low-to-medium × high → MEDIUM-HIGH.
• Description: 152-FZ on-device processing requirements block the direct channel and the WTP test. Per the published evidence#, on-device CV is requires_capital band and not yet ready.
• Trigger to escalate: External counsel flags hosting pattern as non-compliant; OR the Telegram bot path is rejected for the channel.
• Mitigation: On-device inference path scoped in phase A as a parallel track. The Russian-language WTP variant runs as a Telegram-flow CTA without external links if the web path is gated. Counsel review booked for week 2.
• Owner pattern: PM owns scoping; engineering owns on-device pipeline; ops owns Telegram bot operation; external counsel owns sign-off.

R-05 — domain gap propagates into product (rank 5)

• Probability × impact: medium × medium → MEDIUM.
• Description: Zero playing-level expertise in padel leads to flawed interview design or bad recap content. Advisory coach pool fails to scale.
• Trigger to escalate: Two consecutive recap exemplars rejected by advisory coach pool; OR jobs-to-be-done interviews surface frame-mismatch language (using terminology no rated player recognises).
• Mitigation: Advisory coach pool secured as paid retainer in week 1 of phase A. The sits in on coach-led drill sessions in weeks 2–3 to absorb shot taxonomy. Every recap card is reviewed by the coach pool until the reliability gate is met.
• Owner pattern: PM owns retainer setup; advisory coach pool owns content review; founder owns ratification.

Definition of done

Verified URL register (≥ 6 distinct sources)

1. github.com/Joao-M-Silva/padel_analytics — OSS feasibility anchor for capture-then-tag pipeline.
2. foundrycro.com/blog/cac-benchmarks-2026 — CAC benchmarks for organic, paid, ASO, community, affiliate channels.
3. padelfip.com/world-padel-report-2025 — Italy court count growth (12.9% YoY 2025); Spain 17,300 courts; SAM basis.
4. playtomic.com/global-padel-report — Active player benchmarks; SOM low-end basis.
5. fep.es — Spanish federation chapter (coach recruitment, ).
6. fitp.it — Italian federation chapter (coach recruitment, ).
7. hudl.com/products — Coach co-pilot analogue from adjacent sports.
8. strava.com/premium — business-to-consumer subscription pricing anchor (USD 11.99/month).
9. insightaceanalytic.com/report/ai-in-fitness-and-wellness-market/2744 — Adjacent AI-fitness ceiling sizing.
10. padelytics.ai — Counter-example: existing match-level analysis vendor selling to plateau players.