PlayProof — audit drafter for independent playground inspectors

1. One-liner

Reads a CPSI’s site photos and voice notes, then writes the 40-page ASTM F1487 audit before they leave the lot.

2. Trend signal — why now?

The Certified Playground Safety Inspector (CPSI) audit is one of the most paperwork-heavy field jobs in property services. A full audit runs up to 40 pages — a line-by-line pass/fail against ASTM F1487 and CPSC handbook items, with photos, risk ratings, and corrective recommendations for every measured opening, every fastener, every surfacing depth (playgroundprofessionals.com). The inspection takes an hour. The write-up takes the evening.

What changed in the last 12 months: multimodal vision models got cheap and accurate enough to look at a photo of a guardrail gap, a rusted weld, or a worn S-hook and describe the defect in inspector language — and an LLM can now cite the exact ASTM clause and draft the corrective recommendation. That stack didn’t exist at a usable price two years ago.

Meanwhile the money is moving. Playground-inspection operators report that “variable costs for essential inspection tools and software hit 130% of revenue in 2026” (financialmodelslab.com) — the segment is software-hungry and margin-stressed. Incumbents (Playground Guardian/Park Protector, ParkZapp, OpenGov, GoAudits, CityReporter) are funded and expanding modules into trails, restrooms, sports fields. But every one of them is a checklist app sold to the park department doing its own routine self-inspection — Playground Guardian literally markets itself as “a fraction of what it would cost to hire a CPSI” (playgroundguardian.com). Nobody is arming the independent CPSI on the supply side — the person who actually sells the certified annual audit and has to write the liability narrative.

Provenance:
  - Signal 1 (demand): CPSI audits run up to 40 pages, line-by-line ASTM F1487/CPSC pass-fail with photos and corrective recs; report write-up is the bottleneck — https://playgroundprofessionals.com/playground/inspection/playground-inspections-and-audits — 2026-06-20
  - Signal 2 (feasibility): Cheap multimodal vision + LLM can now read defect photos (entrapment gaps, corrosion, missing fasteners, surfacing depth) and draft ASTM-cited narrative — tech-unlock, viable only since ~2024 — 2026-06-20
  - Signal 3 (economic): Playground-inspection operators report inspection tools/software hit 130% of revenue in 2026; incumbents (Playground Guardian, ParkZapp, OpenGov) funded and expanding — https://financialmodelslab.com/blogs/operating-costs/playground-safety-inspection — 2026-06-20
  Category: Underserved niche

3. The opportunity

The audit is priced at $300–$800 per playground structure per year (playgroundguardian.com), recurring, and liability-driven — a kid gets hurt, the certified audit is the first document the lawyer subpoenas. So demand is non-discretionary. The constraint on an independent inspector’s income isn’t sales; it’s throughput. They can inspect six sites in a day but they can only write three reports a night. The write-up — transcribing field notes, matching each finding to the right ASTM clause, formatting the photo log, drafting plain-English corrective recommendations — is a 2–4 hour after-hours grind per site.

PlayProof collapses that to ~15 minutes of review. The inspector walks the playground exactly as they do today — photographing defects, dictating findings — and by the time they’re back in the truck the draft audit is sitting in their inbox: every photo placed, every defect classified by ASTM clause and priority, every recommendation written in their house style. They review, correct, sign. The product doesn’t replace the CPSI’s judgment or signature — it replaces the 3-hour typing job that judgment currently gets buried under.

The incumbents can’t pivot here easily: their entire business model and sales motion is “skip the expensive human inspector, let your maintenance guy run a checklist.” Selling a tool that makes the expensive human inspector 2× more productive is the opposite positioning and the opposite customer.

4. Target market

• Primary customer: Independent CPSIs and small (1–8 person) third-party playground inspection firms in the US who sell certified annual/semi-annual audits to schools, daycares, municipalities, HOAs, churches, and apartment complexes.
• Why they buy: In their words — the inspection is the easy part, the report is what kills the evening. They’re capacity-capped by write-up time, so every hour saved on reporting is a billable audit they can add. A solo CPSI charging $400/audit who reclaims 3 hours/site can do 1–2 more sites a day.
• Rough TAM reasoning: The NRPA maintains a public CPSI registry of currently-certified inspectors (nrpa.org). The credential spans park-department staff (who self-inspect and won’t buy this) and the independents/contractors who sell audits (who will). A conservative serviceable slice is low-thousands of revenue-generating independent inspectors plus small inspection firms — each doing 150–400 audits/year. At a few thousand buyers and $100–150/mo, the ceiling is comfortably in the low-tens-of-millions ARR — small enough to be invisible to the checklist-app incumbents, big enough for a bootstrapped operator.
• Why now for them: Liability scrutiny and insurance documentation requirements keep rising; AHJs and risk pools increasingly demand a current certified audit on file. Volume is up, the per-report burden is unchanged, and for the first time the write-up can actually be automated without losing the inspector’s voice.

5. Product sketch (MVP)

• Field capture: mobile flow where the inspector photographs each defect and dictates a voice note (“S-hook opening on the left swing exceeds spec, recommend replacement”) — no typing on site.
• Vision defect read: model reads each photo and proposes the defect type, severity, and the ASTM F1487 / CPSC handbook clause it maps to; inspector confirms or overrides with one tap.
• Auto-drafted audit: generates the full multi-page report — executive summary, equipment inventory, prioritized findings table (high/moderate/low hazard), photo log with captions, and per-finding corrective recommendations in plain English.
• House-style memory: learns the individual inspector’s phrasing and recommendation language so the draft reads like their reports, not a generic template.
• Standards engine: ships with the ASTM F1487 dimension tables and CPSC Public Playground Safety Handbook checklist baked in, so clause citations are correct and current.
• Client-ready output: branded PDF with the inspector’s logo, CPSI number, and signature block (NFPA-style named-inspector requirement analog for playgrounds — the named CPSI must own the report).
• Re-inspection diff: on the next annual audit, auto-flags which prior findings were corrected and which persist — a feature municipalities and risk pools love.

6. AI angle — what’s load-bearing

Remove the AI and there is no product — it’s back to a checklist app, which already exists and which I’m deliberately not building. The AI does two genuinely hard things: (1) reads field photos and converts “rusty bolt, gap looks too wide” into a correctly-classified, ASTM-cited defect with a severity rating, and (2) drafts the narrative — the prioritized findings, the corrective recommendations, the plain-English explanation a school principal can act on — in the inspector’s own voice. The checklist incumbents make the human type all of that. The whole wedge is that vision + language models now do the 3-hour part. That’s load-bearing, not decoration.

7. Localization angle (if any)

N/A — this is a US-first play. ASTM F1487 and the CPSC handbook are US standards and the CPSI credential is NRPA-specific, so the moat is the standards-knowledge, not a language or payment-rail localization. There’s a credible month-12 expansion into Canada (CSA Z614) and the EU (EN 1176), each a different standards pack on the same engine — but those are later sequencing, not a launch wedge.

8. Business model — path to $1M–$5M ARR

• Pricing: $99/mo solo inspector (unlimited audits), $149/mo “firm” tier for up to 5 inspectors with shared client history and re-inspection diffs. Optional $49/mo seasonal pause for inspectors who go quiet in winter (reduces churn).
• ACV: ~$1,100–1,500/yr solo; ~$2,000–4,000/yr for small firms.
• Rough math to $1M ARR: ~750 solo inspectors at $99/mo × 12 = $890K, plus ~50 firms at ~$2,400/yr = $120K → ~$1M. That’s a single-digit share of the revenue-generating independent CPSI population.
• Rough math to $5M ARR: ~3,000 solo + ~400 firms, or the same base plus the Canada/EU standards packs and a per-audit usage upsell for very high-volume firms. Requires becoming the default tool the CPSI course recommends.
• Expansion path: ACV grows via (a) firm seats, (b) the re-inspection diff and client-portal features municipalities will pay extra for, (c) adjacent inspection verticals the same vision-plus-standards engine unlocks (splash pads, fitness-court equipment, fall-surfacing GMAX/HIC reporting).

9. Go-to-market wedge — first 100 customers

• Mine the NRPA CPSI registry. It’s public (apps.nrpa.org/CPSI_Registry). Filter for inspectors whose listing or website shows they’re independents/contractors (not “City of X Parks Dept”). That’s a named, finite list of exactly the right buyers — cold email a 90-second Loom showing a real photo turning into a finished page. Expect 3–6% reply on a list this targeted.
• Sponsor / present at CPSI re-certification courses. Every CPSI must recertify every 3 years; state recreation & park associations run the courses constantly. Get 10 minutes at the front of the room demoing a report drafting itself. This is where the entire buyer population physically gathers.
• The playground-inspection subreddits/Facebook groups and the IPSI/IPEMA communities. Independent inspectors swap war stories about report nights there. Post the before/after (raw photos in → finished audit out), offer a free first-audit draft.
• Partner with 1–2 inspection-franchise networks whose franchisees do certified audits and currently hate the paperwork — one network relationship can seed dozens of seats.

10. Build complexity — justification

Medium. The web/mobile capture flow, PDF generation, and auth are off-the-shelf. The real work is (a) encoding the ASTM F1487 dimension tables and CPSC handbook checklist into a reliable standards engine so citations are never wrong, and (b) tuning the vision model on real defect photos so the defect-classification is trustworthy enough that inspectors trust the draft. Both need domain partnership with a practicing CPSI and a few hundred labeled report/photo pairs. Realistic v1 for a technical founder plus a CPSI advisor: 3–4 months.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Tool drafts; the certified CPSI reviews and signs. We don’t certify or inspect.
Ethical — no harm / dark patterns	✅	Keeps the human inspector in the loop; sells productivity, not replacement of judgment.
Market exists (evidence above)	✅	Recurring $300–800/audit market, liability-driven, public buyer registry.
1–5 person team can build this	✅	Technical founder + CPSI advisor, 3–4 months.
Launchable with <$50K / ₹40L	✅	Off-the-shelf stack + inference costs; main cost is the CPSI advisor’s time.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	15/20	Real, recurring, billable-capacity-capping pain — but it’s a nightly grind, not a daily money-on-fire emergency. Solidly above checklist-fatigue, below compliance-deadline panic.
Demand evidence	15	11/15	Hard signals: published per-audit pricing, funded incumbents, public buyer registry, “software at 130% of revenue” cost data. Gap: no direct verbatim CPSI complaint quote sourced yet (registry forums are gated).
Build feasibility	15	11/15	Standard stack + standards engine; main risk is vision accuracy on defect photos and getting labeled training pairs. Pair ships in 3–4 months.
Distribution clarity	15	11/15	Named finite buyer list (NRPA registry) + a captive in-person channel (recert courses). Conversion is the unknown.
Revenue mechanics	15	11/15	Pricing benchmarked to the value (one reclaimed evening = one more billable audit). $1M path needs single-digit % of the independent CPSI base.
Time to first revenue	10	8/10	Self-serve trial; an inspector who runs one real audit through it and likes the draft converts fast. Revenue in 4–8 weeks of launch.
Defensibility	10	6/10	Soft moat: accumulated standards-engine accuracy, per-inspector house-style memory, re-inspection history lock-in. A checklist incumbent could bolt on AI, but they’d be fighting their own anti-CPSI positioning.
Total	100	73/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required — needs solid AI/vision engineering and a practicing CPSI co-founder/advisor to get the standards engine and defect classification right.

Key assumptions to validate (3–5)

1. Assumption: Independent CPSIs spend 2–4 hours of after-hours write-up per audit and would pay ~$100/mo to cut it to ~15 minutes. How to test: 20 phone interviews with independents pulled from the NRPA registry; ask directly how long reports take and what they’d pay.
2. Assumption: Vision models can classify real playground defects (entrapment gaps, corrosion, fastener loss, surfacing wear) accurately enough that inspectors trust the draft instead of redoing it. How to test: collect 200 real defect photos from a partner CPSI, measure classification accuracy against their ground-truth findings.
3. Assumption: Inspectors will trust an AI-drafted, human-signed report without fearing liability exposure. How to test: put the “you review and sign, the standard is cited inline” framing in front of 10 inspectors and a playground-liability insurer; watch for the objection.
4. Assumption: The recert-course channel will let a vendor demo. How to test: contact 3 state rec & park associations running CPSI courses.

Risk flags

1. Incumbent fast-follow: Playground Guardian/ParkZapp could add an AI-draft module. Mitigant: their positioning is explicitly anti-CPSI (“skip the expensive inspector”), so serving the inspector is awkward for them — but it’s still a 12-month risk.
2. Liability perception: inspectors are risk-averse by trade; if they fear an AI draft creates legal exposure, adoption stalls. The “human signs, standard cited” framing must be airtight.
3. Niche size: the revenue-generating independent CPSI population is small. This is a low-tens-of-millions ceiling business, not a rocket — fine for a bootstrapper, wrong for anyone expecting venture scale.

14. Structured verdict

Score:                  73/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (AI/vision) with a practicing CPSI co-founder or paid advisor
Time to revenue:        6–10 weeks from launch
Capital to launch:      $15–30K (mostly the CPSI advisor's time + inference during tuning)
Top 3 assumptions to validate first:
  1. Write-up really costs 2–4 hrs/audit and WTP is ~$100/mo — 20 registry interviews
  2. Vision defect-classification is accurate enough to trust — 200 real-photo accuracy test with a partner CPSI
  3. Inspectors trust AI-drafted/human-signed reports without liability fear — framing test with 10 inspectors + an insurer
Kill criteria:
  - Abandon if <30% of 20 interviewed independents report >90 min/audit write-up time (the pain isn't big enough)
  - Abandon if defect-classification accuracy stays below ~80% after tuning on real photos (inspectors won't trust the draft)
  - Abandon if a checklist incumbent ships a credible inspector-facing AI-draft module before your v1

15. Next step — 1-week validation sprint

• Day 1–2: Pull 40 clearly-independent CPSIs from the NRPA registry. Cold-call/email for 20-minute interviews. One question above all: how long does your average report write-up take, and what would you pay to make it 15 minutes?
• Day 3–4: Recruit one friendly CPSI to send 50 real defect photos with their ground-truth findings. Run them through an off-the-shelf vision model and a prompted ASTM-citation draft. Measure: what % did the model classify and cite correctly?
• Day 5: Go/no-go. Go if ≥60% of interviewed independents report >2 hrs/audit write-up AND ≥70% photo classification accuracy on the raw test. Both are falsifiable. If either misses, the pain or the feasibility isn’t there yet — shelve and revisit when vision models improve.