RxCatch — remake catcher for optical dispensaries

1. One-liner

AI cross-checks the doctor’s Rx against your lab order and flags every error before it ships — killing remakes.

2. Trend signal — why now?

The optical industry quietly eats a 15% remake rate on spectacle lens orders — well-run practices get to 5% or below, but the average shop is at 15% (Optogrid / Eye Care Leaders benchmark). Each remake costs $75–$250 all-in (replacement lens $25–150, two-way shipping $16–36, plus 45–90 minutes of optician labor). A mid-sized practice doing 200 jobs/month absorbs over $32,000 a year in direct remake cost — before you count the patient who now thinks you’re sloppy.

A big, named, controllable slice of that is transcription error: misreading the doctor’s prescription, the dreaded “90-off” transposition of cylinder and sphere, entering the wrong axis, or dropping the PD / segment height entirely. The trade itself calls these out as “unambiguous in cause” — the lab built exactly what you ordered; what you ordered was wrong.

What changed: vision-LLMs in 2025–26 can finally read a handwritten or faxed prescription reliably, and the Rx is a tiny, bounded data shape (sphere / cyl / axis / add / prism / PD / seg). That makes automated read-and-validate economical per order for the first time. The optical-software market is moving — Optogrid (digital PD), SpecCheck (Rx order entry), Jelo, Ocuco, VisionWeb all sell into this workflow — but every one of them either measures PD or speeds up clean data entry. None of them sit between the original doctor’s Rx and the transmitted order as a validation gate.

Provenance:

• Signal 1 (Demand): 15% industry-avg remake rate, $75–250 per remake, $32K/yr at a 200-job/mo practice; transcription/axis/transposition named as controllable causes — https://www.optogrid.com/blog/eyeglass-remake-rate-cost/ — 2026-06-18
• Signal 2 (Feasibility): Lab-side complaints confirm “MISSING INFORMATION” is the most recurring, costly intake problem; bounded Rx data shape now readable by vision-LLMs — https://opticiannow.com/2020/05/19/opticians-and-labs-working-together-to-minimize-errors-from-mr-0ptix/ — 2026-06-18
• Signal 3 (Economic): ~22,900–29,000 US optometry practices, $18.2B revenue, sales growing 6%/yr; multiple funded vendors (Optogrid, SpecCheck, Ocuco, VisionWeb) selling remake-reduction adjacent software — https://www.ibisworld.com/united-states/industry/optometrists/1560/ — 2026-06-18 Category: Workflow automation

3. The opportunity

The remake is the optical industry’s oldest leak, and the part nobody’s plugged is the handoff between the doctor’s prescription and the lab order. Today a human optician reads the Rx, reads the frame/measurement data, and re-keys it into a lab portal or LMS while a patient waits. Best practice is “two-point verification” — but the standard verification checks the finished lens against the job ticket, not against the original Rx. So a transcription error made at order entry is invisible until the patient puts the glasses on and complains. Then it’s a $75–250 remake and a burned trip.

The incumbents are aimed elsewhere. Practice-management suites (Ocuco, RevolutionEHR) and lab portals (VisionWeb, Eyefinity) assume clean structured input and optimize production. Measurement tools (Optogrid) attack the PD slice. Order-entry tools (SpecCheck) make typing faster but still trust the typist. RxCatch is the thin, focused gate that reads the actual doctor’s Rx, reads the order about to be sent, and refuses to let a transposed cyl, an out-of-tolerance axis, a sign flip, or a missing PD/seg-height through. Not a suite. A seatbelt.

4. Target market

• Primary customer: Independent optometry practices and optical dispensaries in the US — single-location to ~4-location groups, 100–400 lens jobs/month, 1–3 opticians keying lab orders. Owner-optometrist or office manager is the buyer.
• Why they buy: “At 10% my lab fires the customer!” (LENNY, OptiBoard). Remakes are a known, hated, partly-controllable cost line. Every owner can quote roughly what a remake costs them and roughly how often it happens. They’re already told to “track them and reduce them to zero” — they just have no tool that catches the error before it ships.
• Rough TAM reasoning: ~22,900–29,000 US optometry practices (IBISWorld), the large majority with an optical dispensary. Most are single-location, 8 employees, ~$973K revenue. Even 5% penetration is ~1,150 practices; 12% is ~2,800.
• Why now for them: Remake economics got worse (lens + shipping inflation), labor is tight, and vision-AI finally makes “read the doctor’s Rx and check my work” something software can actually do.

5. Product sketch (MVP)

• Snap-and-check: Upload or photograph the doctor’s original Rx (handwritten, faxed PDF, or printed) — AI extracts sphere, cyl, axis, add, prism, PD, base curve.
• Order cross-check: Paste, type, or import the lab order you’re about to send; RxCatch diffs it against the extracted Rx field-by-field.
• Red-flag rules engine: flags transposed cyl/sphere (“90-off”), out-of-ANSI-tolerance axis, sphere/cyl sign mismatches, impossible or implausible values, and missing PD / seg-height / add for progressives.
• Material/spec sanity checks: warns on known traps (e.g. CR-39 for drill-mount frames, missing tint/coating callouts) — the “MISSING INFORMATION” class labs complain about most.
• One-line verdict before transmit: green “looks clean” or a ranked list of what to fix, with the source field highlighted on the scanned Rx.
• Remake-cause log: every catch is logged; monthly report shows estimated remakes (and dollars) avoided and your trending error mix.
• Advisory only: RxCatch never alters or transmits the order — it flags; the human decides. (Keeps it out of medical-device territory.)

6. AI angle — what’s load-bearing

Two AI jobs, both load-bearing. First, vision extraction: reading a messy, handwritten, or faxed prescription into structured fields is the thing that was impossible cheaply until 2025-26 — remove it and you’re back to manual re-keying, which is the whole problem. Second, contextual validation: not every “mismatch” is an error (doctor changed the Rx, optician intentionally compensated axis) — the model has to reason about which discrepancies are real red-flags worth interrupting an optician for, or it becomes alert-fatigue noise and gets switched off. Remove the AI and RxCatch is just another form to fill in. The deterministic ANSI-tolerance rules sit on top of the AI reads, not instead of them.

7. Localization angle (if any)

N/A — this is a US-first play. The wedge is US remake economics, US optical-lab portals, and US practice density. The same shape ports to UK/EU/Australia later (same Rx format, same ANSI/ISO tolerances), but there’s no payment-rail or language wedge that makes a localized version win first. Start where the wallets and the trade forums (OptiBoard, AOA) are densest.

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

• Pricing: $149/month per dispensing location (flat, unlimited orders). A single avoided remake/month pays for it twice over.
• ACV: ~$1,800/location/year; multi-location groups bundle at ~$120/location.
• Rough math to $1M ARR: 560 locations × $149/mo × 12 ≈ $1.0M. That’s ~2.4% of US optometry practices.
• Rough math to $5M ARR: ~2,800 locations (~12% penetration) at blended $149, OR 1,400 locations plus a paid lab-side tier (wholesale labs validating inbound orders from hundreds of accounts at $400–800/mo) plus a per-seat upsell. Lab-side is the natural expansion: same engine, pointed at inbound chaos.
• Expansion path: start single-location → multi-location rollups → lab-side inbound validation → remake-analytics/benchmarking module practices pay extra for to negotiate insurance/lab terms.

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

• OptiBoard + trade forums: the optical trade lives on OptiBoard, where remake-rate and transposition threads run for pages. Show up with a free “what’s my remake leak costing me?” calculator, then a 30-day trial. This is where the exact buyer already complains.
• AOA / Vision Source / IDOC networks: the AOA Center for Independent Practice and buying groups like Vision Source and IDOC aggregate thousands of independents and actively push cost-saving tools. Pitch a group discount; one network endorsement = hundreds of warm intros.
• Cold-walk the lab relationship: wholesale labs hate bad inbound orders (“MISSING INFORMATION… minor problems that turn into a major problem”). Partner with 3–5 regional independent labs to recommend RxCatch to their accounts — the lab wins fewer remakes, you get a pre-qualified channel into every practice that ships to that lab.
• Trade shows: Vision Expo East/West — booth a live demo: photograph a deliberately-transposed Rx, watch RxCatch catch the 90-off in two seconds. That demo sells itself to anyone who’s eaten a remake.
• Targeted outreach: scrape state optometry-board licensee lists + practice directories, send a 60-second Loom showing a real catch, expect single-digit reply rate on a $32K-pain message.

10. Build complexity — justification

Medium. Vision extraction of a bounded Rx is off-the-shelf vision-LLM work; the ANSI-tolerance and transposition rules are deterministic and well-documented. The custom work is (a) tuning extraction across the zoo of Rx formats and handwriting, and (b) designing validation that interrupts only on real red-flags so opticians don’t mute it. A technical founder with an optical-domain advisor ships a credible v1 in ~8–12 weeks; the hard yards are the false-positive tuning, not the plumbing.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Advisory decision-support; does not alter/transmit orders or diagnose — stays clear of medical-device regulation.
Ethical — no harm / dark patterns	✅	Reduces patient-harming errors; human stays in the loop.
Market exists (evidence above)	✅	15% remake rate, $32K/yr documented loss, active trade complaints.
1–5 person team can build this	✅	Vision-LLM + rules + thin UI; 8–12 wks.
Launchable with <$50K / ₹40L	✅	Inference + standard web stack; founder-led sales.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	15/20	Real, recurring, costs real money — but absorbed as cost-of-business with manual workarounds, not hair-on-fire daily.
Demand evidence	15	12/15	Multiple independent signals: benchmarked remake rate, hard $ figures, funded adjacent vendors, live forum complaints.
Build feasibility	15	12/15	Off-the-shelf vision-LLM + deterministic rules; main risk is false-positive tuning, not infra.
Distribution clarity	15	10/15	Named channels (OptiBoard, AOA, Vision Source, labs) but conversion uncertain and no viral loop.
Revenue mechanics	15	11/15	Clean SaaS benchmarked against $32K loss; one open question is whether they’ll pay for a “checker” vs absorb the loss.
Time to first revenue	10	8/10	Self-serve trial → paid in 4–8 weeks; one avoided remake justifies the spend.
Defensibility	10	5/10	Thin. Workflow lock-in + accumulating remake-cause data, but Optogrid/SpecCheck could bolt this on. Execution + focus moat.
Total	100	73/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required — needs solid vision-AI + a real optician/lab advisor to nail tolerances and avoid alert fatigue.

Key assumptions to validate (3–5)

1. Assumption: Independents will pay $149/mo for a pre-transmit validation gate rather than keep eating remakes. How to test: 30 cold demos to single-location practices; measure how many start a paid trial after seeing a live catch.
2. Assumption: Vision extraction is accurate enough across handwriting/fax that it catches real errors without drowning opticians in false flags. How to test: run 500 real anonymized Rx + order pairs, measure catch rate and false-positive rate against a known remake-cause log.
3. Assumption: Labs will recommend it to their accounts. How to test: pitch 5 regional independent labs; count how many agree to co-market.
4. Assumption: Transcription/transposition errors are a large enough share of remakes that catching them moves the dollar number. How to test: survey 20 practices on their remake-cause breakdown.

Risk flags

1. Incumbent bolt-on: Optogrid or SpecCheck could add Rx-validation as a feature. Mitigate with speed, a lab-side channel, and remake-analytics depth they won’t prioritize.
2. Alert fatigue: if the gate cries wolf, opticians mute it and churn. The product lives or dies on false-positive discipline.
3. Behavior change: asking a busy optician to add a step (scan the Rx) before transmit is friction; the catch has to feel worth it on day one.
4. Regulatory creep: if positioned as “correcting prescriptions” it edges toward medical-device scrutiny — must stay strictly advisory, human-in-the-loop.

14. Structured verdict

Score:                  73/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder (vision-AI) + optician/lab domain advisor
Time to revenue:        6–10 weeks
Capital to launch:      ₹4–6 lakh ($5–7K)
Top 3 assumptions to validate first:
  1. Will independents pay $149/mo to prevent remakes? — 30 cold demos → paid-trial conversion
  2. Catch rate high, false-positive rate low? — 500 real Rx+order pairs benchmark
  3. Will labs co-market? — pitch 5 regional independent labs
Kill criteria:
  - Abandon if <10% of 30 cold-demo practices start a paid trial after a live catch
  - Abandon if false-positive rate forces opticians to mute the gate in pilot (>1 false flag per 5 orders)

15. Next step — 1-week validation sprint

• Day 1–2: Build a throwaway demo — vision extraction on 50 real/scrambled Rx images + a handful of transposition/axis/PD rules. No UI polish; just prove it catches a 90-off and a missing PD live.
• Day 3–4: Book 10–15 calls via OptiBoard DMs and cold outreach to single-location practices. Run the live demo: photograph a deliberately-bad Rx, watch it catch. Ask the close question: “$149/mo to never ship that error again — yes or no?”
• Day 5: Decide go / no-go on a falsifiable bar: ≥4 of 15 practices verbally commit to a paid trial AND the demo’s catch rate on their own sample Rx exceeds 80% with under 1 false flag per 10 orders. Below that, the gate isn’t trusted enough or the pain isn’t paid-for — revisit.