Send 10 completed op sheets and a blank template. In two weeks, we hand back a per-dimension accuracy report on your actual handwriting — at no cost, no commitment. If the numbers earn it, we scope a paid pilot together. If they don't, you walk with the data and we go our separate ways. Same workflow at the bench either way — your inspectors keep writing on paper, the way they work fastest.
| Char. | Dimension | Nominal | Tol | Reading |
|---|---|---|---|---|
| D1 | Bore Ø | 0.7500 | ±.0002 | .7501 |
| D2 | Pilot Ø | 0.3750 | ±.0003 | .3752 |
| L1 | Length overall | 1.250 | ±.005 | 1.252 |
| L2 | C'bore depth | 0.250 | ±.002 | .253 |
| R1 | Surface finish | 16 Ra | max | 14 |
| T1 | True position | Ø.0010 | MMC | .0007 |
| Char. | Dimension | Nominal | Reading | Status |
|---|
98.4%
Now the math behind what you just watched. Every inspection report carries ~23 minutes of work that happens after the part is measured — re-keying, verifying, cross-checking, formatting. Apprend collapses that tail. The numbers below are stated per-inspector so they scale to ATM's actual headcount, whatever it turns out to be.
Headcount, reports/day, and loaded rate are placeholders — these get replaced with measured values from ATM's actual job mix during Phase 2. Phase 1 establishes whether the accuracy is there to make those measurements worth taking.
Four stages. Same paper trail you already keep — a digital copy lands on top.
From the shop walkthrough: a 30-minute inspection is the floor, but the report doesn't ship until the cross-check, re-key, verify, and format steps are done. That tail is what disappears.
~23 minutes saved per report · ~51% reduction. Not because measurement gets faster — it doesn't — but because the four steps after measurement collapse into a photo and a glance.
Three phases. Each one earns the next. ATM can walk away at any point with everything we've built up to that moment — accuracy data, extracted reports, working capture link. Nothing is locked behind a future commitment.
Phase 1 is bounded by design — five steps over two weeks, ~1 hour of ATM's time total, ending with a real accuracy report on real ATM sheets. No agreement to sign, no kickoff onsite, no inspector training. Just proof.
One email from Bao with: the blank op-sheet template, the Excel inspection report you send the customer, and 10 completed sheets from one recent part family. We confirm receipt the same day.
We map every field — fixed text, machine-printed nominals, handwritten readings, signatures — and wire each row to the right cell in your Excel template. Zero ATM time required.
All 10 sheets go through the pipeline. Every reading is compared cell-by-cell against the original handwritten value. We share an early-accuracy snapshot mid-week so Bao sees progress before Day 14.
Anything below the accuracy bar gets a tuning pass. Edge cases (smudged readings, ambiguous decimals, surface-finish-max vs. ±tol) are documented so Bao sees exactly where the model is strong and where it isn't.
One 30-minute call. We walk through per-dimension accuracy, the filled Excel reports, projected time savings against ATM's job mix, and an honest take on whether Phase 2 is worth scoping. ATM keeps everything either way.
Three numbers come out of Phase 1. They're how Bao decides whether Phase 2 is worth scoping — and, in Phase 2, they become the binding success bar that triggers the guarantee.
You pay for what you've already seen work. Nothing earlier. Phase 1 is on Apprend — we cover the engineering time because if we can't read ATM's handwriting at accuracy, there's no point continuing on either side.
ATM runs aerospace, medical, and customer-confidential parts. We treat the data like it matters.
Apprend never touches your ERP, drawings, or customer files. Only the captured op-sheet image and the resulting extraction leave your floor.
TLS in flight, AES-256 at rest. Captured images are deleted on a configurable retention window — default 30 days, can be 0 days on request.
ATM job data is not used to train shared models. Tuning happens against a separate pilot dataset, in a tenant scoped to ATM.
Every extracted row is linked back to the original sheet image and inspector initials — the audit trail your ISO 9001 surveillance reviews expect.
All capture and extraction runs on US-region infrastructure (AWS us-east / us-west). ITAR / EAR-controlled parts can be excluded from the pilot floor trial; SOC 2 Type II audit and full ITAR compliance are on the post-pilot roadmap and can be scoped if either is a gating requirement.
ATM owns the captured images, the extracted readings, and the resulting reports. On termination, Apprend exports everything and deletes our copies within 30 days.
Lifted from conversations with shop owners weighing similar engagements. If yours isn't here, ask Vidhi.
Because the bar is honest: if Apprend can't read ATM's handwriting at the accuracy your inspectors would need to trust the output, neither side should waste any more time. The free phase is how we both find out — quickly, without putting ATM on the hook financially before there's evidence.
One email with three attachments: the blank op-sheet template, the customer Excel report, and 10 completed sheets from a recent part family. Then a 30-minute readout call at the end of week 2. That's it — ~1 hour of Bao's time total.
Apprend says so honestly at the readout and there's no Phase 2. ATM keeps the accuracy report, the edge-case catalogue, and the 10 filled customer reports. Apprend deletes ATM's source data within 30 days. No invoice, no awkward follow-up, no "let's just push through anyway."
4–6 weeks, two part families, a live floor trial with one inspector, measured time savings against ATM's manual baseline. Fixed-fee, priced together with Bao at the readout once we both know what the work actually requires. Success-guaranteed: invoiced 50/50, second half waived if we miss the bar.
Phase 1 doesn't touch the floor — it runs on completed sheets from your existing workflow, so adoption isn't a question yet. Phase 2 is where the live floor trial happens, and it's designed around the inspector's existing flow: keep writing on paper, snap one photo at the end. The trial is also when we'd catch resistance early, while ATM still has the off-ramp.
Phase 1 has no recurring commitment — there's nothing to cancel. ATM keeps the filled reports and the accuracy data regardless. For Phase 2 and beyond, on termination Apprend exports everything we have for ATM in the original Excel format plus the captured images, and any production scripts we built. Your customer reports are not dependent on us being online to produce them.
Yes for non-export-controlled work. For ITAR / EAR-controlled programs, exclude those parts from the Phase 1 sample set — we'd scope a compliant path before Phase 2 if you want to extend coverage. We can sign an NDA at Phase 1 if ATM prefers; not required.
Vidhi Kothari, Apprend's CEO, runs the engagement personally — she is your point of contact from sample handoff through the Phase 1 readout and into Phase 2 if it goes there. One ML engineer pairs with her on the extraction pipeline. No handoff to an account team, ever.
Apprend already runs a production handwriting-extraction pipeline at a Pennsylvania injection-molding customer, reading logbook data straight off the press floor into their review queue. The same pipeline is what you'd run for ATM's op sheets. Manufacturing-specific OCR — tuned for the messy edges of bench writing, tolerance shorthand, and the smudges that come with the job.
That's the entire setup for Phase 1. No agreement to sign, no kickoff onsite, no commitment until Bao has seen real numbers from his shop's sheets. Reply to Vidhi when you have the four items below ready — we run the clock from sample receipt.