PrintQuote — RFQ estimator for small machine shops

1. One-liner

Reads the 2D PDF print and napkin RFQ a small shop can’t get to, and turns it into a costed quote in minutes.

2. Trend signal — why now?

Three things landed at once.

The pain is measured, not anecdotal. CNCCookbook’s quoting survey found spreadsheets are “the number one player by a wide margin,” followed by literal “eyeball guestimate,” and that at least 40% of respondents quote with no CAD model at all — working from “back of napkin drawings.” Quote-to-book ratio is 51% for average shops vs 70% for the top. The survey’s own conclusion: “nobody is happy with any of the available solutions.” Separately, industry coverage states ~85% of job shops admit they skip RFQs because the quoting process is too tedious, and ~75% don’t know whether the work they win is actually profitable. That is revenue walking out the door twice — once on the RFQs never answered, once on the won jobs that bleed margin.

The tech just became load-bearing. As of 2026, multimodal LLMs read 2D engineering prints — title blocks, material callouts, standard dimensions, surface-finish symbols — at 85–95% accuracy, dropping to 70–80% on dense GD&T frames with multiple datums (per RFQ-parsing vendors and tolerance-extraction writeups). Before 2025 this was a hand job: ~75% of organizations still extract GD&T by pencil-and-spreadsheet, and complex parts “can take multiple days.” The print is the legally binding contract and carries the specs a STEP file does not — so reading the print, not the CAD geometry, is the real bottleneck, and that is exactly what’s newly automatable.

Money is moving here, but aimed up-market. Mavlon sells AI RFQ parsing at $15K–$50K/year, explicitly targeting shops doing “20+ RFQs per week” in aerospace/automotive/industrial. Paperless Parts runs custom pricing starting ~$1,000/month. DigiFabster runs $2K–$50K/year. CADDi Drawer is enterprise drawing-search. Every funded player is built for the 250+/enterprise contract manufacturer with clean CAD — none of them are priced or shaped for the shop with 6 machinists and a fax-quality PDF.

Provenance:

• Signal 1 (demand): CNCCookbook quoting survey — 40%+ quote with no CAD, 51% avg quote-to-book, “nobody is happy with available solutions”; ~85% of shops skip RFQs as too tedious — https://www.cnccookbook.com/job-quote-cost-estimation-survey-results/ + https://www.thefabricator.com/thefabricator/article/shopmanagement/5-steps-to-faster-quoting-in-the-job-shop — 2026-05-31
• Signal 2 (feasibility): 2026 multimodal print reading at 85–95% on standard fields / 70–80% on complex GD&T; ~75% still extract GD&T by hand, complex parts take “multiple days” — https://mavlon.co/post/ai-rfq-parsing-for-manufacturing + https://www.infrrd.ai/blog/tolerance-data-extraction — 2026-05-31
• Signal 3 (economic): Funded RFQ-quoting AI priced for enterprise — Mavlon $15–50K/yr (20+ RFQ/wk), Paperless Parts ~$1K+/mo, DigiFabster $2–50K/yr; 5,400+ verified shops, 83% under 20 employees — https://www.capterra.com/p/179259/paperlessPARTS/ + https://www.doss.com/machine-shop-database — 2026-05-31 Category: Tech-unlock

3. The opportunity

The whole funded category solves the wrong half of the quote for the wrong customer.

Paperless Parts, DigiFabster, aShop, FilletPro all start from a clean STEP/CAD file and instantly price geometry. That’s lovely if your customers send 3D models. But 40%+ of small-shop RFQs arrive as a 2D PDF, a scan, a photo of a print on the shop floor, or three lines in an email body — no model. The instant-CAD tools simply can’t ingest that, so the small shop is back to a human squinting at the print and re-keying dimensions into Excel.

Mavlon does read the messy 2D print — but it’s $15–50K/year and built for shops drowning in 20+ RFQs a week. The 5,400+ US shops where 83% have under 20 employees don’t have that volume or that budget. For them the binary is brutal: answer the RFQ (an hour of the owner’s night, after running parts all day) or skip it (85% do, routinely). Either way they’re losing.

PrintQuote is the SMB-wallet version of the part nobody small can buy: drop in whatever the customer sent — PDF, scan, photo, email — and get back an extracted part spec (material, dims, tolerances, finish, qty, due date) plus a first-pass costed quote built from the shop’s own rates and history. The AI does the reading-and-extracting (the slow, error-prone human step). The costing math stays deterministic and tied to the shop’s numbers — deliberately not a hallucinated price.

4. Target market

• Primary customer: Owner or lead estimator at a US job/machine/fab shop with 2–25 employees (CNC milling/turning, sheet-metal, weld shops). Typically the owner quotes nights/weekends or one overloaded estimator does. $500K–$8M revenue.
• Why they buy: In their words (paraphrased from Practical Machinist + CNCCookbook): quoting is “tedious, time-consuming”; they “gave up trying to price every line item” because the effort-to-win-ratio is upside-down on small jobs; one-man shops “do not expect email responses in less than 24 hours” because the owner is on the machine; 40% are working from “back of napkin drawings.” They want to answer more RFQs without hiring an estimator, and stop bleeding margin on blind quotes.
• Rough TAM reasoning: ~5,400 verified CNC/job shops in one database; the broader US metalworking + fab universe (machine shops, sheet-metal, weld, plastics) is tens of thousands of establishments, the vast majority under 20 employees. Even 3,000 reachable SMB shops at $150/mo is ~$5.4M ARR.
• Why now for them: The bottleneck (reading 2D prints) only became automatable in the last 12 months, and the enterprise tools made the problem visible (RFQ AI is now a category their bigger competitors talk about) without ever becoming affordable to them.

5. Product sketch (MVP)

• Drag-drop or email-forward an RFQ in any form: 2D PDF print, scanned drawing, phone photo, or plain email text.
• AI extracts a structured part spec: material, key dimensions/envelope, tolerances + GD&T it’s confident about, surface finish, quantity, due date, revision.
• Confidence flags: every extracted field shows high/low confidence; low-confidence GD&T is surfaced for a 10-second human confirm, never silently guessed.
• First-pass costed quote from the shop’s own setup time, run rates, material markup, and margin — editable, not a black box.
• Shop “memory”: learns from past quotes (this customer, this part family) to pre-fill rates and flag “you quoted this last year at $X.”
• One-click branded quote PDF + email back to the customer.
• Quote log with win/loss tracking, so the shop finally sees its quote-to-book ratio and which jobs actually booked.

6. AI angle — what’s load-bearing

Remove the AI and the product is just another quoting spreadsheet — so the AI is the whole wedge. The load-bearing job is reading inconsistent 2D engineering evidence — a faxed print, a phone photo at an angle, an email that says “need 50 of these, 6061, anodized” — and turning it into a clean, structured part spec the costing engine can price. That’s the multi-hour, error-prone human step that only became machine-doable in 2026.

What is not the AI, on purpose: the price itself. Costing runs on the shop’s deterministic rates and formulas. A hallucinated quote is worse than no quote — it loses money on the job or the customer. So AI reads; math prices. That split is also the trust story for a skeptical machinist audience.

7. Localization angle (if any)

N/A — this is a US-first play. The wedge is the US small-shop install base, English-language prints, ANSI/ASME GD&T conventions, and US distribution channels (Practical Machinist, regional MEPs). A later ISO/European cut (DIN tolerances, metric-default) is a real expansion, but forcing localization now would dilute the wedge.

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

• Pricing: $99/mo Solo (1 estimator, ~30 RFQs/mo), $199/mo Shop (3 seats, higher volume), $349/mo Pro (unlimited seats + win/loss analytics + customer/part memory). Optional usage overage on heavy RFQ months.
• ACV: ~$2,000–2,400 blended.
• Rough math to $1M ARR: ~450 shops × ~$185/mo × 12 ≈ $1M. Out of 5,400+ verified shops (and a far larger fab universe), 450 is ~8% of one database — very reachable.
• Rough math to $5M ARR: ~2,200 paying shops, plus per-RFQ overage on busy shops, plus an adjacent vertical (sheet-metal/fab uses the same print-reading + costing loop with different rates) and an ERP-integration tier. Achievable without leaving the SMB segment.
• Expansion path: seats → win/loss analytics tier → customer/part-family memory (stickier the longer they use it) → outbound “auto-draft a quote the moment an RFQ hits your inbox” automation.

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

• Practical Machinist + CNCZone forums and FB groups. The exact audience lives here, openly complaining about quoting in threads like “Quoting work — how do you do it?” Post a 90-second screen recording: messy PDF in → costed quote out. Be present in quoting threads, not spammy. These communities have tens of thousands of shop owners.
• The “free RFQ teardown” magnet. Shops forward (or upload) a real RFQ they haven’t gotten to; we return the extracted spec + a draft quote for free. It’s the demo and the lead in one. Converts the “I’d never get to this” pile into a live trial.
• MEP / state manufacturing-extension and regional machinist meetups. Every US state has a Manufacturing Extension Partnership working directly with small shops on efficiency — a natural co-sell/referral channel into exactly the under-20-employee segment.
• Cold outreach off shop directories (DOSS / Thomasnet / Manta): scrape small CNC/fab shops, send a personalized Loom that quotes a sample part from their public capabilities page. 5% reply on a hyper-relevant niche is realistic.
• Win/loss hook for referrals: once a shop sees its real quote-to-book ratio jump, that number is the testimonial that sells the next shop.

10. Build complexity — justification

Medium. The reading layer is off-the-shelf 2026 multimodal models with structured-output prompting; the costing engine is a deterministic rules/rates app (standard web stack); the hard part is the confidence-gating and human-confirm UX that keeps a 70–80%-accurate GD&T read from producing a bad silent quote, plus tuning extraction on real, ugly prints. A small team ships a credible v1 in ~3–4 months; the moat-building (per-shop memory, accuracy on messy inputs) accrues after launch.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	No regulated data; quotes are the shop’s own commercial output.
Ethical — no harm / dark patterns	✅	Confidence flags + human confirm prevent silent bad quotes.
Market exists (evidence above)	✅	Measured: 40% no-CAD RFQs, 85% skip RFQs, funded incumbents up-market.
1–5 person team can build this	✅	Off-the-shelf models + deterministic costing app.
Launchable with <$50K / ₹40L	✅	Inference + standard SaaS infra; sub-$50K to first revenue.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	16/20	Real, recurring, money-on-the-table both directions (skipped RFQs + blind margins). Not literally daily hair-on-fire for every shop, but felt every quote.
Demand evidence	15	12/15	Multiple hard signals: survey stats, funded incumbents, active forum complaints. Held below 13 because direct SMB willingness-to-pay at $99–349 is inferred, not yet proven.
Build feasibility	15	11/15	Doable in 3–4 months; accuracy-on-messy-prints and confidence UX are the real engineering, not trivial.
Distribution clarity	15	12/15	Named communities + directories + MEP channel + a strong free-teardown magnet. Conversion math still estimated.
Revenue mechanics	15	12/15	Pricing benchmarked well below incumbents; 450 shops to $1M is reachable. SMB churn risk on a tool used only when RFQs come in.
Time to first revenue	10	8/10	Free-teardown → paid trial can convert in weeks; not pre-sold.
Defensibility	10	5/10	Execution + accumulating per-shop quote memory. Copyable by a funded incumbent shipping a cheap SMB tier — that’s the main risk.
Total	100	76/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy (extraction accuracy + confidence UX), domain-expertise-required (you must understand how shops actually cost a job — setup vs run time, material markup, GD&T impact — or the quote is garbage). A founder with shop/estimating background plus an AI engineer is the ideal pair.

Key assumptions to validate (3–5)

1. Assumption: Small shops will pay $99–349/mo for a quoting aid (not just admire it). How to test: 30 shop interviews + a pre-sale “$99/mo, first month half off” offer off the free-teardown magnet; need ≥15% teardown-to-paid.
2. Assumption: Extraction is good enough on their real prints (faxes, photos, no-CAD) that the time saved is obvious. How to test: Run 100 real RFQs from 10 shops; measure field-level accuracy and minutes-to-quote vs their manual baseline.
3. Assumption: Shops trust an AI-assisted quote enough to send it to a customer. How to test: Track how often the first-pass quote is sent with minor edits vs torn up — target ≥60% sent with light edits.
4. Assumption: The free-teardown magnet actually pulls the “never got to it” pile into trials. How to test: Measure teardown submissions → trial starts from forum + cold outreach.

Risk flags

1. Competitive / platform risk: A funded incumbent (Mavlon, Paperless Parts, DigiFabster) ships a self-serve sub-$100/mo SMB tier with 2D-print reading before v1. This is the primary kill.
2. Accuracy/trust risk: If GD&T-heavy parts produce wrong specs and a shop quotes badly once, trust is gone in a low-trust, word-of-mouth industry. Confidence-gating is non-negotiable.
3. Usage-cadence risk: Shops only quote when RFQs arrive; light-month churn and “I’ll just do it myself this week” are real. Win/loss analytics and part-memory are the retention hooks.

14. Structured verdict

Score:                  76/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       AI engineer + machine-shop/estimating domain expert (2-person team)
Time to revenue:        6–10 weeks (free-teardown magnet → paid trial)
Capital to launch:      $15–30K (inference + SaaS infra + a few months runway)
Top 3 assumptions to validate first:
  1. SMB willingness-to-pay $99–349/mo — 30 interviews + pre-sale, need >=15% teardown-to-paid
  2. Extraction accuracy on real messy/no-CAD prints — 100 RFQs from 10 shops, field-level accuracy
  3. Quotes trusted enough to send — >=60% of first-pass quotes sent with only light edits
Kill criteria:
  - Abandon if <15% of free-teardown users convert to a paid trial after 60 days
  - Abandon if extraction can't clear ~90% field accuracy on standard fields on real shop inputs
  - Abandon if a funded incumbent ships a self-serve sub-$100/mo 2D-print SMB tier before v1

15. Next step — 1-week validation sprint

• Day 1–2: Pull 50 real RFQs (PDF prints, scans, photos, no-CAD) from 5 friendly shops + forum volunteers. Run them through an off-the-shelf multimodal model with a structured-extraction prompt. Measure field-level accuracy (material, dims, tolerance, finish, qty) against ground truth.
• Day 3–4: Stand up a bare landing page + the “free RFQ teardown” offer. Drop it in two Practical Machinist quoting threads and DM 40 small shops from a directory. Count teardown submissions and “would you pay $149/mo for this” replies.
• Day 5: Decide. Go if (a) extraction clears ~90% on standard fields and visibly beats manual time on real prints, AND (b) ≥10 shops submit a teardown and ≥3 say they’d pay. No-go if accuracy is mushy on real inputs or the teardown magnet pulls nothing.

Falsifiable: either real shop prints extract cleanly and shops raise their hand off the magnet, or they don’t.