ShopSense

1. One-liner

Learns how your veteran estimator prices work, so the shop can still quote after he retires.

2. Trend signal — why now?

Three things moved at once, and they point at the same 20 square feet of a machine shop: the desk where one guy prices every job.

The estimator is the bottleneck, and it is measurable. Industry reporting on quote turnaround puts hard numbers on something shop owners have only felt. A shop owner quoting one job a day reported a 50% strike rate; at eight jobs a day that fell to 25%. Same shop, same skill — the win rate halves when the estimator is overloaded. Worse is the loss nobody counts: RFQs that never get quoted at all because the backlog outran the deadline. That number is typically larger than the lost-quote count and, in the trade press’s words, “stays invisible until specifically tracked.” Buyers now blast RFQs to many suppliers at once and often buy from whoever answered first, so a slow quote is read as no capacity.

The person holding that knowledge is leaving. BLS reported 462,000 manufacturing job openings in March 2026 and projects ~34,200 openings a year for machinists and tool-and-die makers through 2034. Deloitte and The Manufacturing Institute put the need at up to 3.8 million new manufacturing employees between 2024 and 2033, with as many as 1.9 million skilled positions at risk of going unfilled. Machine-shop estimator job postings routinely demand 10+ years of hands-on experience, and shops train replacements through inconsistent on-the-job mentoring — senior estimators coaching while still carrying their own quota. The constraint in 2026 is not headcount in the abstract; it is the concentration of pricing judgement in one or two senior heads that walk out with “judgement, setup knowledge, customer history and troubleshooting ability.”

Capturing that judgement got cheap. Knowledge-capture practice has converged on a method that now works: record the veteran talking through his reasoning and run the transcript through an LLM, because experts volunteer diagnostic nuance in conversation that they never write down in forms. The standard guidance is to start capture 6–12 months before retirement. Meanwhile document-understanding models read dimensioned 2D PDF drawings well enough to pull material, tolerance, quantity and finish without a CAD license.

The estimator is the ideal capture target because — unlike a maintenance tech — his job already generated a labeled dataset. Every quote the shop ever sent is a training example: drawing in, price out, won or lost. Nobody is mining it.

Provenance:
  - Signal 1 (Demand): Estimator overload measurably halves win rate (50%→25% from 1 to 8 quotes/day); un-quoted RFQs exceed lost quotes and stay invisible until tracked — https://markovate.com/quote-turnaround-time/ and https://www.mmsonline.com/articles/winning-more-work-as-the-shop-next-door — observed 2026-08-16
  - Signal 2 (Feasibility): Knowledge capture via recorded expert conversation + LLM transcript processing is established practice, recommended 6–12 months pre-retirement — https://learn.uvm.edu/news/how-to-capture-knowledge-retiring-employees-ai/ and https://www.automate.org/robotics/industry-insights/the-great-knowledge-transfer-safeguarding-manufacturing-expertise-as-veteran-workers-retire — observed 2026-08-16
  - Signal 3 (Economic): BLS 462,000 manufacturing openings (March 2026), ~34,200 machinist/tool-and-die openings/year; Deloitte + Manufacturing Institute project 3.8M needed by 2033 with 1.9M at risk unfilled — https://www.cloudnc.com/blog/cnc-machinist-hiring-crisis-2026 and https://themanufacturinginstitute.org/manufacturers-need-as-many-as-3-8-million-new-employees-by-2033/ — observed 2026-08-16
  Category: Underserved niche (with a tech-unlock trigger)

3. The opportunity

The quoting software market solved the wrong half of the problem. Paperless Parts and Machine Research both do geometry-driven pricing: feed a CAD or STEP file, get cycle time and cost from models trained across shops. That is genuinely useful and genuinely well-funded. It answers “what should this part cost in general.”

It does not answer the question that actually determines whether a shop wins money: “what would Dave have quoted this at, and why.”

Dave prices at 1.4× on that aerospace customer because they always change the print twice mid-job. He no-quotes anything from a particular buyer because they shop his number to a competitor. He knows the Haas is a dog on 17-4 stainless and pads accordingly. He remembers that this exact bracket ran in 2019 and the setup took three hours longer than planned. None of that is in the geometry. All of it is in his head, and about a third of it is in the shop’s own quote history — sitting in a folder of PDFs and an ERP nobody queries.

The gap is that incumbents sell a pricing engine trained on the industry. What a 20-person shop with a retiring estimator needs is a memory trained on itself. Those are different products with different price points. A shop that will not buy a $12K/year platform to replace its estimator will absolutely pay to not lose him.

This is also a wedge incumbents are structurally slow to take: their value proposition is cross-shop model quality, and “we learn only your shop’s quirks” cuts against it.

4. Target market

• Primary customer: Owner or general manager of a US contract machine shop, 8–60 employees, $2M–$15M revenue, doing prototype-to-mid-volume CNC milling and turning work. The buyer is the owner; the user is the estimator and whoever is being groomed to replace him. Highest-intent segment: shops where the estimator is 58+ and has no trained backup.
• Why they buy: In their words, from the trade forums — the worry is not software, it is “what happens when your experienced guy leaves.” Practical Machinist threads describe owners handing quoting to an employee where “it took a couple years for the guy to figure it out,” and ask directly whether the knowledge transfers “or do you just eat the pain for a few months while everyone figures it out again.” The buying trigger is a retirement date on the calendar.
• Rough TAM reasoning: Census/industry data shows 12,981 US establishments in NAICS 332710 (Machine Shops) employing ~226,270 people — roughly 17 employees per shop, dead center of the target. Add adjacent sheet metal and fabrication and the reachable pool is larger. If ~30% have a near-retirement estimator and no successor, that is ~4,000 shops in acute pain. At $349/mo I need ~240 of them for $1M ARR — about 6% of the acute segment.
• Why now for them: The retirement is dated, the RFQ volume is up, and the replacement hire wants $62K–$80K and still needs years of seasoning. The alternative to buying is watching win rate fall by half.

5. Product sketch (MVP)

• Backfill from history. Point it at the folder of past quotes and the ERP export. It reads the PDFs and reconstructs a searchable record: this part, this customer, this quantity, quoted at this price, won or lost, and what it actually cost to run.
• The interview. A structured recorded session — 45 minutes a week for eight weeks — where the veteran talks through live quotes and past jobs. It asks the follow-up questions a new hire wouldn’t know to ask: why the premium on this customer, why you no-quoted that one.
• “What would we have quoted this at?” Drop in a new RFQ PDF. It surfaces the three closest jobs the shop has actually run, what they were priced at, what they actually cost, and the rule Dave stated that applies here.
• Customer quirk cards. A per-customer page: historical margin, print-change frequency, payment behaviour, win rate, and the estimator’s stated handling rule.
• Estimate drafts with the reasoning shown. A starting number with every input traceable to a past job or a captured rule — never a black-box price.
• No-quote flags. Warns when an RFQ matches the pattern the veteran said to walk away from.
• Successor mode. The trainee quotes first, the tool shows what the history implies, and the delta gets reviewed — so the handover is measurable rather than hopeful.

6. AI angle — what’s load-bearing

Remove the AI and this is a filing cabinet nobody opens. Three jobs are genuinely model work:

1. Reading the drawings. Shops quote from dimensioned 2D PDFs, often scanned, frequently without a 3D model. Extracting material, tolerances, finish, quantity and revision from that is vision-model work, and it is the reason this was not buildable in 2019.
2. Turning talk into rules. The veteran does not write policy; he tells stories. Converting eight hours of rambling shop-floor narration into discrete, attributable pricing rules is exactly what LLMs became good at, and it is the established method in knowledge-capture practice.
3. Similarity matching. “Which of our 4,000 past jobs is this new print most like” is an embedding problem over mixed text-and-geometry features, not a keyword search.

The deliberate constraint: it retrieves and explains, it does not invent prices. A wrong confident number destroys trust with this customer permanently. The product’s job is to put the shop’s own history in front of whoever is quoting.

7. Localization angle

N/A — this is a US-first play. The wedge is the US machinist demographic cliff, US-style RFQ-blast buying behaviour, and imperial-unit prints. The same shape exists in Germany, Japan and Italy (all with older manufacturing workforces and dense SME machining sectors) and is a credible year-two expansion, but each needs different print conventions and language. Nothing about the wedge requires local payment rails or regulation.

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

• Pricing: $349/mo per shop (up to 5 seats), annual billing preferred. Plus a one-time $2,500 “capture sprint” — the eight-week structured interview program with the retiring estimator, delivered high-touch at first.
• Why that number: Shops already carry ERP line items in this range — ProShop ~$500–$715/mo, JobBOSS² from ~$3,000/yr, Global Shop ~$1,500/mo. Paperless Parts starts around $300/mo and scopes up custom. $349 fits an existing budget category without a board conversation, and the capture sprint is priced against what a shop loses on one mispriced job.
• ACV: ~$4,200 recurring, ~$6,700 in year one including the sprint.
• Math to $1M ARR: 240 shops × $349/mo × 12 = $1.005M. Plus sprint revenue on new logos.
• Math to $5M ARR: ~1,000 shops on a blended $420/mo (seat expansion and a higher tier for multi-site groups). That is ~8% of the 12,981 US machine shops — aggressive but not fantasy, and it likely requires expanding into sheet metal/fab shops with the same estimator problem.
• Expansion path: Seats as the shop trains more than one successor; a second sprint when the next veteran retires; multi-site rollup pricing (small acquisitive shop groups are consolidating and inherit this problem repeatedly).

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

• Sell against a retirement date. Machine-shop estimator job postings on ZipRecruiter, Indeed and Glassdoor are a live signal of exactly the pain — a shop advertising for an estimator with “10+ years” is either replacing a leaver or drowning. Scrape those listings weekly (there are dozens live at any time), find the owner on LinkedIn, and open with “who’s covering quoting until you fill this?” This is the highest-intent list in the business and it refreshes itself.
• Practical Machinist, earned not spammed. The Shop Management and Owner Issues forum runs continuous threads on exactly this (“What happens when your experienced guy leaves?”, “Should a machinist do quoting for the owner?”). Show up as a builder, publish the un-quoted-RFQ math with real numbers, and let the owners come. This crowd sniffs out marketing instantly, so it has to be genuine participation.
• Free “quote autopsy” as the door-opener. Offer to ingest a shop’s last 200 quotes for free and hand back a one-page report: win rate by customer, margin by customer, how many RFQs never got quoted, and which customers are quietly unprofitable. Most owners have never seen this. It requires no commitment, demonstrates the core engine, and the report itself is the sales pitch. Target 5–10% conversion to paid.
• Regional trade nights and MFG Day. State manufacturing associations and community-college machining programs run owner-level events. These are 40-person rooms of exactly the buyer, and one talk on the retirement cliff earns a dozen conversations.
• Referral through the ERP consultants. The independent consultants who implement JobBOSS/E2/ProShop for small shops sit with the owner for weeks and hear the estimator complaint firsthand. Revenue-share them.

10. Build complexity — justification

Medium. The PDF-drawing extraction, the interview-to-rules pipeline, and the similarity search are all off-the-shelf model capabilities wired together — no custom model training required for v1. The real work is unglamorous integration: ERP exports come in a dozen shapes (JobBOSS, E2, ProShop, Excel, sometimes paper), and quote history is messy. Budget 12–16 weeks for a pair to reach a shop that can run a real quote through it, with the first few capture sprints delivered manually by the founder before any of it is productized.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Shop’s own data, used for the shop. Employment-adjacent but the veteran is a participant, not a subject.
Ethical — no harm / dark patterns	✅	One flag worth naming: this must be sold as succession, not replacement. See risks.
Market exists (evidence above)	✅	12,981 US shops; funded incumbents in adjacent quoting; documented owner complaints.
1–5 person team can build this	✅	Pair, 12–16 weeks to first real quote.
Launchable with <$50K / ₹40L	✅	Inference, hosting, and a lot of founder time on the phone with shop owners.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	17/20	Dated, unavoidable, and expensive — win rate halves under estimator overload and the knowledge leaves permanently. Not quite 19: shops have muddled through retirements for a century, and “eat the pain for a few months” is a real, if bad, alternative.
Demand evidence	15	11/15	Strong indirect evidence: hard BLS/Deloitte labour data, quantified win-rate decay, live forum complaints, funded incumbents in the adjacent space. Docked because nobody is currently paying for this specific product — the spend today goes to geometry-based quoting.
Build feasibility	15	11/15	All capabilities off-the-shelf; messy ERP/PDF integration is the tax. 12–16 weeks, not 6.
Distribution clarity	15	11/15	The estimator-job-posting list is genuinely high-intent and self-refreshing, and the free quote autopsy is a strong opener. Docked because this buyer does not buy software online — expect phone calls and shop visits, which caps velocity.
Revenue mechanics	15	12/15	Price sits inside an existing budget line with public comparables; 240 shops for $1M is credible. Sprint revenue helps year-one cash.
Time to first revenue	10	7/10	The capture sprint can be sold and delivered manually before the software is finished, which pulls revenue forward — but the sales cycle is weeks of trust-building, not a credit card.
Defensibility	10	5/10	Per-shop accumulated history and captured rules create real switching cost after a year. But an incumbent with distribution could bolt on “your shop’s history” — the moat is focus and a head start, not technology.
Total	100	74/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

This needs someone who can sit in a machine shop and not be dismissed. Without a co-founder or first hire who has actually quoted parts, the interview design will be wrong and the shops will know within five minutes.

Key assumptions to validate

1. Assumption: Quote history in a typical 20-person shop is complete enough to reconstruct — i.e. the PDFs and ERP records actually connect price to outcome to actual cost. How to test: Get 5 shops to hand over their last 200 quotes. Measure what fraction can be linked drawing→price→won/lost→actual cost. If under half, the “memory” is too thin and the product leans entirely on the interview.
2. Assumption: The retiring estimator will cooperate rather than stonewall. His leverage is the knowledge. How to test: Run three paid capture sprints. Measure whether he volunteers reasoning or gives one-word answers. Structure the pitch as legacy and post-retirement consulting, not extraction.
3. Assumption: Owners will pay $349/mo + $2,500 for succession insurance rather than doing nothing. How to test: Sell 5 capture sprints before writing the software. Cash collected, not LOIs.
4. Assumption: The trainee’s quotes measurably converge toward the veteran’s over a sprint. How to test: Track delta between successor quote and veteran quote across 8 weeks in the first cohort. Flat delta means the product is not transferring anything.

Risk flags

1. Positioning risk — this is the big one. If the estimator believes the tool is there to replace him, he will poison it, and he is the only source of the data. It must be sold as “your judgement outlives your retirement” with the veteran as the credited author. Get this framing wrong and the product is dead on arrival regardless of quality.
2. Incumbent encroachment. Paperless Parts and Machine Research have funding, distribution and the same customers. “Learn from your own quote history” is an obvious feature to add. The bet is that it cuts against their cross-shop model story and that they are chasing larger shops — but a fast follower here is plausible within 12 months.
3. Data quality. Small shops keep quotes in ways that will horrify you: paper, email attachments, an Access database from 2004. The backfill may be far more expensive per customer than modelled, which turns a SaaS into a services business.
4. Slow, high-touch sales. This buyer answers the phone but does not sign up online. CAC in founder-hours is high and does not obviously compress until there is a referral base.

14. Structured verdict

Score:                  74/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical founder paired with someone who has quoted machine work
Time to revenue:        6–10 weeks (sell capture sprints before software ships)
Capital to launch:      $15–25K
Top 3 assumptions to validate first:
  1. Quote history is reconstructable — link drawing→price→outcome→actual cost on 5 real shops' last 200 quotes
  2. Owners pay for succession insurance — collect cash for 5 capture sprints pre-product
  3. The veteran cooperates — run 3 sprints, measure whether reasoning is volunteered or withheld
Kill criteria:
  - Abandon if fewer than 3 of 20 qualified shop owners will pay for a capture sprint after seeing their free quote autopsy
  - Abandon if under 50% of quote history can be linked to outcomes across the first 5 shops
  - Abandon if Paperless Parts or Machine Research ships shop-specific history learning at comparable price before v1

15. Next step — 1-week validation sprint

• Day 1–2: Scrape every live US “machine shop estimator” job posting. Build a list of 40 shops with a named owner. This list doubles as the first sales pipeline.
• Day 3–4: Call them. Not a demo — one question: “If your estimator retired in six months, what happens to quoting?” Offer the free quote autopsy on their last 200 quotes to the ones who flinch. Target 20 conversations.
• Day 5: Take 2–3 shops up on the autopsy. Manually reconstruct their history — no software, just the founder and a spreadsheet — and check whether drawing→price→outcome can actually be linked. Then ask for $2,500 for the capture sprint.

Falsifiable outcome: ≥3 of 20 owners commit cash for a capture sprint, AND ≥50% of quote history proves linkable on the manual autopsies. Miss either and the idea is not ready — the first says nobody buys succession insurance, the second says there is no memory to sell.