Heatmark: Business Plan (v2, revised after red team)

A shop-floor operating system for small structural and miscellaneous steel fabricators. This is the plan for the venture brief titled "FabFloor".

Status: pre-concept. Heatmark has no customers, pilots, LOIs, partners, testimonials or revenue. Every number marked ASSUMPTION is modeled. Every fact marked EVIDENCE has a source URL in the text or in the Sources list (section 18.2).

What changed in v2: three red-team reviews (a bootstrapped vertical-SaaS operator, a long-time small-shop owner, and a CTO) found that the v1 model left out the people needed to run the company and used optimistic churn, pricing and fee assumptions. v2 rebuilds model.json from new-logo funnels, segment pricing and a fully loaded cost base, and the verdict is materially worse. Section 19 lists every critique and the response. The v1 files are kept as plan.v1.md, model.v1.json and model_build.v1.py for comparison.

Research note: the shared web-search budget was exhausted before and during this revision; one search attempt this session was refused. The Census size-band table could not be pulled: the Census API returned a "Missing Key" page when fetched directly this session. Items marked UNVERIFIED should be checked before anyone relies on them: Census size bands for NAICS 332312/332323, small vertical-SaaS sale multiples, SMB logo-churn benchmarks, state SaaS sales-tax rules, AISC 207 traceability clauses, Build America, Buy America (BABA) documentation, trade-association details, SDS2/SigmaNEST ownership, and Trimble's PowerFab API terms.

Brand note: the brief's working name "FabFloor" is generic and hard to defend as a mark. This plan uses Heatmark, which combines the heat number (mill traceability) and the piece mark (shop tracking). Section 18 lists alternatives.

1. Executive summary

The case. Small US structural-steel and misc-metals fabricators, the family shops with roughly 5 to 75 people, mostly run production on spreadsheets, whiteboards and paper travelers. The steel-specific management suites are built and priced for larger shops:

• Tekla PowerFab has a "starting price" of $50,000 for a perpetual license per SoftwareConnect [S1]. Its listed cons are "High price" and "Can have a high learning curve".
• StruMIS is quote-only [S15].
• FabTrol, the third legacy option, is gone. Per a competitor's account, sales and development ended July 22, 2021, and the product was withdrawn September 1, 2022 [S5]. Trimble markets a "transition from FabTrol to Tekla PowerFab" [S6].
• EZIIL, the modern low-cost entrant, lists $140/$210/$338 per month by team size, with BOM, shipments, certificates and inventory as add-ons at undisclosed prices [S4].

Heatmark's wedge is phone/tablet piece tracking at handoffs (cut, to paint/galv, back, on truck) that ends in a load list and bill of lading, fed by whatever the shop actually gets from its detailer: DSTV/NC1, KISS, CSV reports, or a PDF shop-drawing bill of materials. A revision-diff engine keeps the board right when drawings change mid-job. Mill-cert storage and progress-billing backup are a paid "Office" add-on. Pricing is per shop: $400 or $900/mo (a $1,500 tier exists but few are expected), unlimited floor users.

What the Tekla deadline really is (corrected in v2). Trimble's end-of-life page has a notice dated October 1, 2026 with a deadline of December 31, 2027. After it, license activation and re-hosting are not supported for perpetual licenses without a valid maintenance contract, and "a hardware failure will result in the permanent loss of the license" [S8]. The notice names no products and does not say licenses stop working. Tekla EPM maintenance products appear in a separate June 2024 end-of-sale notice; PowerFab appears in a February 2025 subscriptions notice [S8]. v1 overstated this as "every PowerFab/EPM perpetual shop will be re-quoted in 13 months." It is a door-opener for existing PowerFab/EPM users, not a demand driver, and v2 assigns zero base-case customers to it.

Headline numbers, v2 base case (all ASSUMPTIONS from model.json; "fully loaded" means one founder paid as operator from month 6, then a paid GM; "founders unpaid" is the v1 convention):

• Cash: from $150k, the base case stays positive only if the working founder defers about $825k of cumulative pay through Y5, and even then the low point is about $7k in Y3. That is too thin. From $200k with deferral the low point is about $57k. If founder pay is paid in cash, the base case needs roughly $450-650k more capital than the brief allows.
• Market (EVIDENCE plus estimate): NAICS 332312 has 3,434 establishments (Census CBP 2023 via an aggregator) [S11]; NAICS 332323 has about 2,540 to 2,801 [S12][S13]. How many of these are 20-to-75-person shops with real piece-tracking pain is UNVERIFIED; v2 carries two estimates, 1,700 to 2,000 (v1) and 700 to 1,200 (shop-owner red team). The base case acquires about 208 structural shops over 5 years, which is 10 to 12% of the wide estimate or 17 to 30% of the narrow one.

Verdict in plain English

Does it meet the criteria? No, not in the base case.

• $500k to $1M a year of profit on $100-200k: fails in the base case (about $140k fully loaded in Y5; about $360k only if nobody is paid to run it). It is reached only in the bull case: about $620k fully loaded in Y5 on about 344 shops. My subjective estimate, conditional on passing the 90-day tests, is a 5 to 15% chance of $500k+/yr fully loaded profit by Y5. Basis: reasoning only; it requires every bull driver at once (12% churn, about 150 new shops a year by Y5, and Office add-on attach of 30%+). No cited base rate.
• $10M value in 3 to 5 years: fails in the base case (Y5 roughly $3.5 to 6M to a financial buyer on ARR, $7 to 9M only to a strategic buyer). Bull case Y5 is $7 to 12M financial and $13 to 17M strategic. Subjective estimate, conditional on passing the tests: 3 to 10%. Basis: requires the bull case and a strategic buyer, realistically Trimble, which also controls the Tekla ecosystem. The one niche comparable, AVEVA's reported $6.5M purchase of FabTrol in 2015 (per EZIIL, not independently verified) [S5], points to a low ceiling.
• Without a full-time job: fails as written, in every scenario. Through Y3 this is at least one full-time founder and probably 1.5: about 140 to 270 tailored demos a year by Y2-Y3, 20-30 hours of onboarding per shop, 5:30am support on ship mornings, and parser work for every detailer's quirks. The founders' 10-15 hrs/wk combined becomes realistic only from about Y4, and only after a paid GM (the $200-250k/yr line v1 left out). This is fatal to the brief as written, not to the business.
• Capital: $200k (not $150k) is the right commitment, and only with one founder working full-time on deferred pay.

What has to be true for it to work:

1. There are at least about 700 US structural shops in the 20-to-99-employee band, plus a reachable misc-metals segment.
2. Most target shops receive machine-readable piece lists, or Heatmark's PDF-BOM import works well enough that setup stays under about 25 hours.
3. Owners see a clear difference from EZIIL's real (add-on-inclusive) price and pay $400-900/mo plus attach the Office add-on.
4. Floor scanning survives the foreman: annual logo churn of 12% or less, not 18-25%.
5. One founder goes full-time for about 30 months on deferred pay, and a GM is affordable by Y4.

The three cheapest tests, in order:

1. Census size-band check (about $0, one afternoon). Get a free Census API key and pull County Business Patterns establishments by employment size class for NAICS 332312 and 332323. Stop if fewer than about 700 structural establishments have 20 to 99 employees.
2. File-reality and EZIIL-quote test (about 2 weeks, under $1k). Ask 20 shops, found through steel service-center reps and equipment dealers, to send the files from their last 3 jobs, and request 2 to 3 real EZIIL quotes with the BOM, shipments and certificates modules. Redesign or stop if fewer than half get machine-readable piece lists, or if EZIIL's comparable setup is within about $150/mo and owners can't see the difference.
3. Deposit test (days 31 to 60, under $10k including a demo build). After a demo on each shop's own files, at least 5 of 20 owners sign a refundable deposit or an order conditional on a working pilot at $400+/mo; then at least 3 reach 50%+ ship scans by week 3 of a live job.

Bottom line: v1 called this "the cash-flow pick with a modest ceiling." v2 says it is a modest, founder-operated software business that probably does not meet this brief. It is worth the three cheap tests above, which cost almost nothing, but the founding team should not commit the $200k unless they accept that one of them takes a full-time job for about three years.

2. Problem and customer

Who exactly

• Buyer: the owner, president or GM of a family-owned structural or misc-metals fabricator, (v2) most likely with 20 to 75 employees (section 5). Usually second generation. Often personally estimates, or has one estimator and one PM.
• Daily users:
  • the shop superintendent or foreman (status, priorities)
  • fitters, welders, painters and the yard/shipping lead (scanning)
  • the PM/office manager (load lists, BOLs, progress billing)
• Work mix: commercial and light-industrial structural steel (beams, columns, braces, joists bought out), embeds, stairs, rails, canopies and mezzanines. Work is usually subcontracted to a GC or erector. Payment comes as monthly progress billing (AIA G702/G703-style schedules of values) with retainage.
  • Context, not verified this session: G702/G703 are the common progress-billing formats in US commercial construction.

How they buy (ASSUMPTION, based on reasoning about this segment; to be validated in interviews)

• Word of mouth from other fabricators. Regional fabricator and association events. The detailer, nesting or saw vendor they already use. Trade shows.
  • UNVERIFIED this session: NASCC: The Steel Conference (AISC) and FABTECH are the obvious US shows, and NOMMA is the misc/ornamental association.
• The owner signs, usually after seeing the software run on their own job data, but the foreman, the office manager and often a second-generation family member can veto (section 9).
• They are skeptical of subscriptions but already pay monthly for QuickBooks, phones and equipment leases.
• A shop at $400 to $900/mo is within owner-discretionary spend, so there is no procurement committee. The real gate is "will my foreman actually use it?"

Evidence of pain

EVIDENCE: structural displacement.

• FabTrol, historically one of the three main steel-fab management systems, stopped sales and development on July 22, 2021, and was withdrawn September 1, 2022. Source: EZIIL's account [S5], a competitor's blog. I could not find the primary AVEVA notice.
• Trimble's own marketing addresses "With FabTrol retiring..." and offers a migration webinar [S6].
• (Corrected in v2.) Trimble's October 1, 2026 notice ends activation and re-hosting for perpetual licenses without valid maintenance after December 31, 2027; after that a hardware failure means permanent loss of the license [S8]. It names no products and does not say licenses stop working. It is a door-opener with existing PowerFab/EPM users, not evidence of demand among non-users (section 4).

EVIDENCE: price gap.

• PowerFab's "starting price" is $50,000 perpetual (SoftwareConnect) [S1].
• Capterra lists a figure of $8,000 "per user per month". It is implausible, and I treat it as a listing error [S2].
• G2 says final cost must be negotiated with the seller [S3].
• A competitor (EZIIL) estimates PowerFab for smaller shops "possibly in the low five-figures range annually" with "several months" of implementation [S21]. That is a competitor claim, not verified.

EVIDENCE: what users say about incumbents. The volume is very thin, and that is itself a finding.

• SoftwareConnect lists PowerFab cons as "High price" and "Can have a high learning curve." Its one user review (June 2023) says "Nothing worked exactly like what we were hoping for" about labor coding and takeoff import [S1].
• Capterra reviewers of the sister product Tekla Structures complain about support response ("you won't get a response within the same day") and about maintenance costs for "permanent" licenses [S23]. That is evidence about Trimble's support and licensing posture, not PowerFab specifically.
• Vendor content (Msuite, which serves MEP fabrication) describes the "outgrown spreadsheets" pattern: production status and labor data "can no longer be trusted without manual double-checking" [S24]. This is vendor marketing, not user testimony.

What is missing (skeptic's note): I found no organic forum or Reddit complaint volume about lost pieces or ship-status guessing. That pain is plausible to anyone who has walked a fab yard, but it is not evidenced in public sources. The 90-day interviews must produce it, ideally quantified as re-cut pieces per month, hours spent hunting pieces, and back-charges for short-shipped loads.

3. Solution and product

Core loop (pre-customer #1). v2 changes are marked (v2).

1. Import. Build a job piece list (piece mark, assembly, shape/size, grade, length, weight, quantity) from whatever the shop actually has, with confidence highlights for anything inferred:
   • KISS and detailer CSV/XML report exports, which carry assemblies and piece marks.
   • DSTV/NC1 (one file per part). (v2) NC1 has no assembly hierarchy, so it supplements KISS or reports rather than replacing them (CTO red team; reasoning, not verified against the DSTV specification this session).
   • (v2) PDF shop-drawing bill of materials → piece list, with human review. Promoted from "second wave" to core, because many shops without a CNC beam line, and most misc-metals shops, receive PDFs rather than NC files (shop-owner red team; ASSUMPTION to test in interviews).
   • (v2) A free "Heatmark report template" that detailers load into Tekla Structures and SDS2, so export is one click. No Tekla Open API plugin at launch (see 3.1).
2. (v2) Revision diff. When Rev 2 arrives, compare it to Rev 1: added, removed, changed-geometry and changed-quantity pieces. Affected pieces are flagged against their scanned status and a human accepts the merge. Status is stored as an append-only event log keyed to a stable piece identity, not the mark string.
3. Tag. QR tags for assemblies and loose parts, printable on industrial label printers or wire-on poly/aluminum tags.
4. Scan at handoffs only. (v2) Three or four scans per piece: cut list complete, to paint/galv, back from paint/galv, on truck. Scan a tag or type a piece mark when the tag is painted, burned or galvanized over.
   • (v2) Android-first in a native wrapper (for example Capacitor), with a recommended rugged phone of about $150 (ASSUMPTION), plus a shared station tablet in kiosk mode for shops that ban personal phones on the floor. iOS Safari web apps have no Background Sync and can evict stored data (CTO red team; not re-verified this session).
   • (v2) Offline events carry client UUIDs, are idempotent and are ordered on the server; conflicts are shown to the foreman, never silently overwritten. Photos upload lazily.
5. Job board. Per-job percent complete by piece count and by tonnage; holds and missing pieces.
6. Load lists and BOL. Select completed pieces, check weight against truck capacity, sequence by erection sequence, print the BOL and load list, scan pieces onto the truck.
   • (v2) Ship-morning fallback: the device can generate the load list and BOL locally from cached data if the server or connection is down, so a truck never waits on Heatmark.
7. (v2) In-flight job import. Load the detailer report for an open job and bulk-set status by sequence or lot ("all of Seq 1 shipped"). Target about 2 hours per open job, as a standard onboarding step. Without it, a shop with 8 open jobs runs two systems and the shop-wide board is meaningless for months.
8. (v2) Scan-health alerts. A weekly owner email: "job 2417: 41% of pieces have no ship scan." This is the leading indicator of churn.
9. Hours. Crew time against job and station. (v2) Labeled "job costing, not timekeeping"; no payroll export in V1 (see section 16).

Office add-on ($300 to $500/mo; gated on demand):

10. Mill-cert storage V1 (v2, narrowed): store and search mill test reports per job and PO, linked by bundle or receiving. AI extracts heat number, grade, size and values, and every extracted field is confirmed by a human against the source PDF. Heat-at-cut linking comes only in V2, and only via SigmaNEST or saw-line reports that already carry the stock ID, because a saw operator will not type a heat number per cut and Heatmark is not building inventory. Sold mainly to the AISC-certified and DOT-prequalified subset.
11. Progress-billing backup. Earned value by tonnage or line item against the schedule of values, with a backup table of pieces shipped. The PM always approves; nothing is sent automatically. (v2) CSV/IIF export first; a QuickBooks Online push only if 30%+ of pilots use it, because many family shops may run QuickBooks Desktop or Sage 100 Contractor (ASSUMPTION; interview question).
12. (v2) Galvanizer/painter handoff. Scan-out and scan-in at the outside vendor, where pieces actually go missing (shop-owner red team; ASSUMPTION to test).

Cut from the roadmap in v2: AI estimating from drawing PDFs. SketchDeck.ai already sells AI takeoff to steel estimators [S17]; it is no edge, and it competes for scarce engineering time. Partner or link out instead.

Explicitly not building: nesting, 3D modeling or detailing, CNC control, inventory, full accounting, payroll or timekeeping of record.

3.1 Integration reality (v2)

• The Tekla Open API is a .NET API that runs in-process with Tekla Structures on a Windows machine, so a cloud service cannot call it; using it means shipping a Windows plugin across Tekla versions. The Tekla Partners Program is a paid annual subscription with development-only licenses [S26] (per the CTO red team's reading of the product sheet; I did not re-read it this session). Trimble is both gatekeeper and competitor. Core input is export-only (KISS, NC1, CSV/XML report templates, IFC, PDF BOM). A plugin is a Y3 option.
• Beam-line OEM production software (for example from Peddinghaus, Voortman or FICEP) is treated as a competitor until proven otherwise, not only as a channel. Names from the CTO red team; product capabilities UNVERIFIED.
• The moat candidate is a test corpus of 50+ real US detailer file bundles, including mid-job revisions, collected with permission and under NDA. Correct handling of regional detailer quirks is the tedious-to-copy layer; the QR/Kanban/BOL layer is easy to copy (EZIIL in an estimated 3 to 9 months; CTO red team's estimate).

Practical realities (ASSUMPTIONS from reasoning): tags do not survive blasting, painting or galvanizing, so typed-mark search is first-class; shops have poor Wi-Fi and gloved users; "live in a week" applies to the scanning loop on new and in-flight jobs, not to job history.

4. Why now (rewritten in v2)

1. Incumbent re-decisions (EVIDENCE, narrower than v1 claimed).
   • FabTrol is gone [S5][S6].
   • Trimble's October 1, 2026 notice ends activation and re-hosting for perpetual licenses without valid maintenance after December 31, 2027; after that "a hardware failure will result in the permanent loss of the license" [S8]. The notice names no products and does not say licenses stop working. EPM maintenance products were in a June 2024 end-of-sale notice and PowerFab in a February 2025 subscriptions notice [S8].
   • What this means in practice: an affected shop is "one dead server away" from losing its license. Many will back up the box and keep running (shop-owner red team's view). Those shops are also mostly larger (40+ staff) and will want estimating, purchasing and inventory depth Heatmark will not have.
   • Use: a door-opener with existing PowerFab/EPM users ("before your server dies"), feeding the $900 tier. Zero base-case customers are attributed to it. In the 90-day interviews, find 5 perpetual EPM/PowerFab shops and ask what they plan to do on January 1, 2028.
2. Phones and cheap tablets are already in shops. Scanning needs no special hardware beyond tags and a label printer. Reasoning; no adoption statistic verified. Some shops restrict phones on the floor, hence kiosk tablets.
3. AI document extraction is now good enough for reviewed drafts of mill certs and PDF BOMs. SketchDeck.ai's steel-takeoff product [S17], and AI-fabrication bets like 1872 ($15M seed, July 2026 [S19]) and YC-backed Jeevy [S18], show investors see the same shift. The last two are fabricators, not software vendors.
4. A small AI-assisted team can build the wedge, but AI-assisted coding compresses writing code, not learning the domain's edge cases, supporting customers or handling 20 shops' worth of messy files. The hard parts are domain correctness, revisions and floor adoption.

Counterpoint: "why now" also applies to EZIIL (priced from $140/mo [S4], publishing SEO aimed at FabTrol orphans and US shops [S5][S22]) and to Factory (a $4.69M Series A reported January 2026, expanding into Texas [S16]). The window is real but contested.

5. Market size (bottom-up)

EVIDENCE: counts.

Size bands: UNVERIFIED. The official Census size-band table could not be pulled (the API returned "Missing Key" this session). Manufacturing establishment counts typically have a large under-5 and under-10 tail, and an average of 33 employees per establishment is consistent with that. This is test #1 in the executive summary.

ASSUMPTION: addressable shops, two estimates.

Share needed (v2 model, cumulative logos acquired over 5 years):

Conclusion: if the narrow band is right, even the base case needs roughly 1 in 4 of the real buyers to try Heatmark within 5 years, and the bull case is implausible on structural alone. The misc-metals segment and Canada are not optional for any outcome above base, and misc shops pay less (section 8).

Value of the market at realistic price: roughly 1,100 to 2,000 shops in the narrow structural-plus-misc band × about $650 × 12 ≈ $9 to 16M/yr of US spend if every target shop bought. That supports a profitable small company, not a venture-scale one.

6. Competition

Positioning:

• Steel-specific: reads detailer files and tracks piece marks, tonnage, loads and heats.
• US-built: AISC/US billing conventions, not EN 1090.
• Per-shop price with unlimited floor users.
• Live on the next job, and on open jobs, in a week.
• Works from the files you already get, including PDFs and mid-job revisions.

Against EZIIL, the pitch is depth on structural workflows (DSTV/KISS import, load lists/BOL, heat traceability, tonnage-based progress billing) with everything included. EZIIL charges extra for BOM, shipments and certificates [S4].

7. Defensibility and switching costs

There is no strong moat. A realistic defensibility stack:

1. Embedded data (moderate, grows with time). Every job's piece history, heat-to-piece links, shipped loads and hours. Once a shop has 12+ months of jobs and is using heat traceability for QA/audit records, switching means losing searchable history.
   • UNVERIFIED: the specific AISC certification clauses on material identification and traceability.
2. Workflow habit (moderate). Once foremen scan, the office stops calling the floor. Retraining a crew on a new app is real friction for a 20-person shop.
3. Integration depth (moderate, with partner risk; v2's main moat candidate). Maintained parsers for messy real-world KISS/CSV/NC1/PDF variants and revisions, backed by a test corpus of 50+ real detailer bundles. Each detailer's quirks become an asset.
   • Risk: Trimble controls Tekla's API and partner terms [S26] while also selling PowerFab.
4. Price and segment focus (weak). Large vendors rarely price down to $400/mo, but EZIIL already does [S4].
5. Brand and community (slow). Being "the one the small fabricators use" through fabricator-to-fabricator referral.

Switching in is cheap by design: start on the next job and load open jobs with the in-flight importer; no history import needed. Switching out should be possible (CSV/PDF export) to make the sale easier. The lock-in is habit and history, not hostage data.

8. Business model and pricing

Founders' hypothesis: $400/mo (<15 staff), $900/mo (15 to 40), $1,500/mo (>40), unlimited floor users; onboarding $2,500 to $5,000.

Test against evidence:

Verdict on price (v2):

• $400/mo small tier: exposed. It is about 3x EZIIL's headline. EZIIL's real comparable price is unknown and could be close to Heatmark's once add-ons are included; get 2 to 3 real quotes (test #2) before fixing the price. Frame the price as "one re-cut W-beam or one back-charge a month," backed by interview data. Offer $350/mo annual, $400 monthly.
• $900/mo tier: the core. v2 treats the 20-to-75-person shop as the real buyer, so the $900 tier carries the model and $400 is the entry point.
• $1,500/mo tier: drop as a list tier. Shops over 40 staff most often own PowerFab/StruMIS and want depth Heatmark won't have. Sell them $900 plus the Office add-on.
• Office add-on $300 to $500/mo: mill-cert storage/extraction and billing backup, aimed at AISC-certified and DOT-prequalified shops. Attach rate is the key price lever and is modeled explicitly (base 5% in Y1 rising to 30% in Y5; ASSUMPTION, untested).
• Onboarding fee: realized at 50% in v2. $3,000 list for structural, $1,500 for misc, waived with annual prepay at about half of deals (owners negotiate it away; EZIIL charges $350 [S4]). v1 booked 100% collection; in v1's base this overstated Y3-Y5 revenue by about $110-160k/yr.
• Price increases are counted only when they show up in renewals; the v2 model has none. Blended ARPA in v2 is computed from segment and attach, not hand-entered.

v2 ARPA by segment (ASSUMPTIONS): structural base $650/mo (a mix of $400 and $900 tiers), misc base $425/mo (mostly $400), plus attach × $400 (structural) or attach × $200 (misc). Result in base: structural $670 → $770, misc about $435 → $485, blended about $670 → $690, flat because the misc share grows while attach rises. The vertical-SaaS red team predicted $550 to 650; v2 is higher only because of the attach assumption, which the deposit test must check.

Other revenue (ASSUMPTION, not modeled): tag and label consumables at small margin; paid connectors for unusual detailer formats; metered AI pages with a monthly allowance and overage (section 12).

Payment terms: annual prepay discount 10 to 15%; ACH preferred. Annual contracts only after the pilot, priced per shop, no pause option (project shops will ask to pause between jobs).

9. Go-to-market and customer acquisition

Motion: founder-led, on-site demo selling on the shop's own files, then fabricator-to-fabricator referral. Not self-serve. This trade buys from someone who has stood on their floor (shop-owner red team).

Who decides and who can veto (ASSUMPTION from the shop-owner red team): the owner signs. Three people can veto: the foreman/superintendent, who must make the crew scan; the office manager/bookkeeper (often family), who owns QuickBooks and billing; and often a second-generation son or daughter, who is the real champion. Demos must win all three.

Channels, in v2 priority order (ASSUMPTIONS; all untested):

1. Steel service centers and mill depots. They deliver to every small shop, already produce the mill certs, and their sales reps know which shops are busy. Build the interview list through 2 to 3 friendly reps.
2. Beam-line and saw equipment dealers and their service techs.
3. Paint and galvanizing subcontractors. Every shop ships to them; they see the handoff pain.
4. Regional fabricator associations and NOMMA (NOMMA details UNVERIFIED). One national show per year.
5. GCs and erectors who suffer from short-shipped loads.
6. Detailers: downgraded. They are paid per job by the fabricator, are often offshore, and have no reason to sell for Heatmark. The free report template is still worth shipping.
7. Cold outreach from AISC directories and DOT prequalification lists: downgraded. This buyer is on the floor or bidding and rarely answers unknown email.
8. The Tekla deadline: a door-opener for existing PowerFab/EPM users only (section 4).

Funnel (v2 base, ASSUMPTIONS): demo-to-close 22% (the vertical-SaaS red team suggested 20 to 25%).

Capacity assumption: an AE closes about 45 new shops/yr, a full-time founder about 35 (ASSUMPTION; the red team suggested 40 to 60 per AE). New logos are capped by capacity in every scenario.

CAC (ASSUMPTION): Y5 base sales, shows and digital spend is about $450k for 100 new shops, about $4.5k paid; $5k to $9k fully loaded with founder time. Payback about 9 to 15 months at about $690 ARPA and 85% subscription gross margin. Not benchmarked.

Sales cycle (ASSUMPTION): 3 to 8 weeks to a paid pilot for small shops; longer for shops timed to a PowerFab renewal. Seasonality: shops buy when backlog is thin (winter, between big jobs), which is also when they cut costs; push sales in Q4/Q1.

Onboarding effort (ASSUMPTION): 30 hours per shop in Y1-Y2, falling to 20 by Y4 as the in-flight importer and templates mature: config, tags/printer, import of open jobs, foreman training, two weeks of check-ins. The first 10 shops get an on-site day each. If more than about 30% of first-year customer hours go to bespoke imports or on-site work, the business is becoming a services company.

10. Product and build plan

Before customer #1 (target: month 4 to 6):

• Multi-tenant web app, auth, roles (owner/office/floor).
• Import: KISS, CSV/XML report template, DSTV/NC1, and PDF BOM → piece list with human review (v2).
• Revision-diff engine and event-log data model (v2).
• QR tag generation (PDF for label printers).
• Android-first offline scan app in a native wrapper, kiosk mode for a shared tablet; idempotent, conflict-visible sync (v2; budget 4 to 6 weeks for this alone).
• Job board with % complete by count and tonnage.
• Load list builder, truck capacity check, BOL PDF, local offline fallback (v2).
• In-flight job import (v2).
• Basic job hours (labeled job costing, not timekeeping).
• CSV export of everything; audit log; daily backups.
• Demo tenant "Demo Iron Works" with a demo-data banner on every screen.
• Test corpus: every real bundle from interviews and pilots, with permission, becomes a regression test (v2).

Before customer #40 (gated: build a module only when 10 paying shops ask for it): (v2: moved from "before #20" and cut)

• Office add-on V1: mill-cert storage, search and human-confirmed extraction; progress-billing backup with CSV/IIF export.
• Galvanizer/painter handoff scans.
• Holds, NCR/rework flags, missing-piece report, scan-health alerts.
• Owner dashboard: estimated vs actual hours, earned vs billed.
• SOC 2-style controls documented (not certified).

Later (Y3+), each gated on demand: heat-at-cut via SigmaNEST/saw-line reports; QuickBooks Online push; Tekla plugin; misc-metals templates (stairs, rails, gates); erector/field receiving app; Canada.

Dropped: AI estimating (section 3).

Build cost (ASSUMPTION): one founder full-time as product/engineering lead from month 6; $40k contract engineering/QA/security in Y1, $60k in Y2, then about 1 FTE in Y3 ($150k), 1.5 in Y4 ($250k) and 2 in Y5 ($300k). The CTO red team proposed 1 FTE from Y2 and 2 from Y3; v2 is slightly leaner because a full-time founder covers Y2 and AI estimating is cut.

11. Operating model (v2)

Support model (ASSUMPTION): about 1.25 tickets per shop per month at about 35 minutes (9 hours per shop per year), plus a 30% overhead for coverage from 5:30am to 3pm local across US time zones, an on-call rotation for ship mornings, and travel. The first paid implementer starts at about shop #15 (Y2), not #30 as in v1. "Ship-morning incidents per month" is a pilot metric.

Is it secretly a full-time job? Yes. Through Y3 it is at least one full-time founder and probably 1.5: demos on each prospect's files (136 to 268 a year in Y2-Y3), onboarding, early-morning support, parser maintenance and managing 3 to 5 contractors. The brief's 10 to 15 hrs/wk combined is realistic only from about Y4 with a paid GM.

Outsourced: bookkeeping/tax, legal, annual pen test, design, and possibly SDR/appointment setting.

12. Financial plan (v2 base case; all ASSUMPTIONS; see model.json for bear, bull and stress)

Key assumptions (each rests on reasoning, not observed data, unless noted):

• New logos per year by segment, capped by seller capacity (section 9).
• Annual logo churn 18% (v1: 12%). Bull 12%, bear 25%. Basis: small construction subcontractors in a cyclical industry, with closures, ownership changes and "the foreman quit and nobody scans." No verified SMB churn benchmark this session.
• ARPA by segment plus Office add-on attach (section 8).
• Onboarding fee $3,000 structural / $1,500 misc list, 50% realized.
• Y1 pilots at 50% for about the first 3 months, then list (about a 25% effective Y1 discount). This is consistent with gate G2 (list price by month 12).
• Hosting/AI 6% of subscription revenue, $12k floor; sensitivity at 12% for heavy AI use.
• 2% payment processing.
• Implementation/support from hours (sections 9 and 11) at $90k per FTE.
• AE $95k loaded, SDR $60k, 10% commission on new first-year ACV.
• Management: founder-operator from month 6, then a GM (section 11).
• $150k or $200k starting capital.

Notes:

• The criterion is judged on the fully loaded line. v1's 57% Y5 margin came from paying nobody to run the company, about one engineer, about $110k per seller and 12% churn.
• Subscription gross margin (after hosting/AI, processing and ongoing support labor, before onboarding labor) is about 85%, one figure used everywhere (v1 used both 85% and 90%).
• "Deferred" founder pay is a real liability. If it is repaid out of Y5+ cash, Y5-Y6 distributions go mostly to that.

Scenario summary (from model.json):

The downturn stress test cuts Y4 new logos by 40% and adds 10 points of churn in Y4 (construction slowdowns hit all customers at once). The bear case needs either more capital or a cost cut it does not show; in practice the bear case is a stop at gate G2 or G3.

Unit economics (base, ASSUMPTIONS):

• ARPA about $690/mo; subscription gross margin about 85%.
• Fully loaded CAC $5k to $9k; payback about 9 to 15 months.
• LTV capped at a 5-year life: about $690 × 12 × 85% × 5 ≈ $35k (v1: $50 to 70k on an 8.3-year implied life).
• LTV/CAC about 4 to 7x.
• Leading churn indicator: scans per piece and % of pieces with a ship scan, per shop per week.

13. Capital plan (v2)

Recommendation: commit $200k, not $150k, in tranches, and only if one founder works full-time from month 6 on deferred pay. At $150k the base case's low point is about $7k, which leaves no room for a slow quarter. Without deferral, no scenario fits within $200k.

Outside money: not needed in the base case only because founder pay is deferred. If the bull case shows up, revenue-based financing or a small SBA-style line beats equity for the hiring ramp.

What capital does not buy: the founders' time. That is the scarce input, and v2 prices it.

14. Owner outcomes (v2)

Comparable multiples (EVIDENCE plus caveats):

• SaaS Capital (Jan 24, 2025): predicted private SaaS multiples of 4.8x ARR for bootstrapped and 5.3x for equity-backed companies, from a public SaaS index median of 7.0x at year-end 2024 [S20]. The page gives no ARR size range for its dataset, and growth and net retention are inputs to its model (vertical-SaaS red team's reading). Applying it to a $1.7M ARR, founder-dependent, construction-cyclical SMB tool is a stretch.
• UNVERIFIED this session: vertical-software acquirers of the Constellation/Valsoft/Volaris type are said to pay roughly 1.5 to 3.5x ARR for niche SMB vertical software at this size and to underwrite on EBITDA; brokered deals under about $3M ARR are commonly priced on SDE/EBITDA multiples.
• Niche comparable: FabTrol reportedly sold to AVEVA for $6.5M (2015), per EZIIL, unverified against a primary source [S5]. Trimble acquired FabSuite's assets in 2018, terms not found [S7].

v2 method: value on three bases and use the financial-buyer ARR basis as the base case, the EBITDA basis as the floor, and the strategic ARR basis as upside. (Partial pushback on the red team, which asked for the lower of EBITDA and ARR: in Y5 the base case is still spending heavily on growth, so its fully loaded EBITDA of about $140k understates what any buyer would pay for $1.7M of recurring revenue. The EBITDA basis is shown as a floor.)

Plainly:

• $500k to $1M/yr profit: not in the base case. Only the bull case reaches it, at about $0.53M distributable fully loaded in Y5. Subjective likelihood conditional on passing G0: 5 to 15%; unconditional, before validation: lower, roughly 2 to 8%. Basis: reasoning; it needs all bull drivers at once.
• $10M value in 3 to 5 years: not in the base case. It needs the bull case and a strategic buyer (realistically Trimble, which also controls Tekla API access), or the top of the financial range in the bull case. Subjective likelihood conditional on passing G0: 3 to 10%. No scenario reaches $10M at Y3.
• v1 put $10M at 10 to 20% while showing a base range that straddled $10M. v2 uses one framing: the base case does not reach $10M.

Plausible acquirers (ASSUMPTION): Trimble [S7]; construction-software consolidators and vertical-market-software holding companies; nesting/CAM vendors; detailing-software owners (SDS2's owner UNVERIFIED). AVEVA has exited the category.

15. Risks and mitigations

Fatal flaws, stated plainly: (1) the time criterion fails; (2) at realistic churn and costs, the profit criterion is met only in the bull case. Neither kills the business; both mean it likely does not meet this brief.

16. Legal and regulatory checklist (factual items to confirm with counsel; not legal advice)

• [ ] Entity and IP: OpCo structure, founder and contractor IP assignment and confidentiality agreements. Document the deferred founder pay (accrual terms, priority, what happens on sale or shutdown).
• [ ] Trademark: knockout search for HEATMARK and alternates in software classes.
• [ ] Customer contracts: MSA/ToS, order form, SLA (uptime, ship-morning support hours), data processing terms, limitation of liability. Explicit disclaimer that AI-extracted mill-cert fields and billing drafts must be reviewed by the customer, and that Heatmark records do not certify compliance with AISC, BABA or DOT requirements.
• [ ] Customer data and drawings: contract drawings are typically owned by the owner/architect/engineer/GC and licensed to the fabricator. Use restricted to providing the service. No model training on customer documents without written opt-in; AI vendors with no-training/zero-retention terms. The test corpus needs explicit written permission per bundle.
• [ ] Time records: clock-in/out per job can become a wage-hour record in practice (FLSA recordkeeping; California meal and rest rules). Label it "job costing, not timekeeping," no payroll export in V1, and say so in the ToS. UNVERIFIED this session.
• [ ] Employee data: crew names, time records, possibly photos; assess state privacy laws; notice and consent if location is captured.
• [ ] Sales tax on SaaS: taxability varies by state; register where nexus exists; use a tax engine (about $5 to 10k/yr, now in G&A). UNVERIFIED this session.
• [ ] Traceability and QA: read the AISC 207 material-identification clause and BABA documentation rules before any marketing claim. Never market "AISC traceability compliant." UNVERIFIED this session.
• [ ] File formats and APIs: license terms for DSTV NC documentation, KISS, IFC, and the Tekla Partners Program [S26]. No reverse engineering that breaches EULAs.
• [ ] Security: written policy, encryption, backups, access controls, annual pen test. SOC 2 Type I in about Y4 (about $20 to 40k, now in G&A), only when a GC deal requires it; GC security questionnaires from about Y2.
• [ ] Insurance: tech E&O plus cyber; general liability for on-site visits.
• [ ] Marketing claims: no customer logos, testimonials or savings claims without written permission and substantiation. All demo data labeled.

17. 90-day plan and kill criteria (v2)

Before day 1 (one afternoon, $0): Census size-band check (test #1). Free API key; CBP establishments by employment size class for 332312 and 332323. Stop if fewer than about 700 structural establishments have 20 to 99 employees.

Days 1 to 30: listen.

• Build the list of 150 target shops (20 to 75 staff, structural first, some misc) through 2 to 3 steel service-center reps and one equipment dealer, not directories.
• Book 20 owner/GM conversations, on-site where possible: 12 structural, 3 misc, 5 known PowerFab/EPM-perpetual users (ask what they plan to do on January 1, 2028).
• First question: "What files did you get on your last 3 jobs? Send them."
• Then ask about: how they track pieces today; lost or re-cut pieces in the last 90 days and the cost in dollars; how the load list and BOL get made; how progress billing backup is built; what accounting system and what CNC/beam-line software they run; what they pay for software; whether they evaluated PowerFab, StruMIS or EZIIL and why they said no; who on the floor and in the office would have to use it.
• Request 2 to 3 real EZIIL quotes with BOM, shipments and certificate modules (test #2).
• Collect 3+ real file bundles (one with a mid-job revision) and 3+ mill-cert PDFs, with permission and NDA.

Days 31 to 60: show.

• Build "Demo Iron Works": import → revision diff → tag → scan → board → load list/BOL.
• Demo on each interviewee's own files, with the foreman and office manager present where possible.
• Ask for a refundable deposit or an order conditional on a working pilot at $400+/mo (test #3). Friends-of-friends deposits are reported separately.
• Test $300 vs $400 for small shops; test Office add-on attach at $300 vs $500.

Days 61 to 90: pilot.

• 3 to 5 shops on one live job each (new or in-flight).
• Measure per shop: scans per piece; % of pieces with a ship scan; onboarding hours; ship-morning incidents; parse accuracy and AI cost per document; one quantified "saved" event.

Kill or redesign criteria (any one triggers it):

1. Census check shows fewer than about 700 structural establishments at 20 to 99 staff. Stop, or redesign around misc-metals plus Canada.
2. Fewer than half of interviewed shops get machine-readable piece lists and PDF-BOM import fails on their files. Redesign toward beam-line shops only, which shrinks the market further.
3. Fewer than 5 of 20 sign a deposit or conditional order at ≥ $400/mo. (v1 required 8 of 20 signed orders before production code existed, which the shop-owner red team argued would kill a good business on a false negative; I agree.)
4. Fewer than 3 of those reach 50%+ ship scans by week 3 of a live job.
5. Fewer than 5 owners give a dollar figure for a recurring cost of lost, re-cut or mis-shipped pieces.
6. In head-to-head comparison with real EZIIL quotes, most owners say EZIIL or a spreadsheet is "good enough."
7. Pilot onboarding averages more than 40 hours or needs repeated on-site visits.
8. No founder is willing to go full-time from month 6 on deferred pay. Then stop: the model does not work part-time.

18. Brand and sources

18.1 Brand

Primary: Heatmark. It joins the two identifiers every fabricator lives by: the heat number (mill traceability) and the piece mark (shop tracking). It is short, steel-native, and not generic.

Six .com candidates (availability NOT checked):

1. heatmark.com (likely registered; common compound)
2. getheatmark.com
3. heatmarkhq.com
4. piecetrail.com
5. steeltrail.com
6. markandheat.com

Tagline: "Every piece, every heat, every load. Tracked on the floor, billed in the office." Palette, fonts and voice are in brand.json.

18.2 Sources (all accessed 2026-11-13)

• [S1] SoftwareConnect, Tekla PowerFab review: https://softwareconnect.com/reviews/tekla-powerfab/
• [S2] Capterra, Tekla PowerFab listing (pricing as shown in search summary; page blocked direct fetch): https://www.capterra.com/p/10030015/Tekla-PowerFab/
• [S3] G2, Tekla PowerFab pricing: https://g2.com/products/tekla-powerfab/pricing
• [S4] EZIIL pricing: https://eziil.com/pricing/
• [S5] EZIIL, "FabTrol MRP Discontinued? What's Next" (competitor source; dates and $6.5M AVEVA purchase unverified against a primary source): https://eziil.com/fabtrol-mrp-status/
• [S6] Trimble, "Making the transition from FabTrol to Tekla PowerFab": https://www.trimble.com/en/resources/construction/video/making-the-transition-from-fabtrol-to-tekla-powerfab and https://tekla.com/us/resources/webinars/making-the-transition-from-fabtrol-to-tekla-powerfab
• [S7] Trimble press release, acquisition of FabSuite assets (2018-04-20): https://news.trimble.com/2018-04-20-Trimble-Acquires-the-Assets-of-FabSuite-to-Expand-its-Steel-Fabrication-Software-Portfolio
• [S8] Tekla, End-of-life notices (perpetual maintenance retirement, notice dated Oct 1, 2026; deadline Dec 31, 2027): https://www.tekla.com/terms-and-conditions/end-of-life
• [S9] Tekla support, End of Sales of Tekla Maintenance Products for Perpetual License: https://support.tekla.com/article/end-of-sales-of-tekla-maintenance-products-for-perpetual-license (now redirects to help.trimble.com)
• [S11] Vanta Insights, NAICS 332312 (Census CBP 2023: 3,434 establishments, 114,153 employees): https://vantainsights.com/industry/332312-fabricated-structural-metal-manufacturing
• [S12] Mewayz industry data, NAICS 332312 and 332323 (aggregator; year unstated): https://data.mewayz.com/industry/332312 and https://data.mewayz.com/industry/332323
• [S13] NAICS Association, sector 332 table (332323: 2,540): https://www.naics.com/naics-code-description/?code=332
• [S14] Census (archived), NAICS 332312 definition: https://webarchive.library.unt.edu/eot2008/20081106151529/http://www.census.gov/epcd/ec97/def/332312.HTM
• [S15] STRUMIS: Capterra https://www.capterra.com/p/10032743/STRUMIS/ ; Softwarefinder https://softwarefinder.com/manufacturing-software/strumis ; SoftwareConnect https://softwareconnect.com/reviews/strumis/
• [S16] Startup Daily, Factory $4.69M Series A (Jan 28, 2026): https://www.startupdaily.net/topic/funding/manufacturing-software-platform-factory-constructs-4-69-million-series-a/
• [S16b] Startup Researcher, Factory US$3.25M: https://www.startupresearcher.com/news/manufacturing-platform-factory-secures-usd3-25-million-for-us-push
• [S17] Taproot Edmonton, SketchDeck.ai investment (Apr 28, 2026): https://edmonton.taproot.news/briefs/2026/04/28/steel-estimation-startup-looks-to-expand-after-investment
• [S17b] Signalbase, SketchDeck.ai $150k seed: https://www.trysignalbase.com/news/funding/revolutionizing-steel-industry-estimations-sketchdeck.ai-secures-150k-in-seed-round-funding-for-lift
• [S18] Y Combinator, Jeevy Fabrication: https://ycombinator.com/companies/jeevy-fabrication
• [S19] AIST, 1872 closes US$15M seed: https://www.aist.org/1872-closes-us$15-million-seed-round-to-open-autonomous-steel-fabrication-factory
• [S20] SaaS Capital, Private SaaS Company Valuations (Jan 24, 2025): https://www.saas-capital.com/blog-posts/private-saas-company-valuations-multiples/
• [S21] EZIIL, Tekla PowerFab vs EZIIL (competitor claims): https://eziil.com/tekla-powerfab-vs-eziil-for-small-steel-fabricators/
• [S22] EZIIL, 8 Best Production Tracking Software for Steel Fabricators: https://eziil.com/best-production-tracking-software/ ; 16 Best Metal Fabrication Software: https://eziil.com/?p=4148
• [S23] Capterra AU, Tekla Structures reviews: https://www.capterra.com.au/reviews/92964/tekla-structures
• [S24] Msuite, 5 signs your fabrication shop has outgrown spreadsheets: https://www.msuite.com/5-signs-your-fabrication-shop-has-outgrown-its-spreadsheets/
• [S25] Modern Steel Construction, Quality Corner (Jan 2006; 611 AISC-certified facilities): https://www.aisc.org/globalassets/modern-steel/archives/2006/01/2006v01_quality_corner.pdf
• [S26] Tekla Partners Program technical product sheet: https://developer.tekla.com/tekla-partners-program-suite-technical-product-sheet

• [S27] FRED, Employment for Fabricated Structural Metals (NAICS 332312): https://fred.stlouisfed.org/data/IPUEN332312W200000000

• Census Bureau API (fetched directly this session; returned a "Missing Key" page, so size bands were not retrieved): https://api.census.gov/data/2022/cbp

Not verified (search budget exhausted): Census size-band distribution for 332312/332323; small-SaaS marketplace, broker and vertical-acquirer multiples; SaaS Capital's ARR size distribution; SMB vertical-SaaS churn benchmarks; state SaaS sales-tax rules; AISC 207 traceability clauses; BABA documentation rules; trade-show and association details; current owners of SDS2 and SigmaNEST; Tekla Open API runtime model and PowerFab API terms (CTO red team's description, not re-checked); PowerFab Go capabilities; Canadian fabricator counts.

19. Red team and response

Three red teams reviewed v1: a bootstrapped vertical-SaaS operator (VS), a long-time small structural/misc shop owner (SO) and a CTO (CTO). All three reported that the shared search budget was exhausted; between them they fetched the SaaS Capital page, EZIIL's pricing, FabTrol-status and roundup pages, the Tekla end-of-life page and the Tekla Partners Program product sheet. I re-fetched the Tekla end-of-life page this session and it matches their reading. My own one search attempt was refused (budget exhausted), and a direct Census API call returned "Missing Key," so no new market facts were verified in this revision.

Verdict on the critiques overall: they are mostly right, and the combined effect is larger than any single reviewer estimated. VS and CTO each re-costed v1's 270-shop base and landed at about $0.7 to 1.2M Y5 operating profit. Once churn, onboarding-fee realization, segment pricing and funnel capacity are also corrected, v2's base reaches only about 209 shops and about $0.14M fully loaded profit. I accept the direction of every critique and push back on four specific numbers.

Accepted and changed

Pushed back, with reasoning

What survives

• The capital claim holds only at $200k with about $825k of cumulative deferred founder pay through Y5.
• The product wedge (handoff scanning → load list/BOL, robust to real files and revisions) is still a sensible, buildable product.
• The 90-day validation is cheap (under about $15k), and the first two tests cost almost nothing.
• Against this brief, the honest answer is that Heatmark probably fails: it needs a full-time founder, and the profit and value targets appear only in the bull case.