Pre-weld fit-up OS for structural steel fabs that qualifies joints for autonomous welding, routes exceptions early, and cuts scrap.

1. Adaptive welding has shifted from a bespoke robotics promise into a productized workflow, which means more fabricators will soon need software around the cell rather than another custom integration project.
2. Misalignments, tacks, and fit-up variation are explicitly named as the root causes of scrap and rework, so there is now a concrete upstream workflow to productize instead of a vague "AI for welding" claim.
3. Buyers already value the outcome in reduced rework, grinding, and scrap, which gives a new vendor a measurable ROI story tied to throughput and margin.
4. Traceability is becoming part of the welding stack itself, so a decision layer that records why each joint was auto-welded, prepped, or rerouted fits the direction of the market.
5. Structural steel, data centres, and modular construction are already named target sectors, giving the startup a narrow first customer set with repetitive geometry and urgent delivery schedules.

Catalyst. Novarc and Yaskawa are productizing adaptive welding intelligence around the exact pain of fit-up variation, which makes upstream joint-readiness software newly urgent because more shops can now automate if they can trust the handoff into the cell.

The product sits at the fit-up station before the robotic welding cell. It captures joint measurements, images, tack locations, part metadata, and job requirements, then scores whether that assembly falls within a validated process window for a given robot, fixture, and weld program. If the job is borderline, the software recommends the smallest prep action needed such as re-tacking, shimming, grinding, or manual routing, instead of letting the cell discover the problem expensively mid-weld. Each decision generates a traceable record linking as-built joint condition, route taken, and resulting quality outcome, so plants can learn which upstream conditions actually cause scrap. Over time the system becomes the decision layer that keeps adaptive cells fed with winnable work rather than just recording failures after they happen.

What's different. Robot OEM software decides how to execute a weld once the job is already in the cell, and weld-monitoring tools usually explain defects after the fact. This product owns the earlier routing decision: is the joint ready for autonomous welding, what prep action will make it ready, and when should it bypass the cell entirely. Its defensibility comes from the cross-program dataset linking fit-up conditions to actual weld outcomes, which compounds faster than any one fabricator's tribal playbook or an OEM's single-cell telemetry.

Jobs to be done

flowchart LR
  Buyer[Welding engineering lead] --> Pain[Fit-up variation creates scrap and robot downtime]
  Pain --> Product[Preflight OS scores joint readiness and routes prep]
  Product --> Outcome[Higher first-pass yield and trusted autonomous welding]

Executive takeaways

• The whitespace is upstream, not in-cell: OEMs and autonomy vendors are improving seam finding, adaptive welding, and connected-cell intelligence, but fit-up readiness, prep routing, and auditability before the robot still sit mostly in manual workflows [1][6][21][26][27][31][32].
• Buyer urgency is strongest where repetitive structural-steel or enclosure programs meet AI/data-center schedule pressure and a shrinking welding labor pool [2][3][5][15][16][35][36][37].
• The budget path is operational, not experimental: shops already price automation using rework, overtime, productivity, and break-even math, so a preflight layer can sell as robot-uptime insurance rather than generic AI software [17][18][20][25][38].
• Competition is adjacent but credible: Novarc, Path, Hirebotics, Yaskawa, and FANUC all sell execution or control layers around the weld cell, so the startup must stay OEM-neutral and prove value on mixed fleets [1][12][15][21][24][31][33][34].
• The beachhead looks real but not enormous; structural steel is a strong entry wedge, but venture scale likely requires expansion into adjacent heavy-fab categories already named by OEMs, including shipbuilding, heavy equipment, mining, agriculture, and modular [1][13][29][31][35].

Market definition

Decision software that sits between fit-up/prep and robotic arc welding, scoring whether a joint is inside a proven process window, recommending minimal prep, and routing work to auto-weld, prep, or manual lanes before the cell consumes time or scrap. The beachhead is repetitive structural-steel, rack/enclosure, and modular assemblies where cell uptime and first-pass yield matter more than generic robot analytics [1][15][16][21][22][31][32][34].

Customer and buyer

Daily users are welding engineering managers, quality managers, and fit-up supervisors at structural-steel and heavy-fabrication shops. The economic buyer is typically a VP of operations, plant manager, or director of welding automation who already owns robot uptime, staffing pressure, and quality escape risk. The first-account profile is a multi-line fabricator launching or scaling repetitive robotic welding on data-center or modular programs [15][16][17][18].

Buying triggers

• A new robot cell launch or expansion into repetitive data-center, rack, enclosure, or modular work creates a need to keep the cell fed with qualified joints. [1][15][16][22][35]
• A visible scrap, grinding, or manual rescue spike makes management question robot ROI and forces a closer look at fit-up readiness upstream of the cell. [1][6][17][19][23]
• Welder shortages, overtime, and setup bottlenecks make supervisor-judgment routing too expensive to scale. [5][17][18][39]

Willingness to pay

Willingness to pay is framed through operating metrics buyers already track: overtime, turnover, rework, setup time, and robot utilization. Hirebotics makes that explicit with ROI calculators and break-even logic, while the IMS case shows buyers reward solutions that add capacity without new headcount. That supports a pricing conversation anchored in uptime and first-pass yield, not experimental software spend. [17][18][20][25][38]

Category dynamics

Growth signal 17% CAGR in Americas data-center capacity through 2030

Validation signals

• Novarc and Yaskawa explicitly frame structural steel, heavy equipment, data centers, mining, agriculture, and modular as target sectors for adaptive welding intelligence.
• Data-center construction demand is strong enough to influence structural steel consumption and capacity planning.
• Welding labor remains structurally constrained, making capacity-boosting automation easier to justify.
• Both Path and Hirebotics publish customer evidence where automation added output without simply adding more welders.
• The sensing and monitoring stack needed for preflight capture already exists commercially from OEMs and specialist vendors.

Regulatory & technical constraints

• Any workflow influencing structural-steel routing has to coexist with AWS welding standards, robotic arc-welding expectations, and shop quality procedures.
• AISC-certified fabricators are conditioned to documented quality systems and error prevention, so recommendations need traceability and reviewability.
• OSHA robot and welding safety requirements mean fit-up capture must not interfere with guarded-cell workflows or risk-assessment obligations.
• Measurement hardware has to remain reliable in harsh welding environments with spatter, heat, glare, and changing joint geometry.

Today’s alternatives split across five layers: robot OEMs and pre-engineered cells, autonomous cell vendors, cobot/app-first vendors, seam-tracking and inspection vendors, and internal supervisor judgment. That means rivalry is real, but no incumbent clearly owns the upstream is-this-joint-ready-for-autonomous-welding decision across mixed robot fleets [1][12][15][21][24][27][31][33][34].

Why incumbents do not win by default

• Robot OEMs. Yaskawa, FANUC, and peers own the execution stack, sensing accessories, and integrator channels, but they are optimized around making the cell run, not around being an OEM-neutral system of record for pre-weld routing across mixed fleets.
• Autonomous cell vendors. Novarc and Path are closest to the startup’s technical ambition, yet their center of gravity is still in-cell autonomy and cell sales or RaaS; the whitespace is the upstream decision layer that can work across robot brands and customer process windows.
• Cobot app-first vendors. Hirebotics proves buyers want easy programming, cloud telemetry, and ROI visibility, but its strength is quick deployment of simple repetitive welding, not structural-steel-specific fit-up qualification and exception routing.
• Sensor and inspection vendors. Keyence and Binzel provide critical sensing, seam tracking, and inspection primitives, but they stop at measurement and tooling rather than owning the go/no-go workflow for the joint family.

Weld Fit-up Preflight should start as an OEM-neutral decision layer for North American structural-steel fabricators that already run robotic welding cells on repetitive data-center and modular assemblies but still route borderline joints through supervisor judgment. The urgent pain is not programming the robot; it is deciding, before arc start, whether a real joint with gap drift, tack variation, or light misalignment is safe to auto-weld, needs a small prep step, or should bypass the cell. The first product should capture a minimum measurement-and-photo package at the fit-up station, score readiness against a validated process window for a specific joint family, and create an auditable route record tied to downstream quality outcomes. The first buyer should be a VP of operations or director of welding automation at a 200-1,000 employee fabricator with 2-6 robotic cells and repeat work on beam, bracket, or base-plate assemblies. Pricing should be anchored to active robotic welding cells plus onboarding for joint-family qualification because buyers already evaluate automation using uptime, rework, grinding, and labor math rather than seat count. The hard strategic choice is to win one narrow upstream workflow in structural steel before expanding into heavier fabrication categories or into in-cell execution features that OEMs already pursue. Market evidence is strong on pain, buyer trigger, and adjacent competition, but still weak on acceptable operator-input burden and the exact readiness thresholds that predict downstream defects. The company should treat the first 12-18 months as a proof period for workflow adoption, mixed fleet compatibility, and pilot-to-production conversion rather than assume the category already exists.

Problem

• Structural-steel shops can now buy adaptive welding cells, but they still decide robot-ready versus prep-versus-manual routing with tribal rules, manual gauges, and visual judgment at the fit-up station.
• One bad routing decision creates expensive downstream effects including overwelding, grinding, scrap, manual rescue, and lower trust in robotic welding ROI.

Solution

• Build a pre-weld qualification layer that captures gap and tack measurements, photos, joint specs, and prior cell outcomes to score whether a given assembly is inside a proven process window for autonomous welding.
• Recommend the smallest corrective action such as re-tacking, shimming, grinding, or manual routing before the cell consumes time, then store a traceable record linking as-built joint condition, route choice, and quality outcome.

Why we win

• The product owns the upstream go or no-go routing decision that OEM software and weld-monitoring tools usually do not own across mixed robot fleets.
• If it compounds a cross-program dataset of joint condition, prep action, route taken, and weld result, it can build a process-window library and audit trail that internal playbooks and single-cell vendors do not have.

Milestones

flowchart LR
  Wedge[Structural steel preflight wedge] --> MVP[Fit-up station MVP]
  MVP --> Proof[Lower rescue welding and auditable routing]
  Proof --> Expansion[More cells plants and adjacent heavy fabrication]

Founding team

Experiment roadmap

Risk assessment

What must be true

• At least 5 target fabricators confirm that pre-weld readiness routing is a budget-worthy problem separate from generic robot programming or post-weld inspection.
• A minimum measurement-plus-photo workflow predicts route decisions with at least 80% reviewer agreement on the first target joint family.
• Operators can complete required capture in under 60 seconds per joint or automated sensing can replace the slow steps without breaking pilot economics.
• At least half of paid pilots convert into annual production deployments covering 2 or more cells within 6 months of pilot completion.
• OEMs and integrators allow enough data access and channel cooperation for the company to stay mixed-fleet rather than being trapped inside one vendor stack.

Open diligence questions

• What exact KPI unlocks budget fastest in the first account, robot uptime, grinding reduction, scrap reduction, or traceability?
• How repetitive are the first targeted joint families in real plants, and what share of routed work is predictable enough for software to learn quickly?
• What minimum sensing package yields trusted readiness scores without making the fit-up station slower?
• Will OEMs and integrators co-sell a neutral preflight layer, ignore it, or actively treat it as feature creep?
• Does the structural-steel beachhead alone support a large enough early company, or must adjacent heavy-fabrication expansion start sooner?

Model sanity

• Revenue engine. Blended pilot-then-subscription model: $55K one-time pilot fees provide early cash while $100K/year per-plant subscriptions compound from 3 plants (Y1) to 40 plants (Y3), with ARR reaching $4.0M and EBITDA turning positive in Q2Y3 on pre-seed alone.
• Must go right. Pilot-to-production conversion must hold at 67%+ across diverse structural-steel fabricators, because the entire 3-to-10-to-40-plant ramp depends on paid pilots converting rather than stalling as one-off proof-of-concepts that never reach annual subscription.
• Model breaks if. ARPU compresses below $75K/plant—either because OEMs bundle basic fit-up checks or pricing resistance forces seat-based packaging—which cuts Y3 revenue by $780K+, pushes EBITDA breakeven into Y4, and reduces cash to a $900K floor that may require a bridge.
• Next-round proof. Reaching 10 production plants with $1.0M ARR and a proven deployment template by Y2 end demonstrates repeatable beachhead economics and a credible path to the $4.8M SOM, justifying a $5M+ seed round to accelerate the 10-to-40-plant sprint.

Scenarios

Sensitivity

flowchart LR
  Fabricators[Structural-steel fabricators\n2-6 robotic cells] --> Discovery[Discovery and\ndesign-partner]
  Discovery --> Pilot[Paid pilot\n$55K one-time]
  Pilot --> Conversion{67% convert}
  Conversion -->|yes| Production[Annual subscription\n$100K per plant per year]
  Conversion -->|no| Lost[Lost opportunity]
  Production --> GrossProfit[Gross profit\n70-77% GM by Y3]
  GrossProfit --> EBITDA[EBITDA breakeven\nQ2Y3]
  EBITDA --> Cash[Cash floor\n$1.4M Q1Y3\nno new capital needed]
  Production --> Expansion[Expand cells\nand plants]
  Expansion --> Production

Flags: Y1 gross margin 57% is below 70% target due to pilot-heavy revenue mix (65% one-time fees); margin normalises to 75-77% by Y2-Y3 as subscription dominates · Cash floor of $1.4M in Q1Y3 (month 27) leaves only 7-8 months of then-current operating runway; any Y3 revenue miss or unplanned opex spike could require an emergency bridge before Q2Y3 cashflow-positive inflection · Structural-steel beachhead SOM $4.8M at 40 plants may be too small for Series A metrics without demonstrated adjacent-sector expansion into heavy equipment or shipbuilding by late Y2 · Pilot conversion rate of 67% is an unproven assumption; BP kill criteria require 50%+ and BP funnelTargets state 50%+ as the floor; model is sensitive to even a 10-point conversion drop (see sensitivity row 3) · No seed round is modelled in the base case; the company reaches EBITDA breakeven organically in Q2Y3 but a seed round would be required to accelerate to the upside 50-plant scenario and reduce cash-floor risk

• OEM feature creep. Robotic welding vendors may add basic fit-up checks or routing rules into their own software stacks. Mitigation: Start as the cross-cell system of record for readiness decisions and quality outcomes, then integrate with multiple OEMs so customers keep value even in mixed fleets.
• Data capture friction. If operators must enter too much information manually at fit-up stations, usage will drop and recommendations will be ignored. Mitigation: Launch with minimal required measurements and lightweight photo capture, then automate more inputs through vision or metrology integrations after proving ROI.
• Too little repetition at small shops. Fabricators with highly bespoke one-off jobs may not have enough repeated geometry to justify dedicated preflight software. Mitigation: Target shops with repeat structural programs for data centres, racks, or modular components where the same joint families recur at meaningful volume.