Metal Fabrication Software: Quote in 15 Minutes, Not an Hour

Why quoting and job tracking software makes or breaks a fab shop

Walk into most 30 to 120 person fab shops and the real ERP is a workbook called Quote_Master_v14_FINAL_USE_THIS.xlsx sitting on a mapped drive. It has a tab per material grade, a lookup table for laser burn rates that someone built in 2019 from a Trumpf cheat sheet, a bend column that multiplies hits by a flat number, and a weld column where the estimator types a gut number and moves on. There is a hidden column called "fudge" that adds 12 percent. Nobody remembers who put it there.

That workbook is doing the work of a quoting engine, a router, a cost model and a price book. It fails in specific, expensive ways. An estimator opens a customer's STEP file in SolidWorks or Autodesk Inventor, eyeballs the flat pattern, guesses at pierce count, types numbers into the sheet, exports a PDF and emails it. That is 40 to 90 minutes per quote on anything with more than a few parts. At 60 quotes a month and a 25 percent hit rate, the shop is spending roughly 60 hours a month producing quotes, 45 of which lose. Meanwhile the customer who wanted a number in 24 hours went to the shop down the road who answered in four.

Here is the scene that actually costs money. A repeat customer sends a revised print, rev C, with a bend radius change and a new tab. The estimator copies last quote's tab, updates a couple of cells, sends the number. Rev C now needs a second setup on the brake because the flange runs into the tooling. Nobody catches it until the operator is standing at the machine with a nest of 200 parts and a form that will not clear. That is a 3 hour setup surprise, a scrapped nest of 14 gauge 304, and an argument with the customer about who pays. The spreadsheet had no memory of why the last quote was priced the way it was, so it could not tell anyone the new one was wrong.

Problem 1: quotes take an hour and lose to whoever answered first

The pain is speed and consistency. A 12 part weldment assembly with mixed material takes your senior estimator most of a morning. When he is on vacation, the number comes from someone who does not know that this customer's parts always need deburr and that the 10 gauge job runs slow because the fiber laser is throttled on that material. Two estimators, two prices, sometimes wildly apart on the same part.

Off-the-shelf does not fix this cleanly. Paperless Parts and Quoted quote fast, but they price to their own model, not yours: they do not know your brake tooling library, your welder skill matrix, or that job 88214 for that customer ran 40 percent over because of tacking time. Your ERP, whether that is Global Shop Solutions, E2 Shop System or JobBOSS, has a quoting module that is essentially a form for the number you already calculated somewhere else. None of them learn from your closed jobs.

A custom build starts from your part geometry and your actual history. Ingest the STEP or DXF, run feature extraction to pull sheet thickness, cut path length, pierce count, bend line count and bend direction changes, hole count and internal cutout area. Multiply cut length by a machine-specific feed rate table you own, keyed on machine, material grade and thickness, seeded from your Trumpf or Amada job logs, not from a vendor's guess. Bend time comes from hits multiplied by your measured cycle time, plus setup time predicted from tooling changes required versus what is already in the brake. Weld time comes from a joint-length model plus a fixture and tack allowance by weldment complexity. Then, and this is the part nobody buys off the shelf, every quote writes back its estimate, and every completed job writes back actuals from the shop floor. The variance report tells your estimator on quote entry: "Similar parts for this customer have run 18 percent over on weld. Suggested adjustment applied." On the fab builds we have delivered, quote time on repeat geometry drops from most of a morning to under 15 minutes, and the number is defensible.

Where AI genuinely helps here: an RFQ inbox agent. Customers email PDFs, prints, spreadsheets of line items, sometimes a photo of a hand-marked drawing. A document extraction model reads the incoming email, pulls part number, revision, quantity, material callout, finish, due date and any notes, matches the part against your part history by geometry hash and part number, and drafts a quote before your estimator opens Outlook. It gets most of the boring part right. Your estimator reviews and prices the judgment calls. That works because the extraction task is narrow and you have thousands of your own prints to validate against.

Problem 2: the traveler and the actual job diverge by hour two

Someone prints a router. It says laser, deburr, brake, weld, paint. It goes in a folder that walks with the job. Halfway through, the brake operator finds the part needs a hardware insert before weld, so he walks it to the press. The traveler now lies. When the job ships three days late, nobody can reconstruct where the time went, so the next quote for that part repeats the same mistake.

Generic tools cannot fix this because their routing model is a linear list of operations with a standard time, and a real fab job is a graph: parts nest together across jobs, subassemblies converge at weld, some parts go out to plating for four days and come back. E2 and JobBOSS can hold operations, but the nest crosses job boundaries and their data model does not, so your laser time gets allocated by a rule of thumb rather than by actual burn share.

The custom build models the nest as a first-class object. A nest belongs to a machine, a sheet, a material grade and thickness, and contains parts from N jobs. Cut time and sheet cost allocate to jobs by actual part area plus a share of skeleton scrap, not by guess. Routing is a directed graph with real dependencies, including outside processes with promised return dates that block downstream operations. Shop floor terminals, cheap Android tablets at each cell, let the operator scan a job barcode, tap start and stop, log quantity good and quantity scrapped with a reason code, and log a deviation when they add an unplanned operation. That deviation is the signal. It flows into the variance report and into next quarter's quoting model.

Problem 3: nobody knows what the shop can actually promise

The owner promises three weeks because the customer asked for three weeks. The scheduler maintains a whiteboard, or a second spreadsheet, that shows roughly what is on the laser. The brake is the real constraint half the time, and welding is the constraint the other half, depending on mix. Nobody knows which until the job is late.

Off-the-shelf finite scheduling in the mid-market ERPs assumes infinite tooling and a fixed operator pool. It will happily schedule three jobs on the brake at once because it does not know you only own one set of 88 degree acute punches, or that only two of your six welders are certified for the aluminum work. That produces a schedule the floor ignores within a week, and then you are back to the whiteboard having paid for the module.

What a custom build does: model constraints that are actually yours. Tooling as a finite resource with a setup matrix, so jobs sharing tooling cluster and the schedule surfaces the setup savings. Welders as skill-tagged resources with certification expiry, so an aluminum job cannot schedule onto an uncertified operator. Material availability as a gate, so nothing schedules before the coil or sheet lands. Then run a nightly forecast against the actual backlog and give the owner a promise date with a confidence band, plus the one number that matters: which resource is the constraint this week. When sales asks "can we take this," they get an answer in 30 seconds instead of a meeting.

Problem 4: material cost moves and your quotes do not

Steel moved. Your workbook has a per pound number someone updated last quarter. You quoted a 6 week lead time job at last quarter's price and you will buy the material in five weeks. On a job with $40k of 304 stainless, an 11 percent move is $4,400 of margin gone, and you find out at invoicing.

No off-the-shelf quoting tool solves this because it needs your supplier's actual pricing, your remnant inventory, and your policy on how far to escalate on long lead times. It is your commercial judgment, encoded.

Custom: a material master with per-grade, per-thickness, per-supplier pricing, updated by importing your mill or service center price sheets, with an effective date. Quotes carry a material-price-as-of stamp and an escalation rule you set, for example anything with a promise date beyond 30 days quotes at a forward price with a stated validity window on the quote PDF. Remnant inventory tracked by sheet, so a job that can run out of a 48 by 60 drop prices against the drop and tells the nester to use it. Shops we have built this for stop eating price moves silently, because the quote itself carries an expiry and the number reflects what they will actually pay.

Problem 5: the estimator is a single point of failure

The person who has been quoting for 22 years is the pricing engine. Everything above about tooling, materials, weld difficulty, this customer's habits, lives in his head. When he retires or takes a week off, quote throughput drops and accuracy goes with it. No software vendor sells you this person.

The build turns his judgment into rules that a junior can operate. Every override he makes gets captured with a reason: "added 2 hours, this customer's fit-up tolerance always needs rework." After 300 quotes, those overrides are a rules library. New estimators see the suggestion and the reason together. The system records why, so the shop keeps the knowledge when he leaves.

What this costs and how long it takes

Across 2,000-plus projects at Digital Heroes, the pattern in this category holds steady. A focused first release, meaning geometry-driven quoting with your rate tables, quote versioning, quote PDF generation, and shop floor time capture on tablets, lands at $60k to $130k and ships in 12 to 16 weeks. That is the release that pays for itself, because quote speed and quote accuracy are where the money is.

A full platform, adding nest-aware costing, finite scheduling against tooling and welder certifications, material master with supplier price imports, outside process tracking, and two-way sync to your existing ERP for purchasing and invoicing, runs $150k to $400k phased across 6 to 12 months. Phase it. Nobody should sign a 10 month contract before seeing quoting work.

What drives price up specifically in fab: CAD file ingestion is the big one. Reading STEP and IGES reliably and extracting flat patterns, bend lines and pierce counts is real engineering, and if you need native SolidWorks or Inventor file support rather than neutral formats, add cost. Multi-site adds cost when sites have different machines and different rate tables and jobs move between them. Nesting is expensive to build and usually should not be: integrate with SigmaNEST or Radan rather than rewriting them. ITAR or AS9100 requirements add real cost through access controls, traceability, material cert tracking and audit logs, and if you do defense work, budget for it honestly rather than bolting it on later. Machine integration, pulling actual cycle data off a Trumpf or Amada controller instead of relying on operator taps, adds cost but pays back in data quality.

Build versus buy: take the honest position

Buy if you run a job shop under roughly 20 quotes a month on repeat parts with stable pricing and a single location. Paperless Parts at its published subscription pricing, or an E2 Shop System seat, will beat a custom build on total cost and you should not be reading this. Buy if your differentiation is machine capability rather than responsiveness, and buy if you have no one internally who can own a system.

Build when these signals show up together. Your estimator is a bottleneck and quotes go out slower than 24 hours. You have run the same part for the same customer more than five times and still cannot tell whether it makes money. You lost a job you should have won because a competitor quoted in three hours. You have more than one location and they price differently for no defensible reason. You have already bought a quoting tool and your estimators still keep the real numbers in a spreadsheet alongside it, which is the loudest signal of all: the tool does not model your shop, so your people route around it.

The position: in this category, most shops over $8M in revenue with mixed cut, bend and weld work should build the quoting and costing layer and buy everything else. Do not build accounting. Do not build nesting. Do not build CAD. Build the thing that encodes how your shop makes money, because that is the part no vendor can sell you and the part your competitors cannot copy.

How to choose a developer for fab shop quoting and job tracking software

Ask them to describe the data model for a nest before you talk price. If they model a job as a flat list of operations with a standard time, they have built work order software, not fab software. The right answer involves nests spanning jobs, cost allocation by part area plus skeleton share, and routing as a graph with outside process gates. This question alone eliminates most firms.

Ask what they have done with STEP or DXF ingestion. Specifically: which library or kernel, how they handle flat pattern extraction, what happens when a customer sends a solid model with no flat, and what their accuracy has been on pierce count. Vague answers here mean you will fund their learning curve.

Ask how they integrate rather than replace. You have QuickBooks or an ERP, you probably have SigmaNEST or Radan, you may have a controller that logs cycles. A good partner maps those integrations in the first two weeks and tells you which ones are hard. A bad one proposes to rebuild all of it.

If you do any defense, aerospace or medical work, ask directly about ITAR, AS9100 and material cert traceability, and ask for a project where they implemented it. Compliance is not a feature you add in month eight. It shapes hosting, access control and the audit trail from day one, and retrofitting it costs more than building it in.