Custom Manufacturing ERP: What Job Shops Build When E2 and JobBOSS Break

Why ERP (Enterprise Resource Planning) makes or breaks a job shop operator

Walk any custom machine shop at 6:45 on a Tuesday and you will find the same artifact: a printed dispatch list, marked up in highlighter, that disagrees with the system that printed it. E2 says job 20447 is at op 40, final grind. The part is physically back at op 20 because four pieces failed CMM inspection, and the router has no way to represent a rework loop, so the shop floor lead moved the parts by hand and told nobody who works in an office. The scheduler rebuilds the real schedule in Excel every morning from that walk. The Excel file is the actual ERP. The system you pay maintenance on every year is a typewriter for travelers.

This is the normal condition of shops running E2, JobBOSS, or spreadsheets past roughly $10 million in revenue or a second location. Those tools were designed around a linear router: op 10, op 20, op 30, ship. Real custom work loops back, splits into lots, merges, and leaves the building for heat treat. Every gap between the model and the floor gets papered over by a person: the estimator quoting from memory and a 2019 spreadsheet, the controller reconciling job costs three weeks after the parts shipped, the expediter whose entire job is knowing where things actually are.

Those people cost real money, but the bigger leak is decisions made on wrong numbers. When job costing lands 21 days late and burden is averaged across the whole plant, you keep quoting your worst work at your best margin and never learn which is which. Below are the five failures that push shops with real budgets toward building, what a build honestly costs, and when staying on the shelf is the smarter call.

Routers that cannot loop, split, or leave the building

The scenario every job shop knows: a $48,000 titanium housing order, 22 pieces, and op 30 inspection fails four of them. The four need a weld repair and re-machine loop while the other 18 continue. One job is now two lots on two paths. E2 and JobBOSS model a router as a numbered sequence, so the standard workaround is to create a second job, manually journal cost across, and accept that quantity integrity and lot genealogy just died. Do that thirty times a month and your WIP number is fiction.

A custom build models routing as a directed graph, not a list. A failed inspection spawns a rework path with its own operations and costs that roll back into the parent job automatically. Lot splits carry genealogy, so the 18 pieces on the main path and the 4 in weld repair remain one job, one cost roll-up, one promise date calculation. Outside operations like heat treat, plating, and anodize are routing steps with a linked purchase order, expected turn days, and live vendor status, instead of a disconnected PO that a buyer chases by phone every Thursday.

Job costing that arrives three weeks late and lies when it gets there

Your controller closes jobs at month end. Labor comes from time entries that operators batch in from memory at the end of a shift, because clocking onto an op in JobBOSS takes ten clicks at a shared terminal across the aisle. Burden is one plant-wide rate. So a job quoted at 28 points of margin closes at 4, three weeks after it shipped, and nobody can say which operation ate the difference. The off-the-shelf systems technically support operation-level costing. In practice the data collection is so hostile that shops fall back to job-level guesses, and a guess averaged with a plant-wide burden rate is not costing, it is accounting theater.

A custom build treats floor data collection as the product. A barcode scan or a tablet mounted at the machine, under ten seconds to clock onto an operation, no login gymnastics. Cost accrues per operation in real time, and the system compares actual hours to quoted hours while the job is still running. When op 50 crosses its estimate, the shop manager gets pinged that day, while intervention is still possible. Burden rates live at the work center level, so the 5-axis Mazak stops subsidizing the manual deburr bench in your margin reports.

Quoting from memory while margins erode quietly

Your senior estimator turns quotes around in a day using Excel, tribal knowledge, and thirty years of pattern recognition. He retires in eighteen months. Nobody knows the shop's win rate by part family, and the quoting spreadsheet has no idea that job 19882, a nearly identical bracket in the same material, ran 30 percent over on labor last quarter. Meanwhile a new buyer at your biggest account is shopping every RFQ against two other vendors.

The quoting modules bolted onto E2 and JobBOSS are calculators. They multiply rates you type in, and they never reach back into actuals. A custom quoting engine is a feedback loop: it surfaces the three most similar historical parts by material, feature set, and tolerance band, shows quoted versus actual hours for each, and tracks win/loss by customer and part family. Every closed job makes the next quote smarter. Your retiring estimator's judgment becomes data the 28-year-old inherits, instead of walking out the door with him.

Scheduling that assumes infinite capacity and a quiet phone

An aerospace customer calls at 10 a.m. and needs their order moved up to Friday. Saying yes is easy. Knowing which of the other 40 active jobs slip, and by how many days, is impossible in E2, so the answer gets negotiated at a whiteboard and priority goes to whichever customer screams loudest. Infinite-capacity scheduling is not scheduling. It is a wish list with dates on it.

A custom build does finite-capacity scheduling against the constraints that actually bind in your shop, which are never generic: an operator skills matrix when only two people can run the wire EDM, fixture availability, and real outside-process lead times by vendor. The what-if view matters most: drag the hot job forward and see exactly which three jobs slip and whether any of them breach a promise date, before you commit. Then quote the expedite fee from data instead of gut. Dispatch lists live on machine tablets and update in real time, instead of being printed at 6 a.m. and wrong by 9.

Multi-plant transfers and traceability audits that take a week

Plant 2 machines castings that Plant 1 received, and the material certs live in a filing cabinet 40 miles away. Then the AS9100 auditor asks for full genealogy on a shipped lot: heat number, every operation, every operator, every outside-process cert. Assembling that answer takes four days of digging through E2 attachments, email, and paper. If you hold ITAR work, the prints sitting on a network share open to the whole company are a separate and scarier conversation.

Off-the-shelf systems treat multiple locations as separate sets of books and traceability as file attachments. A custom build makes lot genealogy the data spine: heat lot to raw stock to every split, merge, rework loop, and shipment, with certs attached to the exact routing step that produced them. An audit package becomes a one-click export. ITAR access control lives at the part and document level, enforced by the system rather than by hoping. Inter-plant transfers are first-class moves with in-transit status, not a phantom PO and sales order pair that inflates both plants' numbers.

What building actually costs, from delivery experience

Across 2,000+ delivered projects, Digital Heroes sees a consistent pattern in this category. A focused first release runs $60,000 to $130,000 and ships in 12 to 16 weeks. For job shops that almost always means routing, shop floor data collection, and live job costing, because that is where the pain concentrates and where payback is fastest. A full platform covering quoting, finite scheduling, purchasing, multi-plant inventory, and compliance runs $150,000 to $400,000, phased over 6 to 12 months, with each phase live and earning before the next starts.

What pushes price toward the top of those bands in this category specifically: the number of work centers and any machine-monitoring integration, the depth of finite scheduling logic, compliance regimes like AS9100 documentation and ITAR data segregation, migrating years of E2 or JobBOSS job history, and launching multiple plants at once instead of sequencing them. After launch, budget 15 to 20 percent of the build cost per year for hosting, support, and steady small improvements.

Build vs buy: the honest line

Buy when you are under roughly $5 million in revenue, single plant, mostly repeat work with genuinely linear routers, and your complaint is reporting rather than modeling. At that profile JobBOSS or E2 is cheaper and faster than anything custom, and newer shop systems like ProShop and Fulcrum are worth a trial before you spend a dollar on development.

The signal to build is never a missing feature. It is the data model. When your workarounds are spreadsheets that describe reality better than the ERP does, when rework loops and split lots force you to create fake jobs and hand-move costs, when a second location turns every transfer into double entry, no amount of configuration fixes that, because the wrong assumptions are baked into the schema. Our position: if you have a six-figure budget, multiple plants or high-mix work, and two of the five problems above, build a focused system for those two problems and keep your accounting in QuickBooks or Sage. Do not pay anyone to rebuild a general ledger. The money belongs in the routing graph, the floor data, and the costing engine, because that is where your shop is actually different.

How to choose a developer for custom manufacturing ERP

Most software firms have never stood on a shop floor, and it shows in their schemas. Four tests before you sign anything:

• Make them whiteboard the data model. Ask how they would represent a lot split at op 30 where four pieces enter a rework loop. The right answer involves a routing graph and lot genealogy. If you hear "we can add a status field," the interview is over.
• Interrogate their floor data collection. Ask to see a clock-on flow from a past project and time it. Under ten seconds on a tablet or scanner is the bar. Ask what share of floor transactions their last system captured without office correction.
• Check the integrations they have actually shipped. Accounting sync to QuickBooks or Sage, machine monitoring feeds, CAM or nesting output, and EDI for aerospace and automotive customers. Demos count, slide decks do not.
• Test compliance literacy. Have them walk you through an AS9100 audit export and an ITAR access model from prior work. In this category compliance is a data architecture decision made in week one, not a module bolted on in month nine.