What does job costing have to prove before you trust gross margin?
Buying a manufacturing business without that extract is how people capitalize a quoting habit. Sub-sectors change the typical band. They do not replace the target's trailing twelve months after a physical inventory.
What to pull from the ERP — or admit you cannot
Systems commonly seen on the floor: Epicor, JobBOSS / E2, Global Shop, Plex, NetSuite, ProShop, IQMS / Delmiaworks, ShopTech. The brand is less important than whether you can extract:
- Job cost versus quote (won-job margin versus actual)
- On-time delivery and past-due backlog
- Scrap, rework, and first-pass yield
- Machine utilization / OEE if anyone collected it
- Customer and part-number concentration
- Inventory accuracy versus physical
If the shop runs on QuickBooks plus a whiteboard, write that sentence in your notes. It changes quality of earnings cost and working-capital trust. A QoE that rebuilds margins from packing lists is a line item, not a surprise.
Contribution margin by part family tells you whether the 5-axis cell is carrying the manuals. Scrap and first-pass yield tell you whether the quality certificate is a wall hanging. On-time delivery versus promise date tells you whether the backlog is a blessing or a past-due pile.
| Sub-sector (rule-of-thumb, not the target) | Gross margin band | EBITDA / SDE band | What usually breaks the quote |
|---|---|---|---|
| CNC / precision machining | 28–45% | 10–18% | Owner-only estimating; crashed hours |
| Metal fab / stamping | 22–38% | 8–15% | Steel pass-through; project mix |
| Plastics (IM / extrusion / thermoform) | 25–40% | 10–18% | Resin; tooling ownership |
| Electronics / EMS / harness | 15–28% | 6–12% | Component prices; customer-owned designs |
| Aerospace & defense suppliers | 20–35% | 10–18% | Special process yield; overtime |
| Medical device / components | 30–50% | 12–22% | Quality system cost that is real |
| Automotive / heavy equipment | 15–28% | 6–12% | OEM leverage; release mix |
| Food / beverage / co-pack | 18–32% | 8–14% | Giveaway, slots, working capital |
Label every cell as broker rule-of-thumb. Public exact RMA line items are paywalled. Your target's truth is TTM after E&O.
Pass-through is a diligence fact
ISM's August 2026 Manufacturing PMI printed 54.6 with a Prices Index of 71.1 (Source: ISM). Resin, steel, and electronic components that cannot be passed through show up as a margin gap the index already warned you about. Ask for surcharge history, index clauses, and jobs that were quoted before the last move and shipped after it.
Fed G.17 utilization was 76.0% in July 2026 (Source: Fed G.17). A shop quoting 95% utilization on one-shift overtime is telling you the margin depends on premium labor. BLS OEWS May 2025: machinist median $58,750; CNC programmers $68,120 (Source: BLS OEWS). If overtime is "normalized out" of SDE, put it back until you have a second-shift plan.
How job-cost lies — and how a QoE catches them
WIP at hope. Jobs sit at quoted margin instead of cost incurred. The working-capital peg then buys fiction. See inventory and WIP.
Orphan hours. Setup and programming live in a bucket that never hits the job. Gross margin looks clean. The cell is not profitable.
Hero customers. Concentration plus a low-margin hero is a double problem. Revenue concentration on the worst work is how shops stay busy and poor.
Scrap as "startup." Recurring first-article waste treated as one-time. If it happens every month, it is the process.
A buyer-paid QoE should test job-cost versus tax, not invent a new multiple theory. If the report kills $400,000 of EBITDA, the price moves or the deal dies. Main Street files in BizBuySell's 2021–2025 set cleared at 2.75x median SDE on about $268k of discretionary earnings (Source: BizBuySell via Iconic). Moving $80,000 of scrap back into the recast on a file that size is a turn.
Machines and margin are the same conversation
Machine-list analysis tells you whether the hours in the job-cost file could have been cut on the iron that exists. A shop quoting 5-axis work on a tired 3-axis cell will show it in scrap, overtime, or both. Deferred capex is a margin story: earnings assumed free spindles.
SBA 7(a) maximum remains $5 million with a 10% injection under SOP 50 10 8 (Source: SBA). Lenders who bother to read job-cost will not treat a whiteboard as a borrowing base. ABL field exams are blunter.
Stanford's 2024–25 median purchase of $16 million at about 6.2x EBITDA sat on a ~25% EBITDA margin (Source: Stanford GSB). That is a search-fund median. A 9–12% job shop is not that company even if revenue rhymes. Do not "normalize" your way into 25%.
Quoting rules you should capture before they vanish
Ask the estimator to walk three won jobs and three lost jobs. You want the rule: setup hours, machine-hour rate, material factor, scrap factor, and whether the rate includes programming. If the rule lives only in the owner's head, your first 90 days will leak margin. Write the rule into a one-page playbook during diligence if the seller allows it. If the seller refuses, assume the rule leaves.
Lost-job files tell you whether the shop is priced out or sloppy. A pattern of winning only the worst work is a commercial finding. A pattern of losing on price while delivering 18% against 35% quotes is an estimating finding. Both belong in the QoE request list.
Outside processes — heat treat, plating, grinding, painting — often hide in job-cost as a lump. Ask for the vendor file and the markup. If the shop is a broker of processes it does not control, yield risk sits at the vendor. Certificates on those vendors are access, not decoration.
Capacity math versus the P&L
Machine-hour rates that assume 90% utilization on one shift will look profitable until someone is sick. Fed G.17 utilization 76.0% (Source: Fed G.17) is national climate. Your shop's available hours are a roster and a maintenance calendar. If quoted hours exceed available hours, the margin depends on overtime you may have "normalized out."
SBA 7(a) DSCR inside a $5 million cap (Source: SBA) will not care that the CIM said 35% gross. It will care whether cash appears after scrap and premium labor. Combined 7(a)+504 can reach $10 million for fitting borrowers (Source: SBA). It still wants real jobs.
A fourth practical heading: standard cost versus actual in plastics and EMS. Standard-cost systems can bury resin or component moves until a physical. Demand a revaluation policy and the last adjustment. ISM Prices 71.1 (Source: ISM) makes stale standards a diligence fact. Do not capitalize a standard that is a year old.
Main Street 2.75x SDE files (Source: BizBuySell via Iconic) hide estimating problems because the owner is the estimate. Search-fund median files at $16 million / 6.2x / ~25% margin (Source: Stanford GSB) often have a second chair. That median is never a job-shop multiple. If you are buying the owner-estimator shop, job-cost is the whole underwriting.
IBBA's 83% three-offer statistic (Source: IBBA) does not excuse a missing extract. Ask for it early. If it cannot be produced, budget the reconstruction or walk.
Three jobs you should pull on every file
One hero customer job, one average job, one dog. For each: quote, hours, material, outside process, scrap, invoice, cash collected. If the hero is the dog, concentration and margin are the same problem. If the average job cannot be reconstructed, the ERP is a whiteboard. SBA 7(a) DSCR inside $5 million (Source: SBA) wants those three jobs to look like the P&L. Combined 7(a)+504 can reach $10 million (Source: SBA). Same want.
Main Street 2.75x / ~$268k SDE (Source: BizBuySell via Iconic) lives or dies on those three jobs. Stanford $16 million / 6.2x / ~25% margin (Source: Stanford GSB) often has better extracts — and that median is never a job-shop multiple. IBBA's 83% three-offer statistic (Source: IBBA) does not reconstruct hours. ISM Prices 71.1 (Source: ISM) will show up on the material line. Fed utilization 76.0% (Source: Fed G.17) will show up on the overtime line. BLS OEWS programmer $68,120 (Source: BLS OEWS) will show up if the owner still programs.
Jason Taken is a business broker, not a cost accountant. Educational information only. Confirm recast mechanics with a CPA. No outcome is guaranteed.
Questions the estimate log should answer before you capitalize SDE
What machine-hour rate was used, and whether it includes programming. What scrap factor was used, and whether actual scrap matches it. Whether outside process is marked up or passed through. Whether the owner still writes every number. Whether three reconstructed jobs — hero, average, dog — foot to the P&L gross line. If the log is a legal pad, budget reconstruction.
Do not call that pad a pricing opportunity. A shop that quotes 35% and delivers 18% will deliver 18% for you unless the estimator changes and the rate file changes. Capture the rule before it vanishes. Lost-job files tell you whether you are priced out or sloppy. Both findings belong in the QoE request list.
If the ERP cannot export won-job versus actual, say so in the model as a diligence cost. Whiteboard shops are common. They are not moral failures. They are expensive to underwrite if you pretend they are Epicor.
HedgeStone will ask for three reconstructed jobs before treating a recast as bankable. If the shop cannot produce them, say so in the model as cost. A legal-pad estimate log is common. Capitalizing it as if it were a cost system is not.