What should a buyer extract in the first week?
Won-job margin versus quoted margin. Past-due backlog versus future backlog. Scrap / rework / first-pass yield if the system holds it. Machine utilization or OEE if anyone collected it. Customer and part-number concentration. Inventory quantities versus the last physical.
Those extracts feed on-time delivery, scrap, and rework, backlog quality, and inventory. They do not, by themselves, move a shop from the general CNC 2.0x–3.6x SDE cell into an aerospace 6.5x–9.0x $3–10M EBITDA cell. They decide whether the earnings in that cell are believable.
IBBA Pulse Q1 2026: 83% of deals over $5 million drew three or more offers (Source: IBBA). Offers get sharper when two buyers share the same extract. Foggy systems produce price chips, not mystery premiums.
QoE becomes common around ~$2 million+ of EBITDA and earlier if the add-back schedule is aggressive. A CIM recast is not a substitute. SBA 7(a) still caps at $5 million (Source: SBA); lenders will ask how job cost ties to tax returns.
What if the shop runs QuickBooks and a whiteboard?
Tell the truth in the teaser. Diligence cost goes up. The working-capital peg needs a harder physical. Job-cost WIP needs a job-by-job review. None of that is a moral failure. It is why a $900,000-SDE fab file can still sell to an owner-operator who will live in the shop, and why a searcher may pass.
Do not spend 30 days building a dashboard with no history. Buyers will ask for the year-ago export. Empty tables with pretty charts are a finding.
BLS wage data does not value an ERP. It values the planner who actually uses it. If only the owner can print a job traveler, you have owner dependence inside the software.
How do job-cost misses show up in value?
Quoted 38% gross and actual 19% is not a rounding error. It is a pricing or a process problem. Repeat it across the book and the trailing SDE or EBITDA is a look-back, not a run-rate. ISM August 2026 Prices 71.1 (Source: ISM) makes uncaptured cost even more expensive. Fed G.17 utilization 76.0% (July 2026; Source: Fed G.17) makes “we were busy” an incomplete sentence if the jobs lost money.
Percentage-of-completion WIP that cannot ship is inventory, not revenue. Customer-owned material booked as yours is a trust problem.
| Extract | Healthy signal | Diligence problem |
|---|---|---|
| Job cost vs quote | Tight cloud around quote | Systematic miss; hero jobs only |
| Past-due backlog | Small and aging down | Growing; hides OTD failure |
| Scrap / rework | Measured, trending | Not collected; or collected and ignored |
| Concentration | Matches invoice rebuild | ERP customers ≠ billed customers |
| Inventory vs physical | Narrow variance | Chronic variance; peg fight |
| Whiteboard + QuickBooks | Disclosed | Hidden; then discovered |
What sellers should export before the CIM
A 24-month job-margin file, a customer sales file, an aged inventory, and a past-due open-order file. How it works is the process. Sell my manufacturing business is the seller door. If the system cannot produce those files, say so and budget time for a physical and a QoE.
Buyers start at buy a manufacturing business. The valuation pillar still picks the cell. ERP quality is how you decide whether to believe the cell.
BizBuySell’s 2.75x SDE median (2021–2025; Source: Iconic recap) sits on Main Street files that often run light systems. That is expected. Pretending a whiteboard shop has Epicor-grade history is how Main Street deals get ugly in week six.
How to tell a living system from a filing cabinet
A living system has job travelers that match invoices, inventory that has been cycled, and a past-due file someone opens on Monday. A filing cabinet has years of closed jobs and a whiteboard for the live ones. Both can be honest. Only one can give you a 24-month margin cloud in an afternoon.
Ask for a screen-share, not a PDF. Ask the seller to filter last year’s jobs by margin. Ask them to show a job that lost money and what they did. If they cannot, you are in cabinet mode. Budget the physical and the QoE.
Cloud systems still fail this test when nobody used the modules you need. A NetSuite plant that never turned on work orders is a QuickBooks plant with a nicer login.
What “job cost versus quote” should look like as a chart
A tight cloud around the quoted margin, with a known tail of strategic losses. A two-humped cloud (hero jobs and disasters) is a pricing or a process problem. A cloud that only exists for last month because someone started costing for the sale is a finding. Ask for the year-before extract.
Add-backs that assume every job hits quote will die when this chart is printed. Inventory WIP that assumes every open job ships will die the same day.
Buyer-type differences in how hard they pull
SBA individuals may live with a cabinet if they will stand on the floor. Searchers and add-ons will not live with a cabinet on a $2 million-EBITDA file. PE associates will ask for a data-room extract that an API can rebuild. If the shop cannot produce it, say so in the teaser so the right buyer set shows up.
How it works is when extracts get staged. Do not stage a screenshot of the dashboard and call it a room. Stage the CSVs.
Open-order files should separate firm dates from blankets. Backlog is the valuation read of the same extract. This page is whether the extract exists.
How to run a screen-share that tells you if history exists
Ask the seller to share the live system, not a dashboard PDF. Filter last year by margin. Open a job that lost money and ask what changed on the next quote. Open a job that is past-due and ask why. Open inventory by last movement. If any of those four clicks fail, you are in cabinet mode even if the logo on the login is expensive.
Then ask for the year-before extract of the same four views. A system that only became tidy for the sale will fail the year-before test. That failure is allowed if you disclose it. It is fatal if you hide it.
Cloud versus on-prem does not decide this. Unused modules decide this. A plant that never turned on work orders is a whiteboard plant with a nicer invoice. Say so in the teaser so the buyer set is honest. SBA owner-operators can live with cabinets. Searchers on a 2 million EBITDA file usually cannot.
Stage CSVs in the room, not screenshots. Open orders should already be split into firm dates and blankets before a valuation person reads them. The extract is this page. The read is the backlog page. Both should exist.
Jason Taken is a business broker, not an ERP consultant. Educational information only. Not accounting advice. No value is guaranteed.
A four-click live test that decides cabinet versus system in fifteen minutes
Share the live system. Filter last year by margin. Open a job that lost money and ask what changed on the next quote. Open a past-due job and ask why. Open inventory by last movement. If any click fails, you are in cabinet mode even if the login logo is expensive. Ask for the year-before extract of the same four views. A sale-tidy system fails that test.
Cabinets are allowed when disclosed. They raise physical and QoE cost. They fit some SBA owner-operators. They usually do not fit searchers on a 2 million EBITDA file. Stage CSVs, not screenshots. Split open orders into firm dates and blankets before a valuation person reads them. Unused modules make a cloud system a whiteboard. Say so in the teaser so the buyer set is honest.
A closer read on screen share
Share the live system. Filter last year by margin. Open a job that lost money and ask what changed on the next quote. Open a past-due job and ask why. Open inventory by last movement. Ask for the year-before extract of the same four views. A sale-tidy dashboard fails that test. Cabinets are allowed when disclosed. They raise physical and QoE cost. They can fit an owner-operator. They usually do not fit a searcher on a 2 million EBITDA file. Unused modules make an expensive login a whiteboard. Stage CSVs, not screenshots. Split open orders into firm dates and blankets before anyone reads backlog. The extract is this page. The read is the backlog page. Both should exist.
What a year-before extract proves that a sale-tidy dashboard does not
Share the live system. Filter last year by margin. Open a job that lost money and ask what changed on the next quote. Open a past-due job and ask why. Open inventory by last movement. Then ask for the same four views from the year before the teaser. A dashboard built for the sale fails that test. History that exists will still be there.
Cabinets are allowed when disclosed. They raise physical and quality-of-earnings cost. They can fit an owner-operator. They usually do not fit a searcher on a two-million EBITDA file. Unused modules make an expensive login a whiteboard with a license fee. Stage comma-separated extracts, not screenshots. Split open orders into firm dates and blankets before anyone reads backlog.
The extract is this page. The read of that extract as backlog, scrap, or inventory is the sibling pages. Both should exist. A shop that can produce the extract and cannot explain a losing job has a management problem, not a software problem. A shop that cannot produce the extract has both.