Pre-LOI add-back due diligence comes down to one piece of arithmetic. A buyout is priced as a multiple of adjusted EBITDA, so every dollar of add-back is multiplied before it reaches the price. At 10.0x, a $1.8M add-back that does not hold is $18.0M of enterprise value you would be paying for and not getting.
The add-backs arrive in the CIM, weeks before anyone tests them against the ledger. The quality of earnings engagement that does that testing is commissioned after exclusivity, by which point the price is anchored. So the question a deal team actually faces before the LOI is narrower: do the CIM’s add-backs survive the seller’s own databook?
We could not show you that on a live deal. CIMs are confidential, and so are the databooks behind them.
So we built one.
Apex Industrial Technologies is a fictional manufacturer of precision-machined components for industrial automation and fluid-power OEMs: two facilities in Ohio and Indiana, 214 employees, about $71.9M of revenue. We wrote its CIM, in the voice of a fictional sell-side adviser, and the five-sheet Excel databook that came with it. Then we ran both through the askOdin Clarity PE & M&A desk.
Apex Industrial Technologies, its founder, Northgate Systems, Halberd Lane Partners and every figure below belong to a synthetic data room. Any resemblance to a real company is unintended.
The audit at a glance
- Documents
- Apex_CIM.pdf · Apex_Databook.xlsx (5 sheets)
- Analyst inputs
- Guidance EV $140.0M · replacement CEO $1.2M
- Adjusted EBITDA, as the CIM states it
- $14.0M
- Disqualified add-backs
- $4.4M · three of four
- Supported EBITDA
- $9.6M
- Supported EV at the CIM’s 10.0x
- $96.0M · $44.0M (31.4%) below guidance
- PE score
- 56 / 100 · WATCH
- Findings
- 1 HIGH · 2 MEDIUM · reported EBITDA matched
| As the CIM states it | Supported by the databook | Difference | |
|---|---|---|---|
| Reported LTM EBITDA | $9.2M | $9.2M | matched |
| Add-backs | $4.8M (four items) | $0.4M (one item) | −$4.4M |
| Adjusted EBITDA | $14.0M | $9.6M | −$4.4M (−31.4%) |
| Multiple | 10.0x | 10.0x, the CIM’s own | none |
| Enterprise value | $140.0M guidance | $96.0M | −$44.0M |
The multiple is not ours. It is the one the guidance implies: $140.0M divided by the $14.0M the CIM calls adjusted EBITDA. Holding it fixed is what makes the comparison fair. Supported EV is not a valuation opinion; it is the seller’s own price, recomputed on the earnings the seller’s own databook supports.
How does the engine test an add-back?
A language model reads the CIM: each add-back, in the CIM’s own words, with its amount and page. It also maps databook labels to sheet and cell coordinates, and it does this without seeing a number. From there everything is deterministic Go code: it reads every value from the cell the mapping named, decides which rule an add-back falls under from the CIM’s own wording, and applies thresholds that are fixed in the engine and were set before this data room existed.
Five rules run on a PE room:
| Rule | What it tests | Severity |
|---|---|---|
| PE-001 | A “one-time” add-back the databook shows in two or more fiscal years at 50% or more of the amount claimed | HIGH |
| PE-002 | Owner pay added back above what a replacement would be paid | MEDIUM |
| PE-003 | A pro forma or run-rate saving inside trailing EBITDA | MEDIUM |
| PE-004 | The CIM’s reported EBITDA against the databook’s LTM figure | matched within 5% · HIGH · FATAL above 10% |
| PE-005 | The CIM’s adjusted EBITDA against the seller’s own QoE, when one is supplied | matched within 5% · HIGH · FATAL above 10% |
An add-back no rule covers, or one a rule cannot test, stays in supported EBITDA and is counted as untested. Absent evidence never disqualifies anything.
The four add-backs
All four sit on page 7 of the CIM, in the adjusted EBITDA bridge.
PE-001 · HIGH: the “one-time” IT transition recurs. −$1.8M
The CIM: “Non-recurring ERP transition costs of $1.8M, incurred in the LTM period in connection with the migration of the Company’s ERP to the managed platform operated by Northgate Systems, are added back.”
The databook (Opex detail!B4:E4, Northgate Systems, ERP managed service):
| FY23 | FY24 | FY25 | LTM Sep-26 |
|---|---|---|---|
| $1.59M | $1.70M | $1.81M | $1.80M |
Three fiscal years, each well above half the amount added back. A cost the company pays every year is an operating expense, whatever the CIM calls it. At 10.0x, this one add-back carries $18.0M of the guidance.
Question for management: Northgate Systems is in the databook in three fiscal years. Which part of the $1.8M stops after close, and what ends it?
PE-002 · MEDIUM: the owner’s pay has a replacement cost. −$1.2M
The CIM: “The founder-CEO exits at close; the full $1.2M of CEO salary and bonus is added back.”
The databook (Payroll!B2:E2, Payroll!B3:E3): salary $0.80M and bonus $0.40M. The analyst’s input: a replacement CEO at $1.2M.
The job stays when the founder leaves. Only pay above what a replacement costs is a real saving, and here the founder was paid $1.2M against a $1.2M replacement. None of the add-back survives. PE-002 does not run at all without a replacement salary; the engine will not guess one.
Question for management: What will the replacement CEO be paid, and which of the founder’s duties does the role keep?
PE-003 · MEDIUM: the synergy has not happened yet. −$1.4M
The CIM: “Pro forma procurement synergies of $1.4M from the consolidated supplier programme launched in the third quarter of 2026 are included on a run-rate basis, as if the programme had been in place for the full LTM period.”
The evidence here is the CIM’s own wording. “Run-rate” and “as if” describe a saving still to come. It may be real, and it may belong in your forward case. It does not belong in trailing EBITDA, which is what the multiple is applied to.
Question for management: Which signed agreement delivers the $1.4M, and where does it show in the trailing cost lines?
Accepted: the litigation settlement. +$0.4M
The CIM: “A one-off settlement of $0.4M paid to a former distributor in November 2025 is added back.”
The databook (Other expense!B2:E2, legal settlements): $0.00M in FY23 and FY24, $0.40M in FY25 and the LTM. One fiscal year, so nothing shows it recurring. PE-001 tested it and let it stand.
And the reported figure held
PE-004 checked the CIM’s reported EBITDA of $9.2M (page 6) against the databook’s LTM Sep-26 EBITDA (Summary!B6:E6): $9.20M. Matched. Had the two figures been more than 10% apart, the run would have ended at FATAL and no bridge would have been drawn from a reported number the databook contradicts. PE-005 was skipped because there was no seller’s QoE in the room.
How was the score of 56 reached?
Separately from the bridge, the desk scores the business on seven facts, each read from the databook or quoted from the CIM, against bands written down before any room was run. Apex earned 16 of 21 points, a base of 76.
| Factor | Fact | Points |
|---|---|---|
| Revenue growth, FY23 to FY25 | 7.0% a year ($61.24M to $70.11M) | 2 |
| Gross margin stability | 0.8 percentage-point range | 3 |
| EBITDA margin, LTM | 12.8% ($9.20M of $71.88M) | 2 |
| EBITDA trend | LTM is 116% of FY23 | 3 |
| Cash conversion, LTM | 67.6% after $2.98M of capex | 2 |
| Customer concentration | Largest customer 24% of FY25 revenue (CIM p. 4) | 1 |
| Repeat revenue | 83% from parts in production three years or more (CIM p. 3) | 3 |
The three findings then cost 20 points: 10 for the HIGH, 5 for each MEDIUM. 76 − 20 = 56, WATCH.
That is a fair reading of the company. Apex looks like a sound, slow-growing manufacturer whose adviser stretched the bridge. WATCH is not a pass on the business. It says the price is anchored on earnings the documents do not support.
What does the debt look like on supported EBITDA?
Debt is sized on EBITDA too, and usually on the adjusted figure. The desk includes a leverage calculator for that reason. It is arithmetic on two numbers the analyst types, a debt multiple and a covenant ceiling. Clarity does not know any lender’s terms.
Take senior debt at 4.0x and a ceiling of 4.5x:
| On the CIM’s $14.0M | On the supported $9.6M | |
|---|---|---|
| Senior debt at 4.0x the CIM figure | $56.0M | $56.0M |
| Leverage | 4.0x | 5.8x |
| Debt that fits a 4.5x ceiling | $63.0M | $43.2M |
Sized on the CIM, the debt sits comfortably inside the ceiling. Measured against what the databook supports, the same $56.0M is 5.8x, and only $43.2M fits. The $12.8M difference has to come from somewhere: more equity, less debt or a lower price. Better to see that before the LOI than in the first compliance certificate.
This is a scenario on assumed terms, not a finding. Nothing in the Apex documents is a loan agreement, and nothing here says a lender would or would not fund.
Why does a general-purpose AI summary miss it?
Ask a chat assistant to summarize the Apex CIM and you will get a fair account of what page 7 says: four add-backs, $4.8M, adjusted EBITDA of $14.0M. That is the problem. The CIM’s account of its add-backs is internally consistent. The contradiction is between page 7 and a row on the Opex detail sheet, and between the word “run-rate” and the meaning of trailing EBITDA. A summary reports the page. The audit holds the page against the databook. (We wrote about why language models smooth over add-backs in The CIM Add-Back Trap.)
Adjusted EBITDA has no standard definition. When a public company reports it, the SEC’s Regulation G requires a reconciliation to the nearest GAAP measure. A private-company CIM carries no such obligation, which is why the bridge on page 7 deserves the hardest reading in the book.
What this case study is, and what it is not
It is a single run on a synthetic data room, recorded from the engine on 17 November 2026. The files are fingerprinted with SHA-256 when they are read:
- Apex_CIM.pdf
- 50cbfc29bc8edfc90fc1c4961f735e1f13e829b19e101532c906377e3e386027
- Apex_Databook.xlsx
- e6a8dfef2bd3fe1a532d50a5d1a732922f8d351019c01ecce70ecf6dd2618f3c
It is not a benchmark, and it says nothing about any real company.
It is not a quality of earnings report. askOdin does not test the numbers against the general ledger, sample invoices, prove cash or set the working-capital peg; your QoE provider still does that after exclusivity. What changes is the order: the add-backs are challenged before the price is set, so the QoE confirms a bridge you have already tested.
And it is not a valuation. Supported EV uses the seller’s own multiple. Whether 10.0x is the right multiple for Apex is your call.
The desk runs on files your team uploads, with no data-room connector. Documents are handled under the posture on the security page: No Training Use, and raw documents held no longer than 30 days under a documented retention ceiling. To run it on a CIM you are looking at now, request a scoped pilot.