Financial Due Diligence: Automated Quality of Earnings (QoE) Analysis

A Quality of Earnings report is the document that decides whether a deal closes at the LOI multiple, gets repriced 0.5-1.5x EBITDA lower, or breaks. For a typical middle-market buyout at 10x EBITDA on $25M of reported EBITDA, a $2M addback dispute moves enterprise value by $20M. Sponsors have historically accepted 4-6 week QoE timelines and $150K-$500K Big 4 or accounting boutique fees as the cost of doing business. That tolerance is eroding. Bain's 2025 PE survey shows 71% of GPs now run at least one technology-enabled diligence workstream on every deal above $50M, up from 19% in 2022, with QoE the most common entry point because the data is structured and the work is repetitive.

Automated QoE platforms ingest the target's general ledger, sub-ledgers, payroll, AR/AP aging, and customer-level revenue files directly from QuickBooks Online, NetSuite, Sage Intacct, Microsoft Dynamics, or SAP. They run revenue recognition tests against ASC 606, normalize EBITDA across 40-60 addback categories, build customer cohort curves, and generate a draft QoE deliverable in 7-10 days. The accounting firm still signs the report and defends the numbers in negotiation, but 60-75% of the manual tie-out, schedule building, and Excel reconciliation work disappears.

What automated QoE actually does differently

Traditional QoE engagements rely on the seller's management workbook — a 30-tab Excel file with EBITDA bridges, run-rate adjustments, and pro forma synergies. The accounting team validates this workbook through judgmental sampling, typically testing 25-40 transactions per revenue stream and 15-30 journal entries above a materiality threshold. Sample sizes are dictated by hour budgets, not statistical coverage. A $200M-revenue target with 18,000 monthly invoices gets a sample that covers roughly 0.2% of transactions.

Automated platforms invert this. They process 100% of the general ledger — typically 3-5 years of monthly closes, 500K-5M journal entries for a middle-market target — and flag anomalies for human review. Hebbia, Rogo AI, and Eilla AI offer LLM-based deal workstreams that read the data room PDFs (customer contracts, audit workpapers, board minutes) alongside the structured GL. Mosaic, Cube, and Vena ingest the financial system data and run the variance, cohort, and normalization logic. Big 4 firms have built proprietary stacks: Deloitte's Omnia/iDeal, EY's Helix and Diligence Edge, KPMG Clara, and PwC's Aura/Halo all now embed GenAI agents for memo drafting and transaction tagging.

Revenue quality: where most repricing happens

Revenue quality issues drive roughly 55-65% of QoE-related purchase price adjustments based on engagement data shared by the top six US accounting boutiques (RSM, BDO, Grant Thornton, Baker Tilly, Cherry Bekaert, CohnReznick). The recurring questions: how much of revenue is contractually recurring versus repeat-but-discretionary; what does customer cohort retention actually look like at 12, 24, 36 months; how concentrated is the book; and is revenue being pulled forward through ASC 606 misapplication.

Automated cohort analysis is the single highest-leverage capability. Given a customer-level invoice file with first-invoice date and revenue by month, the platform constructs a triangle showing gross dollar retention (GDR) and net dollar revenue retention (NDR) by acquisition cohort. For a SaaS target advertising 115% NDR, the cohort triangle frequently reveals that 2022 cohorts are at 95% NDR while 2020 cohorts are at 125%, indicating that NDR is flattered by an aging book and that new customer economics have deteriorated. This kind of finding routinely supports a 0.5-1.0x EBITDA multiple reduction.

On ASC 606 testing, automated platforms apply rule libraries that check for common misstatements: revenue recognized at contract signing rather than over the service period, set-up fees recognized upfront instead of amortized, channel partner gross-versus-net classification, and bundled arrangements where standalone selling price allocation is missing. Eilla, Rogo, and Hebbia parse the actual customer contracts (typically 200-2,000 documents in a data room) and reconcile recognition timing to the GL. In a recent vertical SaaS deal I reviewed, this approach surfaced $3.8M of revenue recognized 11 months early on multi-year prepaid contracts — a finding that the management workbook had buried and that judgmental sampling would have missed.

EBITDA normalization at machine speed

Reported EBITDA on the management deck is rarely the EBITDA the deal closes on. A typical middle-market target presents 35-60 addbacks totaling 15-30% of unadjusted EBITDA: owner compensation normalization, one-time legal and transaction costs, COVID-era PPP forgiveness, discontinued product lines, rent normalization for related-party leases, stock-based comp, and pro forma synergies for acquisitions completed mid-period. Each addback needs to be (a) quantified to the GL, (b) characterized as recurring or non-recurring, and (c) supported by evidence.

Automated platforms maintain a rule library of 40-60 standard addback categories and tag GL entries automatically. A journal entry to account 6450 (Legal — M&A) above $25K gets flagged as a likely transaction cost addback; entries to 5810 (Owner Salary) get reconciled to W-2 data to compute the gap to market comp; related-party rent entries get compared to comparable lease data from CoStar or REIS. The accounting team reviews each tag and approves, rejects, or adjusts. What used to take a senior associate three weeks of Excel work now takes a manager 3-4 days of review.

Pro forma synergy addbacks remain the most contested category. Sellers want to add back the run-rate impact of cost actions taken in the trailing twelve months — closed facilities, terminated headcount, renegotiated supplier contracts. The validation challenge is proving the action was actually taken and the savings are sustainable. Automated platforms reconcile headcount addbacks to payroll registers (Gusto, ADP, Paychex, Rippling), facility closures to lease termination documents and utility bill discontinuation, and supplier savings to AP run-rate comparisons. The QoE provider will typically accept 60-80% of validated pro forma adjustments versus 30-50% historically — not because they're less rigorous, but because the evidence is stronger.

Working capital, net debt, and the closing mechanics

The QoE doesn't just inform price — it sets the closing mechanics. Net working capital target (the peg), debt-like items, and the cash-free/debt-free adjustments are all anchored in the QoE. Misestimating the working capital peg by $1M on a deal flows directly to purchase price at closing. The peg is conventionally set at a 12-month average of normalized working capital, but seasonality and growth distort the average.

Automated platforms compute daily DSO, DPO, and DIO from the sub-ledger rather than month-end balances, exposing window-dressing patterns where AR collections are accelerated in the last week of each month or AP payments are delayed at quarter-end. In a recent industrial services deal, daily DSO analysis showed AR was 14 days higher mid-month than at month-end, indicating the seller was managing month-end balances. The buyer adjusted the working capital peg up by $4.2M, which translated directly to a $4.2M price reduction at close.

Debt-like items are where contract-reading LLMs add the most value. The platform parses customer contracts for unfunded liabilities (gift card balances, customer prepayments, loyalty obligations), employment agreements for change-in-control payments and unvested deferred comp, and equipment leases for ASC 842 obligations that may not be on the balance sheet. A clean debt-like items schedule built from the actual contracts is materially more defensible than a list compiled from the seller's CFO.

The vendor landscape

The accounting firm selection no longer correlates cleanly to deal size. Mid-market sponsors are using boutiques like Stout, Riveron, Aprio, and Embark on $100M-$500M EV deals because these firms have moved faster on automation than some Big 4 practices. Apollo, KKR, and Bain Capital have all built internal data science teams that pre-process target financials before the QoE provider even engages, compressing the provider's timeline. This in-house preparation is closely related to the work covered in commercial due diligence automation — the same data infrastructure that supports unit economics analysis feeds the QoE workstream.

Implementation: how PE firms actually deploy this

Sponsors typically deploy automated QoE in three patterns. First, in-house data preparation: the deal team pre-loads target financials into a sponsor-controlled environment (often Snowflake or Databricks with a thin overlay from Hebbia or Rogo), runs initial analytics, and hands a cleaned dataset to the external QoE provider. Second, provider-led automation: the sponsor selects an accounting firm explicitly for its automation stack and lets the provider drive the toolchain. Third, hybrid: sponsor pre-screens with an LLM-based platform during exclusivity, and the provider runs definitive QoE on the closing dataset.

Where automation fails — and what to do about it

Automated QoE breaks down on three categories of targets. First, businesses with weak financial systems — Excel-based accounting, mid-period system migrations, or companies that have completed undocumented acquisitions where the books haven't been integrated. The data quality is too poor for automated rules to add value, and the work reverts to manual reconstruction. Second, complex revenue arrangements with significant judgment — long-duration construction contracts, complex SaaS bundles with usage-based pricing, or businesses with material customer-specific accounting. The ASC 606 rule library can flag issues but can't replace professional judgment on percentage-of-completion or standalone selling price allocations. Third, businesses with material related-party transactions that aren't documented in the GL — the platform can flag related-party rent and intercompany flows it can see, but undisclosed related-party arrangements remain a manual investigation.

Sponsors should also build a clear handoff to post-close integration. The QoE dataset — normalized financials, customer cohorts, addback substantiation — is the same dataset the new CFO needs in the first 100 days. The post-acquisition 100-day plan should explicitly inherit the QoE data model rather than rebuilding it. Firms that do this collapse the time-to-baseline-reporting from 90-120 days to 30-45 days, which has direct value creation impact in years one and two.

The QoE dataset is the most expensive financial dataset the portfolio company will ever produce. Throwing it away at close and rebuilding it through the new CFO's onboarding is the single most common operating model mistake I see.

— Former PE CFO, now portfolio operations advisor

Risk, independence, and what the auditors think

QoE is not an audit — it's an advisory engagement performed under AICPA SSCS standards in the US, not AICPA AU-C audit standards. This matters because the same Big 4 firm cannot perform QoE on a target it audits without independence considerations, and the QoE deliverable does not provide assurance in the audit sense. Automation doesn't change this framework, but it does raise questions about who is responsible when the LLM gets something wrong. The current professional consensus, reinforced by the PCAOB's June 2024 guidance on AI use in audit-adjacent work, is that the signing partner bears full responsibility regardless of which tools were used. This pushes firms toward platforms with strong audit trails and explainability, and away from black-box LLM output.

For sponsors, the practical implication is to treat QoE provider selection as a technology decision as much as an accounting decision. Ask for: the specific platforms used, the percentage of work performed by humans versus AI agents, the audit trail architecture, the source documentation citation standard, and the partner's personal experience with the toolchain. Sponsors signing the SPA based on a QoE deliverable need confidence that every number ties to a source the provider can defend in litigation if needed.

Automated QoE is not replacing the accounting firm or the partner-level judgment that goes into price negotiation. It is collapsing the time and cost of producing the underlying analysis, expanding the coverage from samples to full populations, and exposing revenue and earnings quality issues that traditional workflows miss. For sponsors competing in auctions with 10-15 day exclusivity windows, that compression is no longer optional.