AI Won't Do Your QoE Analysis (And Why)

We built an AI Quality of Earnings tool. And we tell every buyer the same thing on day one: this software won't do your QoE for you. It will make you dramatically faster and more thorough — but the judgment is still yours, and that's the point.

The Promise and the Misconception

AI is genuinely changing financial due diligence. It's also generating a lot of marketing copy that overpromises in ways sophisticated buyers, lenders, and CPAs see right through.

Here's the misconception worth correcting: a Quality of Earnings analysis is not a spreadsheet exercise. It's a judgment exercise. The deliverable looks like a workbook, but the value is the opinion underneath it — whether reported earnings are real, repeatable, and useful for making a deal decision.

Push-button QoE doesn't exist. What does exist — and what's actually transformative — is software that lets one analyst do the work of three, with better coverage and a tighter audit trail. That's what we build. That's what this piece is about.

What AI Can Do Well

AI is genuinely good at the painful, time-consuming parts of QoE — the parts that traditionally consume 70–80% of an engagement and produce no opinions, just structure:

Ingest messy financial exports

Pull from accounting integration partners, raw GL exports, bank statements, and trial balances — and reconcile them into a single canonical dataset

Review the full population

Scan 100% of GL transactions instead of sampling 5%. Catch the things sampling misses by definition

Surface anomalies and exceptions

Round-dollar entries, period-end clustering, related-party indicators, duplicate vendors, unusual journal entries — drafted as a queue for human review

Draft adjustment support

Identify candidate addbacks, pull the matching GL lines, link to source documents, and write the first draft of the narrative

Build structured schedules

EBITDA bridge, working capital, proof of cash, debt schedule, customer concentration — populated and tied out to source automatically

Accelerate the workflow

Cut the time from data receipt to first findings from weeks to hours, leaving more runway for the parts that actually require judgment

What AI Cannot Do For You

This is where most AI-QoE marketing falls apart. The list of things AI cannot do is short, but it includes the entire reason QoE exists in the first place:

Decide whether an addback is appropriate

AI can flag a $400K consultant payment as a candidate. Whether it's truly non-recurring, or really the founder's spouse, or a pattern that will continue under new ownership — that's a judgment call

Assess whether management's explanation is credible

AI can transcribe a management call. It cannot tell you whether the CFO is hedging on the customer concentration question because she's nervous, or because she knows something you don't

Determine whether risk is real or contained

Customer churn that looks like 12% in the data might be three contracts that all renewed last week, or three customers who just got acquired and are reviewing every vendor relationship. AI sees the number. The human knows the story

Conduct the diligence conversation with management

QoE depends on the back-and-forth — the follow-up question to the answer to the original question. That's a human-to-human exercise, and it's where most of the real findings come from

Take responsibility for the conclusion

When the buyer's lender calls in six months because cash flow doesn't match what was diligenced, somebody has to answer the phone. The AI cannot. There is nobody to sue when the AI is wrong

QoE is not about generating a report. QoE is about supporting a deal decision — and the value of that support is directly tied to the credibility of the person who stands behind it.

AI-Assisted vs AI-Generated: The Distinction That Matters

Two phrases that sound similar describe completely different products. The difference shows up in every part of the engagement:

Question	AI-Generated QoE	AI-Assisted QoE
Who owns the conclusion?	The software (effectively, no one)	The analyst, the CPA, or the buyer
Who reviews each adjustment?	Nobody — the model output is the answer	A human accepts, modifies, or rejects every adjustment
Who handles management Q&A?	Skipped	The analyst, with AI-prepared question lists
Who signs the report?	Unsigned, or 'AI-generated' disclaimer	The professional whose name is on it
Defensible to a lender or court?	No	Yes — same standard as a traditional engagement
What you're paying for	Output	Coverage, structure, and speed in service of judgment

The shorthand: AI-generated treats the software as the analyst. AI-assisted treats the software as a force multiplier for the analyst. The first one is an overpromise. The second one is what actually works.

The Better Model: Analyst-in-the-Loop

The most useful framing we've found — for ourselves, our customers, and the CPAs we work with — is analyst-in-the-loop. The analyst is still the one driving the engagement. The AI changes what they get to start from.

Instead of starting from a stack of bank statements and a blank workbook, the analyst starts from:

Full-population review, already done

Every GL line classified, every bank deposit reconciled, every anomaly queued — before the analyst opens the workbook

Adjustment candidates with traced support

Each candidate addback already has its matching GL lines pulled and its source documents linked. The analyst's job is to judge, not to dig

A tied-out EBITDA bridge

Drafted from the data, with every line drillable to the underlying transactions. The analyst stress-tests it; they don't build it from scratch

A starting list of diligence questions

Anomalies become questions for management. The analyst refines them and runs the call. The AI doesn't make the call

The result isn't a faster bad QoE. It's a better QoE done faster, because human attention gets concentrated on the parts that actually require it. Full-population GL review instead of sampling. Tied-out adjustment bridges instead of pasted-in numbers. Documented audit trails on every adjustment instead of "see workpaper 3.7, ask Dave."

Who This Matters Most For

Big PE shops with in-house diligence teams have always had structured workflows — they pay for them in headcount. The buyers who benefit most from AI-assisted QoE are the ones who couldn't previously access that level of rigor at all:

Independent searchers and ETA

You're reviewing 50 deals to close one. You can't pay $30K for a full QoE on every LOI. AI-assisted analysis gets you institutional-quality screening at a per-deal cost that actually scales

Independent sponsors

You need defensible diligence that an LP base will accept, without the overhead of a full transaction-services bench. AI-assisted analysis closes the gap

SBA buyers

Lender requirements keep getting tighter. You need the structure of a real QoE on a sub-$5M deal where the math on a traditional engagement doesn't work

Lower-middle-market deal teams

The $1–10M deal segment has always been underserved by traditional QoE pricing. AI-assisted analysis is what makes proper diligence economically viable here

Lenders underwriting acquisitions

You need the same coverage on a $2M loan as on a $20M loan, but the borrower can't fund a $30K QoE on a $2M deal. AI-assisted analysis closes that gap

CPA firms scaling QoE practice

The bottleneck on a QoE practice is usually the manager's hours. AI-assisted workflow lets the same manager oversee 2–3x more engagements without dropping quality

The Real Unlock (And Our Actual Positioning)

Here's the thing we've come around to internally, and it's a sharper way to say what Shepi actually does:

Shepi is not "AI does QoE." Shepi is "AI gives every buyer a structured diligence workflow that used to require a full transaction advisory team."

That's a more honest claim, and it's a more useful one. It explains who benefits and why. It doesn't try to convince you the software replaces professional judgment — it tells you the software is what makes professional-quality judgment possible on deals where the traditional cost structure made it impossible.

For the search fund evaluating its 40th teaser. For the SBA buyer trying to underwrite a $3M acquisition without spending 10% of equity on diligence. For the CPA firm trying to take on three more QoE engagements without hiring three more managers. That is the real unlock — not replacing professional judgment, but making real diligence accessible to buyers who previously couldn't afford it.

If you want the longer, more balanced survey of where AI fits in QoE, read Can AI Replace a Quality of Earnings Report?. If you want to see what analyst-in-the-loop actually looks like in practice, read AI for Financial Due Diligence.