A deal team will spend six weeks and six figures pressure-testing a target's quality of earnings. It will rebuild the working capital bridge, re-cut every revenue cohort, and walk lawyers through each material contract line by line. Then it will assess the management team and the culture with three reference calls and a read in the partner meeting.
That asymmetry is the quiet risk in most control acquisitions. The workstream that most often determines whether the value-creation plan is executable is the one that gets the least rigor. And a new wave of AI-driven HR and org tools is now promising to close that gap. The promise is real, but the way most teams are reaching for it leaves them holding something they cannot defend in an investment committee.
The cost of getting people wrong
The data here is not subtle. McKinsey puts the M&A failure rate at 70 to 90 percent, citing the Christensen team's 2011 work in Harvard Business Review. The interesting part is the cause. When McKinsey ran its 2023 Global Survey on M&A capabilities, the single most common reason integrations failed to deliver expected value was not financing structure or synergy modeling. It was lack of cultural fit and friction between the acquirer and the target. Bain's M&A Practitioners' 2023 Outlook Survey found the same thing from the buyer's seat: nearly half of respondents named cultural fit or difficulty integrating management teams as a primary reason their past deals had failed.
This shows up in the numbers that matter to a fund. Bain analyzed 65 mature buyouts with full access to fund and management projections and found that 71 percent of them missed their projected margins, landing roughly 330 basis points below the deal model on average. People are a large part of why. Research out of MIT Sloan, built on Census data covering about 4,000 acquisitions, found that 33 percent of acquired employees leave within the first year, versus 12 percent of comparable regular hires. Gallup, citing EY, puts departures of key employees even higher: 47 percent within a year, 75 percent within three. The talent you underwrote walks out the door, and your value-creation plan walks out with it.
None of this is a secret. Bain's own work shows that culture is an early focus area in 80 percent of integrations, yet 75 percent of acquirers still run into cultural problems serious enough to require intervention. Firms know people risk is real. They simply do not diligence it with anything close to the discipline they bring to the QoE.
Why the AI wave doesn't close the gap on its own
It is easy to see why AI org tools look like the fix. They are genuinely good at what they do. They can ingest an HRIS export, an org chart, engagement survey results, and compensation data, then surface patterns across a workforce far faster and more consistently than a human team working from the data room. They flag concentration risk, spans and layers that do not make sense, comp that is out of band, attrition clusters. For a workstream that has historically run on reference calls and instinct, that is a real upgrade in speed and coverage.
The problem is what comes out the other end. The output is fast, confident, readable, and unaccountable. It produces a score and a clean narrative, and it produces them the same way whether the underlying signal is sound or misleading. A deal team cannot take a black-box org score into an IC and defend it under questioning. "The model said so" is not a thesis. And the moment a partner asks why the bench looks thin or why engagement dipped in one unit, the tool that generated the number is not in the room to answer.
The trust problem has a name: automation bias
This is the part most teams underrate, and it is the heart of the issue. When people are handed a confident, well-written explanation attached to an automated output, they tend to accept it, even when it is wrong. Researchers call this automation bias, and it is distinct from its opposite, algorithm aversion, where people discard a model in favor of their own judgment.
The most useful study here is Horowitz and Kahn's "Bending the Automation Bias Curve," a preregistered experiment run across a representative sample of 9,000 adults in nine countries and published in International Studies Quarterly. Their finding cuts in a direction that should give deal teams pause: as the stakes of a decision rise, people become more cautious about trusting an algorithm, but that caution is uneven and shaped by how much they actually understand about the system in front of them. People with a little AI familiarity are the most prone to over-trusting it. In other words, the failure mode is not abstract. It is exactly what happens when a smart, busy professional accepts a confident output on a high-stakes call because pushing back would take work.
That is the worst possible dynamic to introduce into a people-driven investment thesis. And the broader market already feels it. Pew Research found that half of Americans are now more concerned than excited about AI in daily life, up from 37 percent in 2021, and 61 percent want more control over how it is used. Regulators feel it too. Under Article 14 of the EU AI Act, high-risk AI systems must be designed for meaningful human oversight, and the text specifically calls out automation bias as a risk that human overseers have to guard against. Those obligations begin to bite in August 2026. Human in the loop is no longer a nicety. It is becoming a governance expectation.
What a human in the loop actually adds
The human layer is not there to slow the machine down. It is there to do the one thing the model cannot: look at a data point and say "this does not mean what it appears to mean."
Consider the kinds of calls a model gets wrong because it has no context:
- A high-turnover number that looks like rot but is actually a deliberate, healthy performance cull the founder ran last year.
- A thin leadership bench that reads as fragility but reflects an owner intentionally staying lean ahead of a sale, with capable people one level down.
- An engagement score depressed by a single one-time event, a botched office move or a delayed bonus cycle, rather than anything structural.
A model will score all three as red. A credentialed CHRO who has sat in the operating seat will recognize which ones are real and which ones are noise, and will tell you which of the real ones is fixable inside the hold period. That is the difference between a data dump and a judgment you can underwrite. Ambiguity, edge cases, and context are precisely where models are weakest and where experienced human judgment earns its keep.
This is also where the broader research points. EY's work with Oxford's Saïd Business School found that the root causes of transformation success and failure are human and emotional, and that leaders who put people at the center of a transformation were up to 12 times more likely to significantly improve performance. You do not capture that with a score. You capture it with someone who knows what to look for.
How PreOrg is built
Every PreOrg assessment is reviewed by a senior CHRO before it reaches the deal team. Not a chatbot. Not an algorithm running unsupervised. A practitioner who has held the seat, reading the machine's output and applying the context that turns a data point into a defensible call.
That pairing is the whole point. AI brings speed, consistency, and coverage across a workforce dataset no human team could process by hand in a diligence window. The CHRO brings interpretation and accountability. Machine speed, human judgment. It is the only configuration that gives a PE buyer something it can actually take into an IC and stand behind.
The bottom line
The firms winning on operational alpha are the ones that stopped treating people diligence as post-close HR cleanup and started treating it with the same rigor as the quality of earnings. The tools to do that well now exist, but a black-box output is not the answer. It just moves the gut read from the partner meeting into a model and dresses it up as rigor.
Organizational due diligence done right looks like the rest of your diligence: fast, thorough, and signed by someone who will defend it. That is what we built PreOrg to deliver.