A private equity commitment I reviewed a few years ago had one of the cleanest DD files I'd seen. The consultant's report was positive. The placement agent memo was aligned. A co-investor had done its own assessment and reached the same conclusion. The GP's own materials were thorough, internally consistent, well-presented. Every source agreed and the IC approved with confidence.
The thesis collapsed within two years. The post-mortem revealed that every document in the file drew its central claims from the same GP data room. The consultant had analysed the GP's track record numbers. The placement agent had repackaged the GP's investment thesis. The co-investor's 'independent' assessment was built on the same underlying data set. Almost all four reports traced to a single informational origin, and nobody had mapped that before committing capital.
In private markets we have all been warned about confirmation bias. Most of us think we've dealt with it by now. The thesis is challenged in IC, hard questions are asked, management teams get meaningful pushback. But the version we've actually been trained to spot is the minor form. Psychological and, with some self-reflection, easily recognisable. There is another type, a major form, which is structural, and in my view the bigger problem by far: the information architecture itself produces convergence. A disciplined allocator genuinely trying to challenge the thesis will still find it confirmed, because every source in front of them traces to the same origin. You can't think your way out of a biased information architecture.
Analytical independence is not informational independence
It is all too often that the consultant's report, the placement agent memo, the co-investor's assessment, and the fund manager's own materials all converge towards the same conclusion. Each participant adds genuine professional judgment, with their own methods and institutional incentives. The analysis is independent. But the raw inputs trace to a single source.
The distinction between analytical independence and informational independence is where the DD process quietly breaks down. Convergence across four analytically independent reports feels like reliability. It would be, if the reports were also informationally independent. When they share a common data origin, their agreement tells you almost nothing you didn't already know from the first report. Bovens and Hartmann established this formally over two decades ago: convergence among sources sharing a common cause carries near-zero additional confirmatory weight. Whether the dependent sources are scientific instruments or investment consultants makes no difference to the underlying probability.
The natural objection: financial data is audited, and the accounts have to come from somewhere. True. Audit is one of the few genuine points of informational independence in the chain. An auditor has access to primary records, bank statements, counterparty confirmations. But consider how narrow a slice of a typical DD pack the audited accounts actually cover. The manager's track record methodology, its pipeline assessment, its market sizing, its team stability narrative, its ESG representations: none of that is audited. These are the claims that tend to drive the investment decision, and they flow through the DD chain unchecked. The audit covers the financials. The investment thesis rests on everything around the financials.
The second objection is just as instinctive: we already benchmark. But where do those benchmarks actually come from? In March 2025, BlackRock completed its £2.55 billion acquisition of Preqin, the private markets industry's leading independent data provider. Preqin's benchmarks are being integrated into BlackRock's Aladdin and eFront platforms. And even before that acquisition, a Kenan Institute study found that both Preqin and PitchBook source their fund performance data primarily from FOIA requests to public pensions and voluntary GP submissions. You're benchmarking GP data against aggregated GP data. That catches outliers. But it won't catch systematic optimism, shared methodology errors, or industry-wide classification drift. It tells you whether a GP's numbers are unusual relative to peers, not whether those peers' numbers are themselves reliable.
Three questions worth asking before you commit capital
These are three questions I apply (on paper) to every DD pack that reaches my desk.
Where does each claim actually come from? Take the five most consequential claims in the file, the ones that would change the investment decision if they turned out to be wrong. Trace each back to its origin. How many ultimately derive from the GP's own data, representations, or models? If the answer is all five, you've got confirmation, not corroboration. Separate the claims with genuine external verification (audited financials, regulatory filings, market data the GP doesn't control) from GP-sourced content repackaged by intermediaries. In my experience, most DD packs run at roughly 80-90% GP-derived content once the formatting is stripped away. That's not necessarily a problem if the IC is aware of it. It becomes a problem when nobody has done the count.
What would catch a material misrepresentation? Suppose the GP's key representations turned out to be materially wrong. Not fraudulent, just wrong. An overstated multiple, an optimistic market size estimate. Which of your DD sources would have caught it independently? This isn't hypothetical: the SEC's Division of Examinations has repeatedly flagged misleading track record presentations, cherry-picked performance figures, and misrepresented return calculations in GP disclosures. When the consultant, the co-investor, and the placement agent all started from the same data, none of them would have caught the error. If the answer to 'who would have caught it?' is 'nobody', your process is testing internal consistency rather than external validity.
Is the information getting richer or thinner? Over the life of a manager relationship, watch for the pattern: quarterly reports arriving later, granular portfolio company data replaced by aggregated summaries, independent fund administrators replaced by in-house functions, reporting format changes that break period-on-period comparability. Conventional DD looks at the content of the information rather than its quality as a signal. But when the information environment degrades, it usually does so before the problems it eventually conceals become visible.
What follows from this
There are good practices available for challenging an investment thesis before commitment: red teams, pre-mortems, independent operating DD across separate workstreams, back-channel references, portfolio company site visits. Done well, these strengthen analytical independence, meaning the quality and rigour of the interpretation. That matters, and it should be part of any serious DD process. But it's generally not sufficient, because it doesn't address the underlying informational independence. Unfortunately you can't discipline your way out of it. A red team working from the same GP data room is stress-testing the interpretation rather than the data. A pre-mortem generates downside scenarios, but those scenarios are still bounded by GP-sourced inputs. Back-channel references and site visits are the exceptions: they introduce genuinely new information from outside the GP's chain. But their scope is narrow. They cover people and operations, not the financial representations that drive the investment model.
The missing discipline sits one layer beneath. If you accept that the informational independence of your evidence base varies across investments (and it does), two things follow.
First, the scenarios you run should reflect it. Standard sensitivity analysis treats all inputs as equally trustworthy and varies them within a fixed band. But an audited revenue figure and a fund manager's self-reported market sizing estimate are not equally reliable, and they shouldn't carry the same confidence interval. Where the underlying data is independently verified, a narrow range is appropriate. Where it's GP-sourced and unverified, the honest range is much wider. The choice of scenario should be driven in part by how much informational independence sits behind each input, not just by what seems like a 'reasonable' range for the variable itself.
Second, the risk premium should reflect the quality of the evidence, not just the risk of the asset. Two investments can have identical expected return profiles but very different evidence bases. An IC that treats them identically is implicitly assuming equal confidence in both. When the evidence behind one rests almost entirely on GP representations, that assumption is wrong. How much additional premium is warranted for thin informational independence is a hard question. But not asking it means the pricing is silently absorbing a risk the IC hasn't discussed.
How deep this analysis needs to go depends on the investment. For a large, concentrated commitment to a first-time fund, a full provenance mapping of the DD inputs (scoring each against its informational source, calibrating scenarios and pricing accordingly) will tighten the decision materially. For a smaller re-up with an established manager and a long reporting history, a lighter-touch assessment may be enough: a conscious acknowledgment of where the evidence base is thin, documented for the IC, without rebuilding the model. The point isn't that every commitment requires the same depth of informational analysis. It's that the depth should be a deliberate choice, not an omission.
That choice starts before the data room opens.
Vincent Piscaer is Founder and CEO of STRAD, an independent asset analysis and strategic advisory firm. He holds the PCF-30 designation as CIO of a CEE-focused private equity fund and was part of the team that developed the CFA Institute's Certificate in ESG Investing. He writes at strad.global.
