A deal team at a large-cap European PE fund receives several 200-page Vendor Due Diligence reports from sell-side advisors. A full sell-side DD package (commercial, financial, legal, tax, ESG) typically costs the seller EUR 500K-2M to produce, depending on deal size and complexity. The deal team reads every page anyway.
The Problem Is Trust.
The conventional explanation is that deal teams are overwhelmed by documents. Virtual data rooms contain thousands of files. Exclusivity periods have compressed. Teams are stretched across multiple parallel deals.
The deeper problem is structural: VDD reports are sell-side documents written to facilitate a sale, not to inform a buyer. The real work is verification: checking whether the claims in the VDD are supported by the underlying data, whether the commercial DD assumptions are consistent with the financial DD projections, and whether material risks are buried in footnotes rather than flagged in executive summaries.
sbb-itb-6ca8558
Why Generic AI Tools Made This Worse
By 2024, 86% of M&A organizations had adopted some form of generative AI in their workflows (Deloitte, 2025 M&A Trends). The result was disappointing.
Generic AI tools (ChatGPT, Copilot, enterprise search products) can summarize documents. But they cannot reliably verify claims against source data nor cross-reference findings across workstreams. They cannot detect sell-side framing: the subtle practice of presenting accurate data in a context that minimizes risk.
The result is what deal teams call the verification trap: AI produces an answer in seconds, but checking whether that answer is correct takes longer than finding it manually would have. By mid-2025, 42% of companies had abandoned their generative AI initiatives entirely, up from 17% the prior year (MIT, 2025).
The architecture was wrong. Due diligence is not a single question answered by a single document. It is a multi-step process: extract data from multiple sources, validate consistency across documents, flag contradictions, assess materiality, and synthesize findings into a view that an Investment Committee can act on. Single-turn question-and-answer tools cannot do this.
What a Purpose-Built DD System Actually Does
The shift is from search (find information in documents) to verification (challenge claims across documents). A purpose-built DD system works differently from generic AI in three ways:
1. Cross-workstream analysis. Traditional DD is organized in silos. The commercial advisor writes one report. The financial advisor writes another. The ESG advisor writes a third. Nobody cross-references them. A purpose-built system reads all workstreams simultaneously and flags connections that siloed advisors miss.
Example: the commercial DD says customer concentration is manageable. The ESG assessment flags supply chain dependency in the same geography. Together, these create a compounding risk that neither report surfaces on its own.
2. Sell-side framing detection. VDD reports are advocacy documents. They present real data, but they choose which data to emphasize. A purpose-built system compares the claims in a VDD against the raw source documents and flags where the framing is selective: where growth rates are presented without the denominator, where "adjusted" metrics diverge from reported figures, where positive trends are highlighted while negative ones appear only in appendices.
3. Source-traced findings. Every finding links to a specific page in a specific document. This is not a summary that an IC member has to trust. It is a citation chain that an investment committee member can verify in minutes. The output is structured: red flags (findings that could change the deal thesis), yellow flags (findings that affect valuation multiples), and green flags (drivers supporting the target IRR).
The Real Cost of Traditional DD
The economics of traditional due diligence are well-documented. For a large-cap buyout, the total DD bill across all workstreams reaches EUR 1-3M. The timeline is 4-8 weeks per workstream. The bottleneck is coordinating across 3-5 advisory firms, each producing their workstream in isolation, none cross-referencing the others. A deal team can bring 20-30% of the routine extraction and cross-referencing work in-house, reducing reliance on the most expensive advisory hours while keeping advisors focused on judgment calls that require human expertise.
The Advisor Question: Replace or Complement?
The question PE funds ask most often about AI in DD is whether it replaces advisors. It does not. A EUR 500M buyout will still require legal counsel, tax structuring, management due diligence through in-person meetings, and negotiation expertise. These are human judgment tasks that AI cannot perform.
What changes is the ratio of grunt work to judgment. Instead of spending 60% of their time on data gathering and 40% on analysis, advisors can invert that ratio. The advisor focuses on the 5 findings that matter rather than the 200 pages that contain them.
For the deal team, the shift is equally practical. Instead of reviewing 200 pages sequentially, the team receives a structured risk assessment: here are the 12 findings across 5 workstreams, ranked by materiality, with source citations. Three are critical. Two of them compound across workstreams in ways that the individual advisor reports do not surface. The time recovered goes to judgment: assessing management quality, evaluating strategic fit, preparing IC materials, and negotiating terms.
The Bottom Line
PE deal teams still read VDD reports themselves because three things are true simultaneously: the reports are not trusted, the first generation of AI tools was not built for verification, and no one has connected the workstreams that advisors deliver in silos.
That is starting to change. The question for deal teams is not whether AI will become part of the DD workflow, but whether they adopt it before or after their competitors do.
Axion Lab builds AI due diligence tools for European private markets. The system uses specialized AI agents to extract, validate, and cross-reference findings across commercial, financial, ESG, legal, and IT DD workstreams, with every finding traced to source pages. Based in Luxembourg.
