AI M&A Due Diligence Benchmark: A Synthetic Data-Room Review

Evidence and classification

Engagement type: Self-run benchmark

Scenario type: Synthetic data room

Execution environment: Production infrastructure

Run date: 2026-07-04

Evidence review date: 2026-08-18

Product workflow: Nexus ยท Enterprise Deep Discovery

Evidence level: Report-backed with negative control, not independently reproducible

Execution-environment labels are first-party historical metadata, not independent verification of hosting, isolation, or workload status.

Named-party boundary: Model providers and competing products named in the comparison

No third party named in this case sponsored, reviewed, or endorsed this work.

Model and provider references describe this run only, not current availability or inventory.

Evidence package

Attached reports are preserved as historical run evidence. Their bytes and embedded metadata are unchanged; presence and hash verification do not validate every claim in a report.

Open claim review: The web narrative below was rewritten on 2026-08-18 to preserve its query intent while removing unsupported historical marketing claims. Unchanged report artifacts, if any, remain first-party historical evidence; this classification block is not independent validation.

Limitations

  • This is a synthetic benchmark, not customer due diligence or a legal/accounting conclusion.
  • The corpus, manifest, answer key and human-review comparison sources are absent.

This evidence block does not establish a verified buyer relationship, independently validate a run, validate checkout, or authorize release.

This was a first-party benchmark using a synthetic data room. It was not customer due diligence, a legal or accounting conclusion, or proof that Nexus reviewed every page or found every planted issue.

The buyer problem

Large data rooms create two different challenges. The first is retrieval: locating documents and passages relevant to a decision. The second is judgment: deciding which evidence changes valuation, deal structure, conditions, or the need for specialist review. Nexus is intended to address the judgment layer after evidence has been retrieved and labeled; it is not a replacement for collection, hosting, privilege review, legal advice, accounting diligence, or responsible deal authority.

A useful benchmark should test both evidence-present and evidence-absent states. The empty-room artifact is therefore conceptually important: when the relevant corpus is unavailable, the workflow should refuse to manufacture findings. But the existence of a negative-control report does not prove detection quality on the populated corpus.

What can and cannot be concluded

The two reports can show the form of historical outputs under different input conditions. They can help a buyer ask whether the system distinguishes missing evidence from negative evidence, identifies unresolved questions, and keeps recommendations conditional on human verification.

The public package cannot establish corpus size, complete reading, recall, precision, issue severity, or superiority to other diligence tools. It cannot support a claim that every planted issue was found or that every reported issue was a deal breaker. It also cannot establish savings against lawyers, accountants, e-discovery teams, or human reviewers without a declared workflow, comparable scope, rates, quality controls, and sensitivity analysis.

Reproducibility requirements

A defensible rerun needs a publishable synthetic corpus or an immutable document manifest; document and page hashes; ingestion, parsing, and deduplication logs; a precommitted issue taxonomy and answer key; evidence citations for every finding; a scoring script; false-positive and false-negative accounting; and a blinded human-review comparison. The empty-room control should be rerun with the same task instructions and should fail safely without inferring absent facts.

The evaluation should report document coverage separately from judgment quality. It should also record where professional review is mandatory and whether a finding changed the recommendation only after a human accepted its source and interpretation.

Product and buyer fit

Enterprise Deep Discovery reduces and organizes a private corpus. Nexus is the overarching strategic-intelligence product that weighs the surviving evidence, alternatives, uncertainty, and dissent. Pointer can test whether a public evidence package is discoverable; it cannot access a private data room or certify diligence accuracy.

This benchmark is useful for scoping a controlled pilot with synthetic or authorized data. It is not transaction advice or evidence of a completed engagement.

Next step

Review the Enterprise evidence and briefing path. The next state opens the Enterprise information page so a buyer can inspect scope, authority, and procurement boundaries before choosing whether to begin a briefing; it does not submit a lead or start paid work automatically.