The Data and Evidence Foundation
The data foundation gives every company, claim, and decision a stable identity and a traceable history. That is what lets an acquisition team reuse prior work without importing stale facts or another deal's confidential information.
Start with the questions reviewers need to answer: which company is this, where did the claim come from, when was it true, and who has checked it?
Define the records that connect the work
Use stable identifiers for records. Names, websites, filenames, and document titles can change.
| Record | Essential fields | Connection |
|---|---|---|
| Company | Identifier, legal names, domains, parent relationships | Targets and operating businesses |
| Thesis | Version, criteria, exclusions, approver | Screening results and deal rationale |
| Source | Version, origin, effective date, ingestion date, permissions | Extracted evidence |
| Claim | Statement, period, unit, source location, verification state | Finding or model assumption |
| Finding | Issue, impact, owner, resolution and evidence | Gate, valuation input or integration task |
| Decision | Outcome, conditions, approver, input versions | Approved baseline |
| Benefit | Baseline, owner, calculation, timing and dependencies | Post-close actuals and forecasts |
Keep relationships explicit. A source document can support several claims. A claim can depend on several sources. A finding may affect both the deal model and an integration task.
Resolve companies before ranking them
A domain is a useful clue, but it is not a legal-entity identifier. Subsidiaries can share websites; an acquired brand may redirect to its parent. A holding company and an operating business may report different financials.
Combine names, domains, addresses, registry identifiers where available, and documented parent relationships. Route ambiguous matches to review. Preserve aliases and merger history rather than merging records solely because a model says the names resemble each other.
Separate discovery deduplication from verified entity resolution. Removing duplicate search results helps the user read a list. It does not establish which legal entity owns the contracts under diligence.
Preserve the meaning of each number
A financial claim needs its value, currency, unit, period, entity scope, and definition. Store whether it is reported, estimated, calculated, or a seller adjustment.
The following hypothetical records contain different measures; they should not be collapsed into one “revenue” field.
| Source statement | Correct interpretation | Open question |
|---|---|---|
| “Revenue: 12m” in a group presentation | Currency and entity scope unresolved | Which companies and reporting period? |
| “ARR: $14m” in a June operating pack | Annual recurring revenue at a point in time | Does it include unsigned commitments? |
| “Sales: $11m” in annual accounts | Historical sales under the accounts' policies | How does this reconcile to the operating pack? |
Missing data stays unknown. Zero means an observed or defined zero. A model should never turn a blank customer-churn cell into a zero-churn assumption without explicit review.
Design retrieval around evidence completeness
Retain page or cell locations alongside extracted text. Record failed pages, unreadable scans, missing sheets, and extraction warnings. Search should show when the corpus is incomplete.
Combine exact search for identifiers and clauses with semantic search for concepts. Apply access filters before retrieval, then check permission again before presenting a source or generated artifact. Indexes and caches need the same boundaries as the original documents.
Record effective dates separately from ingestion dates. A newly uploaded old agreement is not necessarily the current agreement. Preserve amendments and conflicting statements until a reviewer resolves them.
In CorpDev.Ai
Upload the deal materials to AI Room and check processing status before briefing the Analyst. Point it at the relevant files or folder, then inspect the cited source material behind the answer.
Reconcile claims before drafting the story
Build a claim register before drafting the recommendation. Extract the value, currency, unit, period, entity scope, and source location from each document. Group potentially conflicting claims without choosing a winner.
AI extraction can help assemble this register across many files. Require a source location for each entry and leave unsupported fields unknown. Review the underlying page or table before accepting the extracted value.
In a hypothetical case, $14 million of June ARR and $11 million of prior-year sales are different measures. Do not call the difference a $3 million accounting discrepancy until finance has made the measures comparable.
Finance records the accepted definition and reconciliation. Retain the original claims, so the next reviewer can reproduce the reasoning. When a new source arrives, flag dependent findings and models for review rather than erasing the evidence behind the earlier decision.
Continue with continuous sourcing and repeatable diligence.
© 2026 CorpDev.Ai Unified Process for M&A