What Claude Cannot Do in M&A: An Honest Assessment

Short answer. Claude is genuinely useful in M&A. It compresses document analysis, accelerates model-building, improves IC preparation, and produces working drafts of outputs that previously took days to create. It is also consistently overhyped by people who have not used it seriously in a live deal context. This article is about the limits. Not to discourage use, but because understanding where Claude stops being useful is what determines whether it adds value or creates false confidence in a deal process.

The limits that matter in practice

Claude has no real-time data

Claude's knowledge has a training cutoff. It does not have access to live market data, current comparable transaction multiples, today's credit spreads, or recent precedent transactions unless you provide that information directly in the conversation.

In practice this means Claude cannot tell you what a business in your sector is trading at today. It cannot pull the most recent comparable deals. It cannot confirm current market conditions for the financing structure you are considering. For any analysis that depends on current market intelligence, Claude needs to be fed that information by the user or combined with a tool that has live data access.

Deal teams that ask Claude to benchmark a valuation against current market conditions without providing that data will get an answer that sounds credible and may be materially wrong. The risk is not that Claude refuses to answer. It is that it answers confidently with information that was accurate at training and may no longer be.

Claude cannot access documents it has not been given

Claude works with what is in the conversation. It cannot access your data room, your internal drive, your CRM, or any external system unless those documents are explicitly uploaded or pasted into the conversation.

The practical implication for deal teams is that the quality of Claude's output is entirely dependent on the quality and completeness of the input. A document analysis is only as good as the documents provided. A synergy stress-test is only as good as the synergy assumptions shared. An IC briefing card is only as good as the diligence findings surfaced.

This sounds obvious. In practice it creates a workflow discipline problem. Deal teams under time pressure take shortcuts in what they upload. Claude produces outputs that look comprehensive but are built on incomplete information. The outputs are then used with more confidence than the underlying inputs justify.

Claude cannot verify what it is told

If you tell Claude that the target's EBITDA is €10 million, it will work with €10 million. It has no way to verify that figure against the management accounts, check it against the financial model, or flag that the number looks inconsistent with other information you have provided elsewhere.

This is a fundamental limitation in any deal context where data accuracy is critical. Claude is a reasoning engine, not an auditing system. It processes what it receives. It cannot independently verify whether what it receives is accurate, complete, or consistent with other sources.

The output quality therefore depends entirely on the user's discipline in providing accurate, complete, and consistent inputs. In a diligence process where the whole point is to verify what management is telling you, this limitation is significant. Claude can help you think through what the numbers mean. It cannot tell you whether the numbers are right.

Claude cannot replace judgment on deal decisions

Claude can identify risks, structure analytical frameworks, stress-test assumptions, and produce IC-ready summaries. What it cannot do is exercise the judgment that determines whether a deal should proceed, at what price, and on what terms.

Deal judgment is built from pattern recognition across many transactions, deep sector knowledge, an understanding of the specific buyer's strategy and risk tolerance, and the kind of contextual intelligence that comes from years of operating in a market. Claude has processed a large amount of information about M&A. It has not done a deal. The difference matters in ways that are not always obvious until the judgment call arrives.

The risk is subtle rather than obvious. Claude's outputs are well-structured, confident in tone, and analytically coherent. They can create the impression of rigorous analysis where the underlying reasoning is plausible but not grounded in the deal-specific context that real judgment requires. A junior analyst using Claude outputs as a substitute for their own analysis rather than as a starting point for it is making a workflow error that the quality of the output does not flag.

Claude cannot read behavioral signals

Claude can analyze text and identify patterns in language. It can assess whether a management presentation is internally consistent, whether the stated rationale for a transaction holds together logically, and whether there are gaps in the narrative a seller is presenting.

What it cannot do is measure the behavioral patterns that predict execution outcomes. Self-reported information about culture, leadership style, and organizational behavior, the kind that appears in management presentations and diligence Q&A, is exactly the kind of information that self-presentation distorts. Claude can read it. It cannot see through it.

The behavioral signals that predict whether a leadership team will execute under integration pressure, whether a management team's stated commitment to the deal reflects their actual behavioral orientation, or whether the cultural compatibility between acquirer and target is real or performed. These require behavioral measurement methodology that goes beyond language analysis of consciously constructed text. Claude is not a substitute for that.

Claude hallucinates

This is the limit that deal teams are most reluctant to acknowledge because the outputs so often look right.

Claude will occasionally produce information that is plausible, well-formatted, and factually incorrect. A statistic that sounds real but is not. A precedent transaction that does not exist. A regulatory requirement that is either outdated or simply wrong. The frequency of these errors is lower than it used to be and lower than many people assume. The consequences in a deal context of acting on a single material error are significant.

The discipline required is consistent: every factual claim Claude produces that will be relied on in a decision needs to be verified against a primary source. Not most claims. Every claim. This is not how most people use Claude in practice, and the gap between the discipline required and the discipline applied is where the hallucination risk actually lives.

Claude does not know what it does not know

Related to hallucination but distinct from it: Claude does not always signal uncertainty when it should. It will answer questions outside its reliable knowledge range with the same confident structure it uses for questions inside it. The absence of a disclaimer is not evidence that the information is reliable.

In an M&A context where the stakes of acting on wrong information are high, the absence of expressed uncertainty is a risk. Users need to apply their own skepticism rather than relying on Claude to flag it. When in doubt, always ask Claude explicitly to identify where its evidence is thin, where it is inferring rather than concluding, and where it would need additional information to form a reliable view. This prompt discipline does not eliminate the problem but it significantly reduces it.

The workflow errors these limits create

Understanding Claude's limits in isolation is less useful than understanding the specific workflow errors they produce in M&A practice.

Using Claude outputs without verifying inputs. The most common error. Claude produces a well-structured analysis based on incomplete or inaccurate inputs. The analysis looks credible. It gets used. The problem surfaces later when the decision built on the analysis turns out to have been built on data that was not what it appeared to be.

Treating Claude's confidence as accuracy. Claude writes in a tone that implies analytical certainty. That tone does not track with the actual reliability of the underlying information. Deal teams need to maintain their own skepticism about Claude outputs regardless of how confident the language sounds.

Using Claude as a replacement for sector expertise. Claude has broad knowledge of M&A principles and general sector patterns. It does not have the granular, current, relationship-grounded sector knowledge that experienced deal professionals bring. Using it to substitute for that expertise rather than to support it produces analysis that sounds right and is missing the sector-specific context that changes the conclusion.

Over-relying on Claude for behavioral assessment. As discussed, Claude can analyze stated information about culture and leadership. It cannot measure actual behavioral patterns. Using Claude analysis of management presentations as a proxy for behavioral diligence is a category error with potentially significant consequences for integration outcomes.

What Claude is actually for in M&A

The limits described above are not arguments against using Claude. They are arguments for using it correctly.

Claude is a reasoning and drafting tool. It compresses the mechanical work of analysis, synthesis, and output production. It is most valuable when it is given accurate, complete information and asked to reason from it, structure it, stress-test it, and express it in a format that serves a specific decision.

It is not a research tool, a verification system, a behavioral assessment, a substitute for sector expertise, or a replacement for deal judgment. Every application that treats it as any of those things will produce outputs that look valuable and may be materially misleading.

The deal professionals getting the most from Claude are the ones who understand this distinction clearly. They use Claude to do the things it does well faster than they could do them manually. They apply their own judgment, their own sector knowledge, and their own verification discipline to everything Claude produces. And they treat Claude outputs as starting points rather than conclusions.

That combination, Claude's analytical speed and their own professional judgment, is significantly more powerful than either alone. The combination requires knowing exactly where one ends and the other begins.