What AI Can't Tell You: Where Primary Research Still Wins in Investment Due Diligence

Nearly half of dealmakers now use AI tools every day, according to a Sourcescrub survey. Hedge fund analysts are feeding transcripts into large language models. PE associates are using AI to screen targets and summarise CIMs. Corporate strategy teams are automating competitive landscaping.

And most of it is genuinely useful.

But here's what isn't being said clearly enough: AI is exceptionally good at processing information that already exists. It is fundamentally unable to generate the new, proprietary information that separates a good investment decision from a consensus one.

This guide isn't anti-AI. It's pro-clarity. If you're a PE deal team member, a hedge fund analyst, a corporate M&A professional, or a consultant running commercial due diligence, you need to know exactly where AI adds leverage — and where relying on it creates blind spots that can cost you a deal, a thesis, or your credibility.

The AI Hype Cycle in Investment Research: Where We Actually Are

Let's ground this. AI tools — specifically large language models and their derivatives — are now embedded in investment workflows in three main ways:

1. Summarisation and synthesis: Condensing earnings transcripts, CIMs, industry reports, and news flow into digestible briefs.
2. Deal sourcing and screening: Scanning databases, identifying targets that match acquisition criteria, flagging signals in public filings.
3. Pattern recognition: Identifying comparable transactions, clustering companies by characteristics, surfacing statistical relationships across datasets.

These are real, valuable capabilities. A PE associate who used to spend four hours reading a CIM can now get an AI-generated summary in minutes. A hedge fund analyst tracking 30 names can monitor earnings sentiment shifts across all of them simultaneously.

But notice what all three capabilities have in common: they operate on publicly available or already-collected data. They reorganise, repackage, and re-present existing information. They do not create new knowledge.

This is the critical distinction that most "AI in investing" content glosses over — and it's the one that matters most for due diligence.

The Five Things AI Cannot Do in Investment Due Diligence

These aren't theoretical limitations. They're the gaps that show up on live deals, in real portfolio decisions, and during the diligence processes where capital actually gets committed or pulled.

1. AI Can't Tell You What Customers Actually Think

AI can scrape G2 reviews, NPS data if it's public, and social media sentiment. What it cannot do is talk to 15 enterprise customers of a B2B software target and ask them:

• Would you renew this contract at a 20% price increase?
• How does this vendor compare to the two alternatives you evaluated last quarter?
• If this company were acquired and the account team changed, how would that affect your decision to stay?

These are the questions that determine revenue quality in a PE deal. They're the questions that tell a hedge fund analyst whether a company's reported net retention rate is sustainable or masking churn risk. And they require a human asking another human a direct question, in context, with the ability to follow up.

Where primary research wins: Customer interviews and B2B surveys deliver proprietary signal on retention risk, pricing power, competitive vulnerability, and switching costs — none of which exists in any dataset an AI model can access.

2. AI Can't Validate What a Management Team Is Claiming

During a sell-side process, the management presentation will tell you the TAM is $8 billion, the product is best-in-class, and the pipeline has never been stronger. AI can help you check whether the TAM methodology is reasonable by cross-referencing market reports. But it can't tell you whether the claims are true on the ground.

That requires talking to:

• Former employees who know the internal reality behind the metrics
• Channel partners who can confirm or deny sell-through volumes
• Industry practitioners who can benchmark the company's capabilities against competitors
• Customers who can speak to actual product quality, not marketing claims

AI processes narratives. Primary research pressure-tests them.

Where primary research wins: Expert interviews with former executives, competitors, channel partners, and customers create a triangulated view of reality that no amount of public-data processing can replicate.

3. AI Can't Surface What Isn't Being Said

This is perhaps the most underappreciated gap. AI works on text — it analyses what has been written, spoken, and published. But some of the most valuable insights in due diligence are things that nobody has said publicly:

• A key product line is quietly losing share to a new entrant that hasn't shown up in industry reports yet
• The target's largest customer is in late-stage evaluation of a competing vendor
• A regulatory change in draft form is expected to significantly impact the target's core market within 18 months
• The company's technical talent has been quietly leaving for a specific competitor

These signals don't exist in structured data. They live in the heads of people who are close to the situation — practitioners, customers, former employees, competitors, regulators. Getting to them requires asking the right questions to the right people, which is the definition of primary research.

Where primary research wins: Channel checks and expert calls uncover emerging risks and opportunities that haven't entered the public information ecosystem yet — the precise information that generates differentiated investment edge.

4. AI Can't Judge Context, Nuance, or Credibility

An AI model can summarise what an expert said on a call. It can even flag contradictions between two transcripts. What it cannot do is assess:

• Whether an expert's perspective is biased by their specific role, tenure, or departure circumstances
• Whether a confident-sounding answer is actually hedging
• Whether the real insight came from a throwaway comment in minute 38 that contradicted the expert's initial framing
• Whether three separate data points from different sources form a pattern that changes the entire thesis

Judgement under uncertainty — weighing conflicting inputs, assessing credibility, knowing when to push harder on a line of questioning — is a fundamentally human skill. It's what experienced research analysts and diligence professionals bring to every project. AI can assist with organising inputs, but the synthesis that drives conviction requires human cognition.

Where primary research wins: Skilled interviewers adapt in real time, probe inconsistencies, weigh source credibility, and synthesise conflicting inputs into a coherent view — capabilities that remain beyond AI's reach.

5. AI Can't Generate Proprietary Data on Demand

Need to know what percentage of orthopaedic surgeons in the UK are considering switching from Vendor A to Vendor B in the next 12 months? No AI tool can answer that. The data doesn't exist until someone designs a survey, recruits qualified respondents, fields it, and analyses the results.

Need to understand how mid-market CFOs in the DACH region are thinking about their ERP migration timelines? Same thing. That data has to be created through structured primary research.

AI is a consumption engine. Primary research is a creation engine. They serve fundamentally different purposes in the diligence process.

Where primary research wins: Custom B2B surveys and structured interview programmes generate bespoke datasets that answer your specific investment questions — data that didn't exist before you commissioned it and that your competitors don't have.

A Practical Framework: When to Use AI vs. Primary Research in Your Workflow

Rather than treating this as a binary choice, the most effective investment teams use AI and primary research as complements. Here's a practical framework for knowing which tool to reach for at each stage:

The pattern is clear: AI handles breadth and speed on public information. Primary research handles depth and proprietary insight on the questions that actually drive the decision.

The Real Risk: AI-Driven Consensus

Here's the strategic risk that few people are discussing openly. As AI tools become ubiquitous across investment firms, everyone is processing the same public information through increasingly similar models. The outputs converge. The summaries look alike. The screening criteria overlap.

This creates a new form of consensus risk. If every PE firm is using AI to screen the same databases and summarise the same CIMs, the differentiation doesn't come from the AI layer — it comes from what you do after the AI has done its work.

For hedge fund analysts, this is the classic variant perception problem: if your research process is identical to your competitors', your insights will be too. AI accelerates this convergence.

For PE deal teams, the implication is that commercial due diligence — the primary research that validates or kills a deal — becomes more important in an AI-saturated environment, not less. It's the remaining source of genuine informational advantage.

For corporate strategy and M&A teams, the risk is building acquisition recommendations on AI-synthesised secondary research that looks rigorous but lacks the ground-truth validation that prevents expensive mistakes.

What This Means for How You Staff and Budget for Research

If you accept the framework above, the practical implications for your team are:

1. Don't cut primary research budgets because you've adopted AI tools. The tools make secondary research faster and cheaper. They don't replace the proprietary insights that come from talking to customers, experts, and market participants.
2. Use AI to make primary research more targeted. Let AI do the initial landscaping, identify the key questions, and narrow the focus. Then deploy primary research against the specific unknowns that matter. This makes every expert call and survey more efficient.
3. Invest in research partners who can execute primary research end-to-end. If AI is handling more of the desk research, your team's time is better spent interpreting primary research findings and making decisions — not scheduling expert calls and writing discussion guides. Delegating the execution of primary research to a specialist provider frees your team to focus on judgement, which is where humans still have an irreplaceable edge.
4. Build workflows that combine both. The best research processes in 2025 look like this: AI-assisted screening → targeted primary research → AI-assisted synthesis of primary findings → human judgement on the investment decision. Neither AI nor primary research alone is optimal. The combination is.

The Bottom Line

AI is a powerful tool for investment professionals. It accelerates desk research, improves screening efficiency, and helps teams process more information faster. None of that is in question.

But AI operates on the information that already exists in the world. It cannot create new, proprietary knowledge. It cannot talk to a customer. It cannot sense that an expert is hedging. It cannot design a survey to answer a question nobody has asked before. It cannot tell you what's happening on the ground in a market that hasn't been written about yet.

Primary research — expert interviews, B2B surveys, channel checks — remains the only way to generate the differentiated, proprietary insights that drive conviction in investment decisions. In an environment where AI is making public information more accessible to everyone, the competitive advantage increasingly belongs to the teams that are better at generating and interpreting private information.

That's what primary research does. And that's why it still wins where it matters most.