The Gap: Where Alpha Lives When AI Reads Every Document

✦

An analyst used to spend 40 minutes reading a confidential information memorandum. Highlighting key sections. Taking notes. Summarizing for the deal team. AI tools now do that work in two to three minutes. The analyst reviews the summary, flags questions, and moves on. Multiply that across every CIM, every contract, every financial statement in a deal process, and the time savings are substantial. One estimate puts it at 14 hours per analyst per month freed up from document work. But here is the problem. Every serious private equity firm now has access to these tools. Nearly two-thirds of PE general partners run AI pilots. When every firm uses the same technology to read the same documents, nobody gains an advantage from reading faster. Speed becomes table stakes. Meanwhile, the expert network industry generated $2.5 billion in 2024, up from $1.5 billion in 2020. That growth happened alongside AI adoption, not despite it. The reason is simple. After AI summarizes every document, deal teams still do not know what customers think about the product, whether the sales pipeline is real, or if the competitor everyone dismisses is actually winning deals. Documents contain what management wants investors to see. People who use the product, sell against it, or left the company know what is actually happening. AI reads documents. It cannot have a conversation with a customer who is thinking about canceling. The gap between what documents say and what people know is where investment returns get made. The firms winning with AI understand this. They use automation to free up analyst time, then deploy that time toward conversations AI cannot have.

What AI actually does now

AI reads documents and produces summaries. That is the core capability transforming deal work.

A confidential information memorandum is typically 50 to 100 pages. It contains the company's financials, market position, growth story, and management's view of the opportunity. Before AI, an analyst would spend 30 to 60 minutes reading it, highlighting sections, and writing up key points for the deal team.

Now an analyst uploads the document. AI extracts the financial summary, identifies the key claims, flags inconsistencies, and produces a structured summary. The analyst reviews the output in five to ten minutes, notes questions worth exploring, and moves on.

The same process applies to contracts, financial statements, customer lists, and legal documents. Work that required reading every page now requires reviewing AI-generated summaries and spot-checking the source material.

One study found that AI reduced the time analysts spend on document review by 70 to 80 percent. A task that took an hour takes 15 minutes. A due diligence process that required three weeks of document work can be compressed into one week.

These gains are real. They free analysts from mechanical reading to focus on judgment calls. But the gains are available to anyone who buys the same software.

Why speed stopped being an advantage

Nearly two-thirds of private equity general partners now run AI pilots. Over 40 percent use AI in active deal processes. Adoption is no longer early. It is widespread.

When one firm had AI and others did not, speed was an advantage. That firm could review more deals, move faster on promising opportunities, and spend analyst time on higher-value work while competitors were still reading.

That advantage disappeared when everyone adopted the same tools.

Consider two firms evaluating the same acquisition target. Both use AI to summarize the CIM. Both extract the same financial metrics. Both identify the same questions worth exploring. Neither has an information advantage from the document work. They read the same materials and reached the same starting point.

This is the commoditization problem. When every firm analyzes the same documents with the same AI, the output converges. Everyone starts with the same summary, the same flagged issues, the same questions. The document work becomes infrastructure, like having electricity or internet access. Necessary, but not a source of competitive advantage.

Quant funds discovered this dynamic years ago. When too many firms access the same data with similar models, returns compress. The edge erodes precisely because it becomes widely available. Renaissance Technologies capped its most successful fund after finding that increased scale led to worse returns. More capital chasing the same signals meant less alpha per dollar.

The same logic applies to AI-powered document review. More firms using the same tools on the same documents means less differentiation from the document work itself.

"When every firm uses the same AI to read the same documents, speed becomes infrastructure. It stops being edge."

What documents cannot tell you

A CIM tells you what management wants investors to believe. It contains the growth story, the market opportunity, the financial projections. It does not contain what customers actually think.

Consider a software company being sold. The CIM shows 95 percent customer retention. Strong net revenue retention. Growing average contract values. The numbers look good.

The CIM does not tell you that mid-market customers are frustrated with the product roadmap. It does not tell you that the sales team discounts heavily to close renewals. It does not tell you that a competitor launched a product six months ago that three of the company's largest customers are now evaluating.

That information exists. It lives in the heads of customers who use the product, salespeople who sell against it, former employees who left because they saw problems, and channel partners who hear complaints. None of it appears in the documents AI reads.

This is the gap. Documents contain curated information presented to support a transaction. People who interact with the business daily know what is actually happening. The distance between those two is where investment decisions get made or broken.

After AI summarizes every document in a deal room, deal teams still need to answer fundamental questions. Do customers actually like this product? Is the competitive position as strong as management claims? Will the growth continue or is it already slowing? Are the key employees going to stay?

Documents provide management's answers to these questions. Primary research provides reality's answers.

Why expert networks grew to $2.5 billion

The expert network industry generated an estimated $2.5 billion in global revenue in 2024. That figure is up from $1.5 billion in 2020. The growth happened during the same period AI adoption accelerated.

This is not a contradiction. It is cause and effect.

AI made document work faster. Analysts who spent most of their time reading now had time for other work. The firms that used that time well directed it toward conversations with people who have firsthand knowledge of the companies they were evaluating.

Expert networks connect investors with those people. A former sales director at the target company. A customer who has used the product for three years. A competitor's regional manager who wins and loses deals against them. A channel partner who sees how the product performs in the field.

These conversations produce information that does not exist in documents. A customer can tell you whether they plan to renew. A former employee can tell you why the best engineers left. A competitor can tell you which accounts they are winning and why. None of this appears in a CIM or a financial statement.

The $2.5 billion flowing to expert networks reflects a simple reality. Documents, no matter how fast AI reads them, cannot replace conversations with people who know what is actually happening. The more commoditized document work becomes, the more valuable primary intelligence becomes.

"Documents tell you what management wants you to believe. People tell you what is actually happening."

Three things AI cannot do

AI processes existing information. It reads documents, identifies patterns, and generates summaries. It does not create new information that did not exist before.

Three capabilities remain fundamentally human.

Getting people to share what they actually think. A customer on a reference call gives polished answers. Getting past the talking points requires reading tone, adjusting questions, pushing gently on inconsistencies, and knowing when to stay silent. The most valuable moments in expert conversations happen when someone decides to tell you what they really think instead of what they are supposed to say. Recognizing and creating those moments is human work.

Making sense of conflicting information. Talk to five people about a company and you will hear five different stories. One customer loves the product. Another is about to cancel. A former employee says the culture is toxic. A current employee says it is the best place they have worked. Turning these contradictions into an investment view requires judgment about who to believe, which perspective reflects reality, and when a contradiction is the signal rather than noise. AI cannot weigh credibility or detect when someone is lying.

Deciding what is solid enough to act on. The hard part of primary research is not gathering information. It is deciding that the information is reliable enough to change a position. That decision requires accountability. A human has to stand behind it, explain the reasoning, and accept responsibility if it turns out to be wrong. AI can summarize what people said. It cannot decide whether to believe them.

What the best firms do with freed-up time

AI frees analyst time from document work. What happens next determines whether the firm gains an advantage or just does commodity work faster.

The pattern at firms that capture value from AI looks similar. Analysts spend less time reading and summarizing. They spend more time talking to people. Structured interviews with customers across different segments. Conversations with former employees about what really happens inside the company. Channel checks with partners and competitors to validate or challenge management's claims.

These conversations take time. An expert call requires preparation, scheduling, the conversation itself, and synthesis afterward. Before AI, analysts often did not have time for enough of these conversations because document work consumed their hours. Now they do.

The firms winning with AI treat automation as a way to shift analyst time from reading to conversations. The documents get processed faster. The real work moves to the information AI cannot access.

The firms not capturing value do something different. They use AI to read more documents faster. The same public information, processed more efficiently. The research stack accelerates. The information advantage does not improve.

Closing thoughts

AI changed how deal teams process documents. A CIM that took 40 minutes to read takes three minutes to summarize. Contracts, financial statements, and legal documents that required hours of review now take minutes. Analysts have more time.

The efficiency gains are real. They are also available to every competitor. When every firm uses the same AI on the same documents, speed stops being an advantage. It becomes infrastructure.

The expert network industry grew to $2.5 billion during the same period AI adoption accelerated. That growth reflects what documents cannot provide. After AI summarizes every page, deal teams still do not know what customers think, whether the competitive threat is real, or if the growth story will hold. That information exists in the heads of people who interact with the business daily. Getting them to share it requires human conversation.

AI processes existing information. Primary research creates new information. The gap between what documents say and what people know is where investment returns get made.

The question for every deal team is simple. AI freed up analyst time from document work. Where does that time go now? Toward more conversations with customers, competitors, and former employees? Or toward reading more of the same documents, faster?

One approach closes the gap. The other just moves faster around it.