AI Can Read Millions of Pages. So Why Do Investors Still Call Experts?
A language model can summarize a decade of 10-Ks before an analyst finishes their coffee. It can compare financial disclosures across companies, flag an unusual passage buried in a footnote, and produce an initial investment thesis in the time it takes most people to open their inbox.
Given all of that, it would be reasonable to assume artificial intelligence has already replaced the phone call to an industry expert. CFA Institute’s broader research on AI and finance frames this as a genuine transformation of investment decision-making, not a passing trend. And yet investors keep making those calls anyway. The reason has less to do with what AI can never do than with what a specific kind of question requires when public data simply cannot answer it.
None of what follows is a knock on the technology. AI genuinely excels at financial document analysis, pulling structured facts out of filings and earnings-call transcripts faster than any analyst could by hand. It compares companies across dozens of metrics at once, spots patterns across enormous datasets, and generates a plausible investment hypothesis worth testing. CFA Institute’s research describes generative AI already handling data extraction and report preparation in real workflows today, freeing analysts to spend more time interpreting information instead of typing it up. That shift is real, and it is valuable.
The trouble starts with a different kind of question. Why are customers actually switching suppliers this quarter? Is a company’s product genuinely better than its competitor’s, or just better marketed? What are rivals saying privately about a company’s prospects? How difficult would it really be for a new entrant to break into this market? Public filings and historical text rarely answer these questions reliably, because part of the answer often exists as tacit knowledge, the kind of thing that has never been written down anywhere at all.
The Difference Between Information and Firsthand Knowledge
There’s a useful distinction hiding underneath all of this. Information is what has already been written down: a filing, a transcript, a press release. Firsthand knowledge is what a person carries because they were actually there when something happened.
A former executive can offer context about how a strategic decision really got made, beyond the sanitized version that eventually reached a press release. A customer can describe how a product performed against their actual expectations, not the marketed ones. A supplier may see pricing pressure building months before it becomes visible in reported results. A distributor, an engineer, a regulator, a competitor, each holds a version of the story that public documentation was never designed to capture in the first place.
Context and motive resist translation into a spreadsheet. Industry culture shapes decisions in ways an outsider reading a transcript will often miss entirely, and unwritten rules govern how a market actually works in ways no published standard fully captures. Customer sentiment can shift well before it shows up in a churn number, and competitive dynamics sometimes move faster than any quarterly filing can keep pace with. T. Rowe Price frames this succinctly: AI compresses the information edge, which makes human judgment the differentiator that remains.
What Investors Are Actually Trying to Learn
The specific questions an investor brings to an expert call vary enormously by sector, but they tend to cluster around a handful of recurring themes: is the market really growing as fast as the reports suggest, and is that growth structural or cyclical? Who is actually winning customers right now, and what gives them an edge that competitors can’t easily copy? Would this customer buy the product again, and what almost stopped them from buying it the first time? Is management’s stated strategy credible, and has this team actually delivered on what it promised before? What has changed in the last twelve months that hasn’t hit the news yet?
None of these are questions a document can answer on its own, because the honest answer usually depends on reading between the lines of what several different people, with different incentives, are willing to say out loud.
Why a Conversation Reveals Things a Summary Can’t
A conversation can surface tacit knowledge, understanding a person has but has never written down, along with recent developments not yet visible anywhere in the public record. It can also surface contradiction, and that contradiction is itself useful. Two experts in the same industry often disagree with each other, and that disagreement is a signal worth sitting with rather than smoothing over. A single AI-generated summary, by its nature, tends to flatten disagreement into something that reads like consensus, even when no real consensus exists.
Compliant expert interviews are built around avoiding confidential information and material nonpublic information. Reputable networks screen experts for conflicts of interest, provide compliance guidance before a call begins, and instruct participants not to disclose anything restricted, though ultimately, both the expert and the client carry responsibility for staying within those lines.
It Was Never Really a Contest
Even the researchers building AI specifically for equity research seem to agree the technology isn’t there yet, at least not on its own. A 2024 FinRobot preprint identifies clear limitations in existing AI equity-research systems, pointing to a narrow focus on technical factors and a limited capacity for discretionary judgment, precisely the kind of ambiguous, hard-to-quantify territory where a fund manager’s judgment about why a stock is undervalued doesn’t show up cleanly in the data. CFA Institute’s more recent analysis makes a related point, describing competitive advantage shifting toward governance, process, and judgment as AI capability continues to grow.
That framing matters, because it suggests the real skill set isn’t AI versus human judgment. It’s knowing which kind of question you’re actually asking. AI is very good at finding information across enormous datasets, processing it at scale, comparing companies across dozens of metrics, and generating a starting hypothesis. Humans remain better at explaining why something happened, providing context a document never captured, deciding what actually matters among everything AI surfaces, challenging an assumption that sounds right but might not be, and interpreting a signal that’s genuinely ambiguous or contradictory. Neither side of that list is optional if the goal is a research process that actually holds up.
Where Expert Networks Fit Into the Process
This is where an expert network earns its place, rather than feeling bolted onto the workflow as an afterthought. Finding the right industry expert and screening them for relevant experience and conflicts of interest is a research task in its own right, one a fund’s own analysts rarely have time to run alone, long before the actual interview even happens. Providers ranging from established networks to pay-per-engagement options like Nexus Expert Research exist specifically to run that search on a deal or research timeline that the underlying decision actually requires.
The output of that process is primary research: qualitative, firsthand, and specific enough to genuinely validate or challenge an investment thesis, rather than simply adding another document to the pile of things an analyst has already read.
A modern research workflow tends to move through a fairly consistent sequence, even if nobody writes it down as a formal process. AI finds and organizes the available public information first. An analyst reviews that output and identifies the specific questions it leaves unanswered. An expert network then locates people with genuine firsthand knowledge of those exact questions, and expert interviews test the assumptions the analyst couldn’t confirm from documents alone. The investment thesis gets updated based on what those conversations actually revealed. AI and the analyst then keep monitoring the thesis as new information arrives, ready to flag when something shifts enough to justify another round of calls.
A well-run expert call tends to work through the same handful of themes almost regardless of sector: the real growth rate behind a market and how durable that demand actually is, customer satisfaction and switching behavior, who is genuinely winning share and why, where pricing power sits and how it’s shifting, which regulatory changes are actually coming and how they’ll land in practice, what’s genuinely differentiated about a product versus what’s just marketing language, where the real operational bottlenecks and dependencies sit, whether a management team’s stated strategy matches what practitioners on the ground actually see, and what’s changed in the last year that hasn’t made it into the news cycle yet.
Neither Side Replaces the Other
AI makes information abundant. Experts make information meaningful. Neither claim replaces the other, and treating this as a contest misses what’s actually changing in how research gets done. When a question can’t be answered from public information alone, investors with rapid access to relevant experts are simply better positioned to test what remains genuinely unanswered, whether that access comes through an established provider or a pay-per-engagement network like Nexus Expert Research built to arrange that conversation quickly, on the timeline the decision actually demands.
