When Anyone Can Produce the Average, What Will YOUR Research Be Worth?
Reflections from moderating Singletrack's Q3 2026 Benchmark panel...
As AI makes research faster to produce and easier to find, its value is concentrating in the insight nobody else has and the relationships clients trust.
AI-driven speed and scale are coming to sell-side research. This, of course, does not mean that the research will always be better. Nor that the buy side will pay more for said research. The size of the check will still be set by relative value among providers, and anything that becomes easy for everyone quickly stops commanding a premium. What clients have typically paid a premium for is insight and access they can't get anywhere else, delivered by a person they trust. Neither of these scales the way research production and distribution now can.
I tested this idea on September 15 as moderator of Singletrack's Q3 2026 Benchmark webinar, built around its Quarterly Benchmark Report, with Julia Ashworth (Arete Research and Euro IRP), James Ferguson (Loop Capital), and Vineet Jobanputra and Brijesh Malkan of Singletrack. The panel reinforced what the Report suggests: as AI takes over more of the production and distribution of research, value is concentrating in distinctive insight and trusted relationships. That is true across the industry, and the firms most likely to capture that value are those that meet two conditions: 1. producing something a machine can't approximate, and 2. building their AI solutions on client data they can trust.
When a machine reads the research first, what will set yours apart?
For most of my career, every analyst was taught that nobody reads past page five of a research report, and that's on a good day. That rule assumed a human reader. According to the Report, aggregators and client portals now account for 62% of research downloads, and clients are increasingly running research through AI tools that can ingest a 40-page note in seconds.
The obvious reaction may be to start writing for the machine, but that misses a crucial point. Published research is a marketing engine and leading signal for the meetings and access clients actually pay for, and it's differentiated analysis and relationships that drive those bookings.
What can't be ignored is that the same tools that summarize anyone's research can also approximate much of it, quickly and easily. As a client had just told Brijesh, AI has "raised the bar of what we consider commoditized." What stays valuable is what exists only inside your firm, so no model has seen it and none can reproduce it: proprietary data, a distinctive investment framework, and an analyst's clearly argued view on where a stock or sector is about to turn.
Bad data and generic algorithms: the fast track to obsolescence
Any quantitative AI solution a firm builds or buys is only as good as the client data underneath it. I've been banging the table on this for a long time, to mixed reception. Brijesh put the risk bluntly, warning of "a very efficient path to obsolescence": servicing clients from poor data, with the wrong meeting slots ending up in front of the wrong clients. In his words, "it's not just junk in, junk out, it's a very, very quick junk in, junk out."
Many AI providers offer excellent algorithms but with no specific expertise in finance and no effort to create a layer of trustworthy client data to work from. Any experienced salesperson or analyst will quickly stop trusting outputs that don't reflect the realities of their business or their clients. A vertically integrated solution that understands the financial services ecosystem and handles the data accuracy first is crucial. As I argued in Building AI Without Fixing Your Client Data? Good Luck, a single client can be reflected differently (or duplicatively) across a dozen internal systems, and until that is resolved, a "next best action" recommendation is an overconfident guess. There are vendors that get it right, and I've even worked with some of them, but they are few and far between.
AI can't scale exclusivity
According to the Report, corporate access meeting volumes rose 17.6% year-over-year, driven mostly by large-format group meetings. Group formats are easy to scale, so firms are scaling them, resulting in marginal additional revenue at best. Exclusive one-on-one time with management can't be scaled, yet it is the service that actually pays. AI's role here needs to be to optimize the revenue driven through each allocation.
Corporate access desks are wildly overworked, and as I wrote in The AI-Powered, Human-Driven Corporate Access Revolution, the challenge in optimal allocation is one of scale rather than the talent or effort of these teams. Quality AI can (accurately) rank which investors to invite, surface clients who should have asked for a meeting, and show each client's allocations across every event at once. On the panel, Brijesh noted that allocation is still human-first but is starting to be checked against an optimized allocation, and that the real efficiency gain is in event administration. I agree the administrative savings are real, but I would push further: the bigger prize is revenue. Applying models and pattern recognition at scale, across every allocation decision, shows which combinations of clients, companies and formats actually drive the broker vote and revenue, something no manual process can see.
The same is true of the exclusivity value of the salesperson. One who speaks with twelve other PMs about a note knows how the market is reacting to it, and no model can summarize that because it was never written down. James described his overarching view of AI as automating low-value activities to focus on high-value ones, and every hour saved on administration is an hour spent in exactly those conversations.
When AI strips the byline, the revenue goes with it
Even the best research won't get paid for if no one knows where it came from, which makes intellectual property its own obstacle. As AI tools and platforms ingest, summarize and redistribute research, the analyst and firm that produced it risk losing both control and credit. Julia called research "the golden source," and when Brijesh pointed out that leakage is as old as the emailed PDF, her reply captured what has changed: "AI accelerates the problem."
Firms should be open to walking away from any platform that can't explain how their research will be attributed or what the commercials will be, because there is no revenue without attribution. If a client's AI tool delivers the insight without naming the analyst or the firm, the client doesn't know whom to call, doesn't request the meeting, and doesn't reward that broker.
What the winners will do differently
There is a topic that deserves far more attention than it gets: the engagement experience of the buy-side, and using AI and technology to service clients optimally. Like the industry as a whole, the panel gave it hardly any airtime, yet I think it is one of the biggest differentiators available to the sell side right now. Nearly every other major industry has built systems focused on this because ease of engagement drives revenue. As I argued in The Constraint On Research Modernization is Not Technology, the capability to deliver personalized, AI-enabled client engagement at scale already exists, and the question is whether firms will commit to applying it. The firms that move first will set the expectations everyone else is forced to meet, while the rest simply produce the average, faster.
My thanks to Julia, James, Vineet and Brijesh for a candid discussion, and to Singletrack for hosting. As always, I'd love to hear what you think.
KTB
