On June 11th, 2026, we were delighted to welcome over 450 leaders from the investment research industry in North America to the Metropolitan Club in New York for our 9^th annual Unbundling Uncovered conference. Our panels delved into everything from AI’s implications for buy and sell-side research relationships, to the integration of research and data in investment workflows and budgets, and the likelihood of Europe aligning with American research funding models once again. While the following insights sometimes contradict each other, they demonstrate the healthy debate among key market stakeholders over the course of the day.  

Here are our 5 key takeaways:  

AI is transforming how the buy and sell-side supplies and consumes research 

The research industry is moving to frictionless API/MCP ingestion rather than sell-side portal access, which many panelists declared 100% dead. AI consumption of research is also beginning to replace some traditional email engagement, as panelists noted the first significant drop in email open rates from Q1 2025. Content that previously had a one-day shelf life with 20% open rates is now providing contextual value over time within LLMs that surface the other 80%. LLM-driven curation of content is also changing readership market share on major aggregation platforms which could influence future market share numbers, as payments for direct analyst access broaden. 

From a commercial perspective, whether that usage falls under existing agreements or warrants incremental payment is still being negotiated, but the direction is toward more granular licensing and MCP-standard delivery. The biggest contracting friction right now is pricing. Buy-side firms stated that existing full-service licenses should cover AI-assisted consumption of content that they’re already paying for, while the sell-side gave the opposite view (especially if the overall relationship isn’t a meaningful one!).  

Research providers that treat AI as a shortcut to replace analyst relationships are misreading the market, as that will always be the core value proposition. However, analysts will need to have a well-rounded skill set to succeed, as the “numbers-only” analyst approach may not work in this new environment. One output of all this is that spec sales is growing in importance. Deep sector knowledge, combined with deep relationships, is what clients increasingly value.  

The “junior analyst harnessing AI” use case is a notable bright spot but also a growing concern. AI is successfully compressing the time it takes for newer analysts to reach productivity, but graduates who have relied on AI may lack the analytical grit that defines the future great analysts. Sell-side heads of research are actively designing programs to preserve that development path. 

Transparency is still a core issue for the sell side as they assess their value delivery 

As the new technology takes hold there is the potential for richer, more granular attribution, independent of what the buy-side decides to provide in their research evaluation process. Attribution, transparency, and consumption tracking are table stakes if buy-side firms want to ingest sell-side research into AI/LLM workflows. The “black box” concern is real but solvable. The technology now exists to track whether an LLM surfaced a piece of content, used it in summaries, or prompted a full read-through – not just “they opened the note,” but which paragraphs, KPIs, and data points were used in a model or to get to an investment decision.  

This will be important, because once again more granular, transparent voting was the clear ask from the sell-side and strongly advocated by a couple of vendors. Current voting frameworks underweight quality versus quantity, fail to capture the value of archives and AI-surfaced content, and don’t adequately highlight emerging talent, according to a selection of speakers. Indeed, buy-side firms are investing in clearer broker feedback loops, rate card communication, and scorecard consistency, but attribution to named buy-side individuals remains a hard limit for many, driven by the past experience of it being “weaponized.” (Note from SR – that’s the third conference that the term “weaponized” has been used!) The sell-side responded that any instances of transparent buy-side research valuation data being used to hassle PMs are incredibly rare and would be stamped out swiftly by any reputable broker.  

Advice came from panelists that the most effective transparency approach was to incentivize the behaviors you want to see, but you don’t need to provide raw granular data that creates noise – there’s a happy medium that still moves us along the transparency curve. (Note from SR – that’s why we are so excited about our imminent analyst rankings!) 

Rate cards should drive behavior and often don’t. For example, in corporate access, funds that want premium access (true one-on-ones with senior management) need their compensation structure to reflect that, and misalignment between what firms say they want and how they pay remains common.  

How firms implement AI infrastructure will dictate winners and losers 

The buy-side is in active experimentation mode, buying third-party tools while also building internally. AI FOMO is a real concern, and firms are now mandating usage internally to drive adoption. Buy-side firms are using only ~7% of generative AI’s capacity and two thirds of CEOs report 20%+ ROI on AI investments (though there was significant skepticism about those claims!). 

The models themselves (Claude, GPT, Gemini) are moving to a more commoditized phase. The potential for differentiation lies in how well your content is structured, connected, and made available to those models. On the supply side firms that invest in linking research, filings, CRM content, sales interactions etc. will generate meaningfully better AI outputs.  

Agent orchestration beats single-agent thinking. The most effective implementations break analyst workflows into discrete agents with specific roles (one builds, one reviews, one validates), while overloading a single prompt degrades output quality significantly. 

However, cost management is a growing focus. Token costs are the new headcount cost – as AI usage scales, token spend will become a core P&L line item. The smart approach is model arbitrage, using frontier models only where genuinely needed and routing routine tasks (document review, data extraction, formatting) to cheaper, lighter models. As AI providers move from subsidized to a truly monetized stage, the cost of AI-driven workflows will become visible. Firms need to plan for this as part of their technology budgeting – as we know from market data, the word “enterprise” doesn’t mean you are protected from large price increases on renewal once you are hooked! 

AI is blurring the lines between research and data   

The focus has shifted from finding and accessing data to making it AI-ready. Firms have enough data (kind of)! The largest buy-side firms are now evaluating vendors on the quality of their MCP layers, semantic files, and structured data delivery, not just the UI or insights themselves. 

The buy-side goal is to scale large unstructured datasets across all their teams in areas that provide compelling value. For example, AI is enabling quants to work with unstructured data in ways previously reserved for fundamental analysts, and reducing historical data requirements from 8 years to as few as 2-3. The lines between data-driven and research-driven investing are blurring. 

AI has forced existing, as well as new, potential providers of data to go through an intensive data cleansing period, allowing for compelling monetization in areas that would have been challenging previously. And data providers, in common with brokers earlier in the day, made the case that the creation of a translation layer, with the associated feeds and MCPs of historical datasets should be accompanied by a new price tag. On the other hand, buy-side firms don’t expect to increase total data and research budgets beyond the current trends, so will have to reallocate budget within fixed pools. The broker/data budget split is expected to converge within 1-2 years, giving investment teams more flexibility to determine their own mix but also creating tough choices if this becomes a zero-sum game.  

If CSAs in Europe are going to happen, the dominoes will be falling by year-end 

Panelists said that CSAs in Europe will be good for asset managers, asset owners and research providers alike – the benefits to performance outweigh the single basis point cost, hence the US’s happiness to continue with 28e throughout MiFID. However, the buy-side “first mover” logjam persists – a growing group of asset managers are indeed ready to adopt global CSAs, but none want to be the leaders of this move. Many large buy-side firms have completed their extensive preparations, but project teams are now at risk of being deprioritized as competing internal initiatives (AI, OMS implementations) demand attention. Some smaller asset managers have already made the transition, so a “trickle up” dynamic may be more likely than the originally anticipated trickle-down from the large globals.  

Many firms with EU-domiciled funds were waiting for the new EU MiFID rules to be implemented in early June before transitioning all funds simultaneously, and this explains why several firms that targeted H1 2026 didn’t move, and why 2027 research budgets remain the target implementation date many firms are working toward. The next 6 months will be telling, as this should be the time when a comprehensive industry move would gain pace.   

The consensus trigger required for a clear buy-side domino effect to happen is either a critical mass of mid-sized firms moving first to build momentum, or a more prescriptive regulatory pronouncement from the FCA/EU regulators. Urgency was building from firms keen for this to become an industry trend before momentum dissipated, and the point was made that having the research cost conversation with clients in a stable market is far preferable to introducing it during a correction, when CFOs will be cutting budgets and research spend becomes an easy target yet again. 

Conclusion 

The case was made by research and data providers that packaging and delivering content creates costs and delivers previously untapped value, and therefore commercial models should recognize this. The buy-side’s assertion that budgets aren’t going up arbitrarily, and existing agreements and relationships should be adequate, obviously clashes with this view. So how will this play out? It’s clear if you represent a big enough client relationship for a provider then the conversation may be a more agreeable one – so perhaps this dynamic creates even greater concentration in a market already dominated by the biggest players on both sides? 

For this not to be the case it needs to become obvious that AI is augmenting the investment process not just in terms of efficiency but in actually delivering alpha. If the pie becomes bigger, then a dynamic and diverse research market is eminently possible, but without gains that filter through to end investors this may accelerate the (originally MiFID II-driven) market concentration trend. Joint payments may also provide some greater budgeting flexibility in the short term, but in the longer-term AI will need to impact the fundamentals of the investment process in order to fund the required upgrades that everyone seems to be implementing. At the very end of the conference one delegate told us to not forget one thing: “MiFID II softening achieved political buy-in in Europe because leaders wanted to enhance the allocation of capital in the wider economy – great research is at the core of this goal.” It was a timely reminder that may be useful to remember as we all ride the AI rollercoaster this year – see you all on November 4^th in London to take stock!