Efficiency in Investment Research
There's a version of AI-assisted research that everyone in financial services is chasing. The analyst who asks a natural language question and gets an answer grounded in the best available content. The publishing team that generates periodic content without spending half a day on formatting and workflow. The portfolio manager who surfaces relevant analyst views without logging into another portal.
That version exists. Parts of it are already running in production at leading firms. But for most of the industry, the gap between the promise and the reality is still significant. And the reason isn't that the AI isn't good enough. It's that the infrastructure around the AI wasn't built for research.
The Generic AI Problem
The tools that exist for AI-assisted knowledge work are remarkable. But investment research isn't general knowledge work. It operates under constraints that generic AI tools were never designed to respect.
Entitlements matter. Not every analyst at every buy-side firm is entitled to see every piece of sell-side research. A generic AI tool that pulls from a broad corpus has no mechanism to enforce those permissions at the query level.
Attribution matters. When an AI synthesizes an answer from multiple sources, the identity of those sources isn't a nice-to-have. It's how buy-side firms demonstrate the value of their sell-side relationships, how sell-side firms get commercial credit for their work, and how compliance teams maintain audit trails.
The result is odd. The industry has access to some of the most powerful AI tools ever built, and most of them aren't fit for purpose in the research workflow.
The Workflow Problem Nobody Talks About
Analysts are already using AI. That ship has sailed. The more interesting question is where AI sits in the workflow, and for most analysts, the answer is that it sits at the edges. It helps with ideation and drafting in a personal AI environment. But when it's time to publish, they have to leave that environment, navigate into a publishing system, find the right template, and manually transfer everything they just built.
The context switch isn't just friction. It's a signal that the tools aren't actually integrated. The AI is sitting alongside the research workflow, not inside it.
That gap compounds at scale. Publishing teams managing high-volume periods, earnings season, quarterly reviews, periodic client updates, still carry significant manual burden for content that's largely formulaic. The structure is known. The data is available. The bottleneck is the step between having the content and getting it into the system.
What the Buy Side Is Missing
The workflow problem isn't limited to sell-side publishing teams. Buy-side firms face their own version of it.
When AI mediates research consumption, the connection between the original content and the decision it influenced gets harder to trace. Which broker's view actually shaped the portfolio call? Which research relationship is generating the most signal? If the answer lives only in a black-box AI system, the buy-side loses the ability to manage its research relationships strategically.
The firms thinking about this clearly understand they need AI tools that are governed and attributed, not because regulation requires it yet, but because attribution is what makes the investment in research relationships legible.
The Integration That's Been Missing
What the research industry needs from AI isn't more powerful models. The models are already there. What it needs is integration. AI that sits inside the workflows that research teams and their clients actually use, with the governance and attribution layers that make it safe to deploy in a regulated environment.
That means AI that understands entitlements and respects them at the query level. AI that preserves attribution as content moves through synthesis and derivation. AI that fits into existing compliance and approval workflows rather than bypassing them. AI that returns signal to the people and institutions that produced the content, not just to the end user consuming the output.
It also means reducing the friction between where ideas are developed and where research gets published. The analyst who thinks in an AI environment should be able to publish from that environment without starting over. The operations team managing high-volume periods should have programmatic tools that automate what can be automated without sacrificing the governance that can't be.
Efficiency Without Governance Is a Risk, Not a Benefit
The efficiency argument for AI in research is compelling. Less time on formatting. Faster discovery. Better signal from client interactions. More research reaching more people in more useful forms.
But the firms that deploy AI fastest without the right infrastructure aren't ahead. They're exposed. Exposed to entitlement violations they can't see. Attribution loss they can't measure. Compliance questions they can't answer.
The efficiency gains that matter are the ones that come with the governance layer built in. Not as an afterthought. As the foundation that makes the efficiency sustainable.
The infrastructure to make this real is closer than most firms realize. The standards, the schemas, the distribution architecture - much of it already exists. What's being built now is the layer that connects it all and makes AI a native participant in the research workflow rather than a tool that sits alongside it.