At many investment firms, the current AI strategy is to take the investment process that already exists and make each individual step faster. The memo is drafted quicker, financial models are updated with less manual work, calls are summarized automatically, companies are screened more efficiently, and the CRM finally gets populated. These are all useful applications, and some will create meaningful gains in productivity and cost. They also reflect a surprisingly narrow interpretation of what is happening.
I have watched this change from three different seats: seven years inside a crossover fund making alternative data part of the investment process, then at a venture firm where I built the investment operating system from scratch, and over the past two years as an operator running a cross-border goods business on AI infrastructure I built without an engineering team. Across each experience, the lesson has been the same. The largest gains do not come from the tool itself. They come from redesigning the process around what the tool makes possible.
The more important shift is that analysis no longer needs to exist as a series of tasks that a person initiates, assembles, and completes. Once a firm’s data, models, and institutional knowledge are connected, analysis can become a continuous system. New information can update assumptions, trigger new scenarios, challenge existing conclusions, and trace the implications across companies and the portfolio. The process no longer has to restart every time an earnings call, management meeting, expert call, or new piece of data arrives.
This is not simply the same investment process happening faster. It is a different investment process.
The process was built around human capacity
For most of the industry’s history, serious analysis was expensive. It took capable people days or weeks to build a view with enough depth to support an investment decision, and much of the organization was built around that constraint. Analysts covered a finite number of companies because a person could only follow so many names. Research began once an opportunity became important enough to justify the work. Investment committees met on fixed schedules because analysis and synthesis arrived in batches. Memos compressed weeks of thinking into a format that senior investors could absorb quickly. Knowledge lived in individual people because properly capturing, structuring, and connecting all of it was too labor-intensive.
These were not bad processes. They were rational responses to the economics of human analytical capacity. But as those economics change, the processes built around them need to be reconsidered.
To be clear, memos and investment committees do more than ration analysis. They create accountability, coordinate people, challenge assumptions, establish decision rights, and control risk. Those functions remain important, but the function should not be confused with the format. The fact that a firm needs accountability does not mean its knowledge must live inside a static memo. The fact that a firm needs challenge does not mean competing views should first be considered during a weekly meeting. The fact that a firm needs risk controls does not mean those controls should operate only during periodic reviews.
A well-designed system could make many of these functions more rigorous. It could preserve the evidence available at the time, the assumptions behind a recommendation, the data and models used, the challenges raised, the confidence expressed, and any overrides that followed. That record does not replace accountability. The people and institutions with authority over the decision remain accountable for it. But it creates far greater auditability than the polished memo most firms preserve after much of the uncertainty and disagreement has been edited away.
Most firms will avoid reconsidering these structures because doing so is an organizational project rather than a technology project. It is much easier to purchase a copilot than to rethink how research is conducted, how decisions are made, what information is retained, and what the analyst is ultimately there to do. AI will therefore be inserted into each existing box. The same people will produce the same documents for the same meetings, only faster.
This will create efficiency, and efficiency is valuable. But when every competing firm can purchase the same tools, that efficiency will be difficult to sustain as a meaningful advantage.
The workflow is the edge
The obvious objection is that more analysis does not necessarily lead to better decisions. I agree. A poorly designed system could produce an endless amount of noise, false confidence, and increasingly sophisticated ways to justify whatever the firm already believes. The objective is not to maximize the quantity of analysis. It is to improve the process through which information becomes a decision and the result of that decision becomes learning.
That process determines which data enters the firm, how it is evaluated, how assumptions are formed and challenged, which issues receive attention, how conclusions translate into portfolio decisions, how risk is controlled, and how outcomes affect the next decision. AI can dramatically reduce the cost of the analytical work taking place throughout this process, but access to a capable model does not give the firm a differentiated system. The firm still has to determine what it is trying to achieve, which information it trusts, how evidence should be weighted, which constraints should be enforced, what requires escalation, and how the process should improve over time.
Two firms can therefore use the same models and produce entirely different results. One may combine those models with differentiated data, explicit assumptions, independent checks, and a disciplined method for updating its views. Another may use the same technology to create more persuasive versions of its existing conclusions. One system is designed to identify when the firm may be wrong. The other is designed, perhaps inadvertently, to make the firm sound more right.
The underlying technology can be identical while the quality of the workflow is entirely different.
This is why I believe the proprietary workflow will become one of the most important assets inside an investment firm. The workflow is not a memo template or a sequence of automated tasks. It is the full logic through which the organization converts information into action: what the system monitors, how it interprets what it finds, how it tests its own conclusions, how it allocates attention, how it controls risk, and how it learns.
Proprietary does not automatically mean valuable. Many firms already have highly customized processes that are slow, political, and difficult to audit. A bad workflow remains bad regardless of how unique it is. The workflow becomes an advantage when it is built around the firm’s mandate, incorporates differentiated information, makes its assumptions visible, measures its forecasts and confidence over time, and improves as more experience passes through it. A proprietary process without a feedback mechanism is simply custom bureaucracy.
This is also why access to a commercially available model is unlikely to become a durable edge. Whatever the leading model providers release, competing firms will generally be able to purchase access to the same or comparable capabilities. Model selection, evaluation, implementation, and security will remain important, but the model itself will increasingly resemble infrastructure. What cannot be purchased as easily is the system the firm builds around it.
Most firms have storage, not memory
Every investment firm has accumulated information that no other firm possesses: management meetings, expert calls, diligence materials, rejected deals, exited positions, failed theses, internal disagreements, and decisions made under uncertainty. Most of this information is scattered across inboxes, stored in disconnected documents, entered into CRM fields nobody revisits, or left in the memory of the person who experienced it.
Even when the material survives, the reasoning that connects it is usually lost. The expert call is saved, but not the assumption it changed. The investment decision is recorded, but not the competing interpretation. The final recommendation remains, but the dissent and uncertainty disappear. The outcome is known, but the firm can no longer reconstruct what it predicted before that outcome became obvious. This is not institutional memory: it's storage.
Institutional memory exists when evidence is connected to beliefs, beliefs are connected to decisions, and decisions are connected to outcomes. It preserves what the firm knew and believed at the time, rather than reconstructing a cleaner story after the result is visible. It also distinguishes between a sound process that produced a bad result and a poor process that happened to make money. Without that distinction, the firm risks learning the wrong lesson and compounding bias rather than knowledge.
Done properly, this memory allows the firm to learn not only about markets and companies, but about itself. Which sources does it consistently trust too much? Where are its forecasts poorly calibrated? Which types of businesses does it understand unusually well? Which assumptions repeatedly prove incorrect? Which internal disagreements identify real risk, and which are mostly noise?
A firm capable of answering these questions can improve the system supporting every future decision. Its history becomes part of the workflow rather than a collection of artifacts left behind by it. Every management meeting, thesis, forecast, disagreement, and outcome adds context to what the system does next.
That is where the advantage can compound. A competitor can rent the same model. It cannot rent the exact history of what your firm saw, what it believed, what it decided, and what it learned.
What would you build today?
None of this requires the conclusion that human judgment disappears. In many investment strategies, access, trust, negotiation, relationships, context, and responsibility remain fundamental. The narrower point is that human judgment will operate inside a much more capable system. Investors should spend less time recreating context, moving information between disconnected tools, updating static documents, and performing work the system should already be capable of performing. Their time can move toward determining what matters, obtaining information the system cannot reach, interpreting ambiguous evidence, and deciding what deserves the firm’s attention and capital.
The exact division between human and machine will differ by firm, strategy, and asset class. That is a design choice. The mistake is assuming that the current division of labor is permanent, then building an AI strategy whose primary objective is to preserve it.
The most important question for an investment firm is therefore not, “Where can we add AI?” That question assumes the existing process is correct and the technology simply needs to be attached to it.
The more important question is: “What would we build if we were starting today?”
Would research remain a sequence of isolated projects, each initiated by a person and reconstructed from the beginning? Would the firm’s knowledge continue to live inside static documents and individual memories? Would challenge occur primarily in scheduled meetings? Would coverage remain limited by the number of companies a team could manually follow? Would the process preserve the final decision while discarding much of the reasoning that produced it? I suspect very few firms starting from zero would answer yes.
Most firms will purchase AI tools and use them to make the organization they already have more efficient. There is nothing inherently wrong with that, but it is unlikely to produce the most meaningful outcome. The firms that create a durable advantage will redesign the system through which they collect information, develop and challenge their views, make decisions, and learn from the results.
The model will be available to everyone. The firm built around it will not.