AI made the call. Now explain it to your client.

A client calls Monday morning after reading the quarterly statement. She wants to know why the portfolio holds a healthcare name she doesn't recognize, and why a technology position she does recognize was trimmed in September. As someone who spent the first part of their career in wealth management, this conversation has a familiar shape. Talking about sector views, a fund manager's thesis, where we see valuations. The answers, while never perfect and often debatable, were always available.

Those conversations are changing. The decision to buy the healthcare name came from a model running calculations through a few thousand data points a second. The model was right, or wrong, for reasons the client is entitled to hear. The old answer, built around human conviction, doesn't quite fit the new question.

AI is already on the shelf

The reason this matters now, and not in some vague future, is that AI investing has stopped being something advisors opt into. It is often inside products advisors already distribute. Some separately managed accounts now use machine learning to select securities. Model portfolios rotate based on signals with no human in the loop. Global banks have rolled out AI-driven indices across asset classes, and a growing share of thematic ETFs rebalance on algorithmic schedules.

The practical consequence for advisors is that an increasing share of any book of business now contains positions whose rationale sits inside a model rather than in a manager's commentary. When a fund manager underperformed in the past, you could quote their commentary and walk the client through what changed in the thesis. A decision made by a model leaves the advisor reaching for language that doesn't exist yet. "The algorithm flagged it" is an honest answer, and it is the kind of answer a client remembers for the wrong reasons.

The black box gets a window

Until recently, a knock on AI investment models was that they were black boxes. A system might produce reliable returns and still offer no intuition for why it favored one stock over another. The accuracy was there; the reasoning was not. For an advisor recommending the strategy, that was a structural problem.

The last few years have changed this. The broader AI research community has built what's called explainable AI, a set of techniques that open the box without breaking the model. Systems still run their complex calculations. They can also now report which factors drove a given decision and roughly how much each factor contributed.

Think of it the way you'd think of a thoughtful analyst. Ask a good analyst why they like a stock and you'll get structured reasoning: earnings quality is improving, valuation is reasonable, sentiment has turned, and there's some concern about how much the stock has already run. Modern AI systems can now produce the same shape of answer. The underlying math carries a name, SHAP, borrowed from game theory. Advisors don't need to carry the math themselves. What matters is knowing the capability exists and what it lets them say in a meeting.

Three conversations, reframed

Here is what that changes in practice. Start with the drawdown conversation. The portfolio has dipped over a quarter and the client wants to know what happened. The advisor can say something like: "The model was positioned for rates to stay higher for longer. That call was wrong. When the data shifted in August, the system repositioned into rate-sensitive names, and it has been recovering ground since." That is a conversation about investment thinking.

Next, the concentration conversation. The client sees that a single technology name is over 5% of the portfolio and asks whether the position is too big. The advisor can walk through what the model is weighting: the name's earnings trajectory, its valuation against peers, how it sits against the rest of the book, and which factors would have to change for the position to come down. The conversation moves from "the model said so" to a thesis the client can push back on, which is the point.

Finally, the rotation conversation. A client has noticed the portfolio shifted into healthcare last month and asks why. The advisor can describe what triggered the shift: a shifting regulatory picture, an improvement in sector earnings revisions, a sentiment read that had been negative for most of the year. The language is still the advisor's own, and the thinking is still investment thinking. The difference is that both can now be drawn from a model that, a few years ago, would have refused to show its work.

Where the explanation runs out of road

This capability is an advance, and it is worth being direct about where it runs out of road. Explainability tells you which factors the model leaned on for a given decision at a given time. Whether those factors will keep working is a separate question, and the explanation can't answer it.

Models are trained on data that already exists. When markets do something they haven't done before, the model may still produce an explanation that looks coherent while the underlying reasoning has quietly gone stale. Seeing the reasoning is not the same as knowing the reasoning will hold.

The practical implication is familiar to any advisor who has ever overruled a fund manager. Judgment still matters. When the explanation for a position doesn't square with what the advisor is seeing in the market, the advisor should say so and push back. Treating the model as infallible because it can now show its work is a new version of an old mistake.

Questions to ask before you distribute

For advisors being pitched AI-driven strategies, this leaves a short list of useful questions. The answers reveal more about a product than most glossy decks do.

First, if a client asks why we own this name, what can you give me to help answer that? The right answer is a live attribution report, updatable on demand, that shows what drove the position. A promise to send marketing materials doesn't cut it.

Second, when performance disappoints, can you show me what the model was thinking and what it was repositioned into? A product that can only explain itself when things are going well is not worth distributing.

Third, what are the conditions under which this model is likely to struggle, and how do you monitor for them? Anyone who can't answer that question hasn't thought hard enough about their own product.

The advisors who thrive in an AI-driven market will be the ones their clients trust to speak for the machine.

About the Author

Art Amador is Co-Founder and President of QuantumStreet AI, an AI-driven investment platform with $8 billion deployed globally. He previously co-founded EquBot.