Beyond the Hype: The Strengths and Limits of AI in Advice

For most advisors, the ceiling shows up every day — and it rarely announces itself as a capacity problem. It shows up as another hour of meeting prep that could have been thirty minutes. A follow-up that falls through the cracks because three other things needed attention first. A client conversation that gets cut short because the data wasn't ready in time.

These aren't failures of effort. They're the predictable result of building a practice on workflows that were never designed to scale. And for a long time, the answer was just: work harder, hire more, or serve fewer clients than you could.

Reed Colley, Orion's President of Advisor Technology, puts it plainly: AI isn't additive the way past technology waves were. It's a multiplier. Every manual task, every process that requires a human to hold it together, represents a ceiling on what the firm can do. AI removes that ceiling — not by replacing advisors, but by removing a significant portion of what gets in the way of doing the work that matters.

But it has real limits. Understanding both sides honestly is what separates useful adoption from expensive experimentation.

Where AI Delivers Real Value

Meeting Preparation

Before a client meeting, there's a version of prep that takes ninety minutes and a version that takes fifteen. The difference is usually access: how quickly can the advisor pull together what's relevant, flag what's changed, and walk in with a clear picture of where the conversation needs to go?

AI closes that gap. It surfaces household context, summarizes recent interactions, and flags anomalies without the advisor having to go looking. The result isn't just saved time — it's a different quality of attention in the room. Advisors who aren't mentally reconstructing the client's situation during the first ten minutes of a meeting are better at listening during it.

One question worth asking any vendor: is preparation grounded in live firm data, or working from static snapshots? That distinction matters more than most demos will make obvious.

Search and Retrieval

Most advisory firms aren't short on data. They're short on usable data. Information sits in CRM notes, planning files, custodial feeds, and email threads — technically accessible, practically buried. AI lets advisors query that information conversationally and get organized starting points instead of raw results.

This isn't analysis. It's the work that used to happen before analysis could begin, and it's where a surprising amount of advisor time disappears.

Early Risk Detection

There's a version of risk management that's reactive — you find out something went wrong when a client calls. And there's a version that's proactive — you see the signal before it becomes a situation.

AI supports the second version. It can flag concentration risk across households, identify clients whose service patterns suggest they may be disengaging, and surface early indicators that tend to get missed when advisors are managing large books without much support. The value isn't just catching problems sooner. It's the confidence that comes from knowing your book is being monitored in ways that would be impossible to do manually.

Workflow Continuity

This one is less visible than the others, but it may be where AI creates the most durable value. The administrative drag in most advisory firms isn't one large problem — it accumulates through dozens of small ones. A task that requires three system logins. A handoff that depends on someone remembering to do something. A report built manually every month because no one has set up a better way.

When technology is fragmented, Reed observes, the advisor becomes the integration layer — the person manually connecting systems that should be talking to each other. AI embedded into existing workflows addresses that at the source. It doesn't ask advisors to change how they work. It makes the way they already work faster and less effortful.

The Limits of AI in Advisory Work

Fiduciary Judgment

This is the clearest limit, and the most important one to state plainly: AI does not make recommendations. It supports the advisor who does.

Suitability, trade-offs, and the judgment calls that come from knowing a client's full situation — those remain the advisor's responsibility, and they should. AI can organize the information that goes into a decision. It can't make the decision, and it can't be accountable for it. Any tool that blurs that line deserves scrutiny before it gets anywhere near a client relationship.

Client Relationships

Clients don't stay because the advice is efficient. That sounds obvious, but it's easy to lose sight of when you're evaluating AI on the basis of time saved and tasks automated.

Trust is built through consistency, empathy, and the kind of judgment that comes from knowing someone across years and life events — not just accounts and allocations. AI supports the operational work that surrounds those relationships. It doesn't touch the relationships themselves. Firms that keep that distinction clear tend to use AI better, because they're not asking it to do something it was never designed to do.

Oversight and Validation

Here's the risk that doesn't get talked about enough: AI produces confident-sounding output. It's designed to. And that polish can quietly reduce the impulse to check.

Using AI-generated content in a client conversation without validating it first isn't just a compliance concern — it's a trust concern. The advisors who use AI well treat its outputs as a starting point, not a conclusion. They stay curious about where an answer came from. That habit is more important than any guardrail a vendor builds in, because guardrails don't survive every edge case and curious advisors do.

Starting From a Broken Foundation

AI amplifies what's already there — including the problems. That's its strength in a well-run firm and its liability in one that isn't.

Reed is direct about this: data quality is the precondition, not an afterthought. In Orion's research, only about 3% of firms feel their data strategy is fully solidified. That's not a criticism — it's the reality of where most practices are. But it matters enormously for AI, because a system working from fragmented, inconsistent, or poorly permissioned data doesn't produce better insights. It produces confident-sounding ones.

What makes this harder than it looks is that building a data foundation isn't just a technical problem. Getting the right data to the right people — and making sure the wrong people can't access what they shouldn't — turns out to be one of the most complex problems in the space. Firms that underestimate it tend to find out mid-implementation.

The practical implication: before evaluating what AI can do, it's worth asking honestly what your data can support. AI doesn't fix a broken foundation. It shows you, faster than anything else would, exactly where the cracks are.

The Right Frame: AI as Co-Pilot

The co-pilot model is the most useful way to think about this — not because it's a catchy metaphor, but because it accurately describes the division of labor. AI handles the background work: preparation, retrieval, pattern detection, administrative continuity. The advisor handles interpretation, relationships, and every decision that carries real weight.

What that frame does, practically, is shift the question. Instead of asking "what can AI do?" — which is a product question — firms start asking "what should AI do here?" That's a strategy question, and it leads to much better adoption decisions.

The firms getting real value from AI right now aren't the ones that moved fastest. They're the ones that were clearest about what they were asking it to do.