Essay · Agentic wealth
Getting off batch
Agents carry three jobs on the back end of an advisory business. Only the first one runs on data that arrived overnight, and it is the only one the industry is about to buy.
I wrote last week that adding AI onto antiquated rails isn't likely to work well. That was an assumption and it was vague. A day later the CEO of a clearing firm made it precise, in public, without being asked.
William Capuzzi runs Apex. Under a post of mine, he wrote this:
LPL makes the same mistake I have seen over and over. Focus tech spend on the front side of the stack. The UI, UX. The CRM. The portfolio construction software.
The challenge is that, at the core of all of this, is a batch process. Creates band aids over band aids.
AI only works on well structured data. Even better when it's real time. Clients will demand real time insight.
He named a mechanism. I had named a mood.
The three jobs
Agents carry three things on the back end of an advisory business.
The first is the workflows and the steps that facilitate the administration. Filing a document, routing an approval, populating a form, moving a case to the next stage.
The second is organizing the information. Sorting what is relevant context from what is actionable, and then sorting the actionable into planning, administrative and investing. This is the work I really don't want.
The third is anticipation. Pushing the work forward the way a team does, so that the expert has what they need to provide high level insight and tailored judgment. That preparation takes a lot of work, and it's work most experts lose interest in.
I can say that from the inside. From 2020 to 2023 I built sales and marketing automations, along with the process of getting the information I needed before I could help someone. It worked and I stopped anyway. I did it a bunch, I became proficient, and my mind didn't want to keep doing it.
What batch means
Batch means data moves on a schedule. Positions, transactions and balances arrive in files, mostly overnight. By the time anything reads them, they describe yesterday.
That is enough for the first job. None of filing, routing, populating or advancing a case needs today's number. Those need the right steps in the right order, and they will run on a file that landed at four in the morning.
The second job needs something the first one doesn't. Deciding what is relevant and what is actionable is a judgment. Automation routes what it is told to route, and it cannot decide what matters.
The third job needs both. Anticipation is noticing that something changed. A held-away balance moved. A distribution posted. A client sold something. Batch data cannot tell an agent that a thing changed. It can only report what was true this morning.
So the first job can run on batch. The other two can't.
What the next three years of spending buys
A firm on batch rails can still automate a great deal. It can take the steps and put them on rails, and it will be able to point at a reduction in clicks and a list of enhancements. However, this type of work does not really require AI, it's just easier and faster for AI than a team of humans.
It also isn't the thing that changes the economics. The headcount in a practice is not spent on clicking. It is spent on the preparation a person does before an expert can be useful, and that is the job that needs to know what changed since yesterday. The expert also needs organized context, and AI is the first technology that can infer what matters rather than relying on automations without intelligence.
The near term looks like this. The industry buys a large amount of workflow automation, calls it AI, and reports the enhancements. The advisor feels a little less friction and the same amount of work.
What the rails are for
Putting wealth management services on AI rails should let you serve more customers in improved ways with less people. The intelligence sits on integrated customer-specific data, does all the work that doesn't require judgment, and organizes and prompts the work that requires expert human judgement. It's time to reduce the number of layers between a client and someone who can actually solve complex financial problems or deliver actionable planning items.
What that looks like in practice is teams of agents by discipline, each with its own context and memory, taught to work together. This isn't designed to replace humans but to reconfigure the work and enhance humans in new roles designed specifically for the future of work. People make sure the steps run. Then the expertise gets separated out, into actionable planning items, investment solutions and sale closing.
We end up being surgeons who come in, strap our gloves on for fifteen to thirty minutes, solve the problem, and go on to the next surgery.
None of that runs on a file that arrives at four in the morning.
What this doesn't fix
The objection is that none of this moves the ceiling, because the ceiling is relationships. An advisor can hold a certain number of families in their head. Take the administration off their desk and you have given them free afternoons.
That is largely right, and it is worth saying rather than arguing around.
The top of that hierarchy — planning, behavior, helping a family decide what the money is for — is the part that still costs an advisor time per client. That's why the 100-client ceiling holds: the work that adds the most is the least automated. Agentic rails change that by taking the mechanical layer off the advisor's desk, not the judgment at the top.
Depending on the size of the firm, that layer is also people, and someone is managing them. Every layer also puts more distance between the expert and the information.
The economics arrive underneath, in the layer whose whole job is preparing the work so an expert can be useful for fifteen to thirty minutes. Agent rails are the first technology that can help reduce the headcount needed to source, onboard and serve customers.
That is a narrow claim, and it is the one an operator will recognize.
The timeline is not set by the firms
I believe that over the next decade the percentage of financial services buyers who expect a digital experience without friction will increase, to the point where it will be a massive competitive disadvantage not to make it easy for them to pick human or AI in the sales cycle and the onboarding process. Done right, those that embrace this will be difficult to compete with, in front of a user who over time will reject the idea of having to do wet signatures more and more.
There is a version of this argument that is about vendors. Every firm in this industry is assembled from parts, and the parts are mostly fine. The core is the question. A firm that spends on the front of the stack and leaves the core on a schedule has bought the first job and told itself it bought all three.
Getting off of batch will determine how much value can be accrued with AI.