This is part 2 of a three-part series from Future Proof. Read part 1: Before AI Can Scale, Wealth Management Has a Data Problem to Solve.
If one theme from Future Proof deserves more attention, it is this: AI in wealth management is already working. The next step is making it work for teams, not just individuals, and that takes AI workflow orchestration.
The progress is real
Panelists were candid about how much has changed in a short time. One executive with more than 30 years in the business described how long it used to take to build an estate plan diagram, run quarterly tax projections, prepare a financial plan, and rebalance portfolios. He now watches colleagues finish work in 15 minutes that once took him three hours.
Another panelist had recently spent time with a group of top performing advisors. They described basic AI capabilities like meeting notes, meeting preparation, and next best action as transformative for their practices.
These are real gains, and most of them come from enterprise AI tools used by one person at a time.
The ceiling: where AI workflow orchestration begins
Kelly Wagman of Cresset and Jordan Campbell of Impruve called this the “single player” phase of AI. An individual advisor working in a chat window can move much faster. But the work that consumes the most time in a firm usually isn’t one person’s work. Client onboarding, account opening, and service requests move across several people and several systems before they are done.
That kind of work needs what they called orchestration. They were careful to strip away the buzzwords. Orchestration simply means defining a workflow step by step:
-
what needs to happen at each step
-
what the output of each step is
-
what business logic applies
-
who owns each step, and how handoffs work
It also means deciding deliberately which steps benefit from AI judgment and which should stay rule based. The goal is not to hand an entire process to a language model. The goal is to give the model a clearly defined role inside a process the firm controls.
Own the harness, rent the intelligence
One panelist gave a sharp warning. The worst move a firm can make, in his view, is to add an AI layer on top of an already convoluted mix of 18 software products and hope it works.
He described two paths forward. A firm can let a large model provider build everything for it, accepting that the provider may learn a great deal about how the firm operates. Or a firm can own what he called the harness: control the execution layer, meaning the workflows, the integrations, and the business logic, while renting the intelligence layer from model providers.
The Cresset and Impruve session reached a similar place. They advised firms to build so that models can be swapped over time, choosing the right model for each task instead of locking into one provider.
Three ways to get there
The speakers outlined three options for firms that have hit the limits of the chat window:
-
Buy. Many booths at the conference positioned themselves as AI operating systems. If a vendor offers a ready made workflow for a firm’s specific pain point, it can be a strong place to start.
-
Extend. Use the AI capabilities built into the CRM, planning, and portfolio management tools the firm already owns.
-
Build. This is the right choice when a workflow reflects how a firm uniquely serves its clients. It lets the firm design for its own client experience instead of the industry average.
They added one important caution about building. AI makes code easier to write, but making a system robust and maintainable still requires people with real engineering experience.
Process comes before automation
There was one more point firms tend to skip. Before a workflow can be automated, the firm has to agree on what the workflow actually is. When different advisors handle the same process in different ways, someone has to standardize it first. That is business work, not technical work, and it shapes everything built afterward. We look at why pilots stall when the workflow itself is never rethought in Why 95% of AI Pilots Never Reach Production.
Building the execution layer
This is where Cheesecake Labs does its work. We help firms map the workflows that matter most, design how AI fits inside them, and build the orchestration and integrations that connect people, systems, and models. We build them to be maintainable and model agnostic, so the firm owns the execution layer and keeps its options open as models improve.
In the final article of this series, we turn to the two issues every COO at Future Proof raised: governance and adoption.
