What Does AI for Wealth Management Actually Cost?

Nobody publishes this number, which is why everyone asks for it. The honest answer is that the licence is rarely the expensive part, and the costs that overrun are the ones nobody put in the business case.

Here is the whole shape of it, in the order the money actually leaves.

The five real line items

1. The licence

Usually per seat or per firm, annually. It is the number on the proposal and the one finance focuses on, and it is typically the smallest of the five. A sensible way to sanity-check it: compare it against the fully loaded cost of the operational hours the system is meant to remove. If the licence is not obviously smaller than that, the business case is thin before you start.

2. Integration

This is the one that moves. An assistant is only useful if it can reach your portfolio system, your CRM, your document store and your custodian feeds, and each of those is its own piece of work with its own authentication, formats and failure modes.

The cost scales with the number of systems, not the number of users. A firm connecting two systems and a firm connecting eight pay very different amounts for the same licence. Ask any vendor to price integration per source rather than as a single block, because a single block is where surprises hide.

3. The security review

Rarely budgeted, never free. Someone has to answer where client data goes, who at the vendor can see it, what is logged and for how long, and what happens when the thing is wrong. In a bank that is weeks of internal time across security, compliance and legal, plus whatever your vendor charges for a penetration test or a questionnaire response.

You can shrink it, though not by skipping it. A vendor who arrives with a data-flow diagram, a retention policy and a completed security questionnaire turns a two-month review into a two-week one. Our note on what your compliance team will ask is the list to pre-empt.

4. Your own people

The largest hidden cost in most deployments. Someone defines what good looks like, tests the answers, decides which questions matter, and tells colleagues why the thing exists. That is typically a part-time commitment from an operations lead and a senior advisor for the first quarter.

Firms that skip this do not save the money. They spend it later, in a rollout nobody adopts.

5. Running it

Models get deprecated, your source systems get upgraded, and answer quality drifts. Someone owns that permanently. With a bought product, most of it sits with the vendor and shows up in the renewal. If you build, it is a standing engineering commitment — which is the real conclusion of any honest build versus buy comparison.

What drives the number up or down

How many systems you connect. The single biggest multiplier. Two sources is a project; eight is a programme.

Where it has to run. A deployment inside your own environment costs more to set up than a shared one and is non-negotiable at many private banks. Decide this before you shortlist, because it changes who can even bid.

How clean your data is. If the client record lives in three systems that disagree, somebody reconciles them before anything intelligent happens. That work is worth doing regardless, but it belongs in this budget rather than appearing as a surprise.

How many people actually use it. Per-seat pricing rewards starting narrow. A desk of six proving the value beats a firm-wide rollout nobody asked for.

What it is worth, stated carefully

The return is easiest to defend where the work is mechanical and measurable: hours spent matching, assembling and re-keying. Those are countable before and after.

It is hardest to defend where it is most often claimed. "Better client conversations" is real and nobody can price it, so leave it out of the business case and let it be the upside. A case built on measurable operational time survives scrutiny; one built on advisor satisfaction does not.

The one genuinely large number is error cost — a figure sent to a client that did not match the custodian statement. Most firms know what that has cost them at least once.

A way to scope it before you ask for a price

Write down the five questions you most want the system to answer. For each, list the systems it touches. Count the distinct systems across all five.

That count is your cost driver, and it lets you have a specific conversation rather than a brochure one. It also tends to reveal that two of the five questions need one system and three need six — which is how a sensible phase one defines itself.

Common questions

Is it cheaper to build? Almost never, once connectivity, compliance engineering and permanent maintenance are counted. The prototype is cheap; the production system is not.

What is the typical first-year split? Expect integration and internal time together to exceed the licence comfortably. If a proposal implies otherwise, ask what it assumes about your systems.

Can we start small? Yes, and you should. One desk, two systems, a defined set of questions. It produces the evidence that funds the rest.

What is the cost of waiting? Not zero. It is paid in the operational hours that continue, and in the ungoverned tools people already use while the formal project waits on a review.

Get a real number

A price that means anything depends on your systems, your deployment constraint and the questions you want answered. Book a demo with that list of five questions and we will scope it against them.