Every AI project in wealth management gets measured, at least on a slide. The question is whether the measurement is doing real work or producing retroactive justification. Most of what shows up in board decks is the latter — "hours saved per advisor per week" multiplied by "advisor count" times "loaded advisor cost" divided by nothing. The resulting number is defensibly enormous and almost entirely fictional.
Honest ROI measurement for AI investments in wealth is harder and more valuable. It also tends to produce uncomfortable answers, which is why firms keep not doing it.
Why standard productivity math fails
The dominant measurement approach is productivity-based: estimate time saved per task, multiply by task volume, translate to dollars. This is straightforward and mostly wrong. Four reasons.
Time saved is rarely time freed. An advisor saves twenty minutes on meeting prep. They do not work twenty minutes less. They use those twenty minutes on something else — sometimes valuable, sometimes not. The productivity math assumes direct conversion to output. It rarely happens.
Task time estimates are fantasy. Advisors and operations staff give estimates for time savings that do not survive contact with measurement. Self-reported "hours per week" is not data. It is wishful thinking dressed up with a decimal point.
Quality effects are ignored. AI tools change output quality, sometimes better and sometimes worse. Productivity math treats the output as constant. A tool that saves ten minutes but produces slightly worse client communications is not a win.
Adoption is assumed, not measured. ROI projections assume all advisors use the tool as designed. Actual adoption is typically 30–60% meaningful usage in year one. Productivity math multiplied by actual adoption gives very different numbers.
| Measurement type | Validity | Useful for |
|---|---|---|
| Self-reported time savings | Low | Internal politics only |
| Observed task completion time | Medium | Operational tuning |
| Throughput (tasks per period) | Medium-high | Capacity planning |
| Quality-adjusted throughput | High | True productivity measurement |
| Outcome metrics (retention, AUM) | Highest | Business case validation |
What actually counts as an outcome
For most AI investments in wealth, the outcomes that ultimately matter are client-side, not operations-side:
Client retention. Does the tooling improve the advisor's ability to serve clients well enough that fewer clients leave? Attrition rates moved by even 100 basis points against a large book are material. This is where AI ROI shows up at scale.
Share of wallet. Do clients consolidate more assets with the firm because the advisor's planning and service capability is better? This is slower to materialize but larger in dollar impact.
Advisor retention. Good tools attract and retain good advisors. Advisors who leave take books of business with them; advisors who stay grow them. The tooling's effect on recruitment and retention is a legitimate ROI input, though rarely measured.
Capacity per advisor. How many households can an advisor serve well? AI that meaningfully increases this is valuable even if no individual advisor works less. A firm that can support 20% more households per advisor is a materially different business.
Typical improvement in client retention that, applied to a $1B advisor book, generates roughly $10M in retained lifetime value — dwarfing the reported "time savings" from most AI projects.
Building a measurement framework that works
A framework that produces honest answers requires three things in place before the project starts.
Baseline measurement before deployment. Current state of the metrics that matter. Not "about 30 minutes" — actual measured times on actual samples of work. Retention rates by advisor cohort. Share of wallet by client cohort. Without baseline, there is no comparison.
Control groups where possible. Not every advisor gets the tool on day one. A phased rollout with deliberate control and treatment groups allows genuine comparison. Firms that deploy to everyone simultaneously lose the ability to attribute effects to the tool.
A pre-registered evaluation. Define the success metrics and thresholds before the project launches. Not after. Firms that wait to see which metrics moved and retroactively declare those the success metrics are engaging in theater.
What to track in year one
For a typical AI investment in wealth management, a minimum viable ROI dashboard includes:
Adoption metrics. Percentage of intended users actively using the tool. Frequency of use per user. Breadth of use (which features are actually engaged). These are leading indicators — low adoption guarantees low impact regardless of tool quality.
Task-level metrics. Time to complete defined tasks (meeting prep, plan update, client response). Error rates on those tasks. Variance in output quality across advisors.
Client-facing metrics. Response time to client inquiries. Meeting frequency per household. Client-initiated escalations. Net promoter or equivalent where available.
Outcome metrics. Retention. Share of wallet. Advisor-level capacity (households served at acceptable service level).
The adoption and task metrics move within a quarter. The client-facing metrics move within six months. The outcome metrics take 12–24 months to move materially. A firm that kills an AI project in month six because retention has not moved is operating the wrong clock.
For firms building AI governance and measurement practices, the AI operating model capability map maps measurement against adjacent capabilities like model governance, adoption enablement, and vendor management — useful for scoping the operating model required to make AI investments productive rather than performative.