JPMorgan / Wealth Analytics
Portfolio analytics built around the kind of work the Asset & Wealth Management team would actually run. AUM concentration by tier, risk-adjusted returns, advisor productivity scoring, and a churn risk model on synthetic client accounts hitting 87% classification accuracy on holdout.
RoleSelf-directed project
Duration2 weeks
Year2026
StatusShipped
The brief.
JPMorgan's Asset & Wealth Management group runs analytics across $4T+ in client assets. I built this project to push my own thinking on how to connect portfolio-level metrics to advisor behavior and surface risk before it materializes — the kind of work I want to be doing day-to-day in a wealth analytics team.
The work.
AUM_CONCENTRATION / Q1LIVE
UHNW retention
94.2%
↑ 2.1pp YoY
Sharpe (top 10%)
1.87
↑ 0.12 vs index
Churn risk flagged
$24.4M
12 accounts
How I approached it.
- Modeled 25 synthetic client portfolios across ultra-HNW, HNW, and mass-affluent tiers with realistic AUM distributions and asset class mixes.
- Built risk-adjusted return benchmarks using Sharpe and Sortino, with index comparisons by tier.
- Trained a churn risk classifier in scikit-learn (logistic regression baseline, random forest production) reaching 87% accuracy on holdout.
- Surfaced the strategic insight automatically: a 1% improvement in UHNW retention is worth ~$24.4M in preserved AUM at the modeled concentration.
The outcomes.
- Reusable as a template for any wealth or portfolio analytics workflow.
- Proved the financial domain modeling end-to-end, beyond pure BI dashboarding.
- Surfaces concrete dollar figures from abstract risk metrics — the way a real advisory team would consume the data.
The stack.
Power BI
Stakeholder-facing dashboard layer with DAX measures
Python
Data generation, churn model training, evaluation
scikit-learn
Logistic regression and random forest classifiers
pandas
Portfolio time series and tier aggregations