From mortgage servicing compliance to customer retention at scale, I've led product teams at Amazon, USAA, LoanCare, and ServiceMac—delivering measurable business impact through AI/ML innovation.
Content creation for fintech marketing teams was painfully slow—weeks to produce product explainers, compliance-approved messaging, and customer education materials. Generic AI tools couldn't handle the regulatory requirements and domain expertise needed for financial services content.
Built an AI content-generation platform using RAG pipelines and vector search specifically trained on financial services knowledge. The system combines LLMs with compliance guardrails, ensuring generated content meets regulatory standards while maintaining brand voice.
Delivered beta to 50 design partners with 92% weekly retention. Automated prompt-authoring workflows cut content creation time by 75%, saving clients an estimated $1.2M annually. Currently preparing for July 2025 public launch with experiment-driven pricing strategy.
Generic AI tools fail in regulated industries. Success requires domain-specific training, compliance guardrails, and workflows that match how regulated teams actually work—not how Silicon Valley thinks they should work.
LoanCare's mortgage servicing platform was built in the 2000s. Call centers handled payoff requests identically—whether the customer was a serious buyer or just curious about their balance. The 15-20% conversion rate meant massive wasted effort, and retention strategies were reactive rather than predictive.
Led ML modernization across two major initiatives:
Lifted call center conversions 18 percentage points (from ~17% to 35% for high-intent segment). Reduced average handle time 22% by routing low-intent calls to automation. Boosted CSAT scores 9 points because customers got appropriate attention based on their actual needs. Position concluded upon completion of fixed-term transformation program.
Optimizing for average performance across all customers is a trap. Better to spend more time with high-value customers and less with low-value ones. Smart routing isn't about efficiency—it's about allocating human expertise where it matters most.
Prime Day was using demographic segments from 2019. Customer behavior had shifted dramatically post-pandemic, but targeting still treated all "millennials" the same. Conversion rates were declining year-over-year despite increasing ad spend. The marketing team needed better segments but couldn't wait 6 months for a traditional segmentation study.
Co-designed an AI segmentation engine using behavioral clustering (K-means + DBSCAN) that created 87 micro-segments based on actual shopping patterns, not demographics. Built propensity models to predict which offers would resonate with each segment. Integrated with existing campaign management tools so marketers could deploy personalized offers without needing data science support.
Also launched automated A/B testing service supporting 200 concurrent experiments with sub-2% failure rate, enabling rapid experimentation across all Prime campaigns.
Generated $38M in incremental revenue during Prime Day. Improved offer conversion rates by 34%. The segmentation system became the foundation for all Prime marketing campaigns going forward, processing millions of customer scores daily with sub-100ms latency.
Behavioral data beats demographics every time. A 65-year-old retiree and a 28-year-old professional might shop identically—age tells you nothing about propensity to buy. Also learned that segmentation without action is just trivia. The real value came from operationalizing it so marketers could act on insights in real-time.
USAA was losing members to competitors but only realized it when they called to close accounts—far too late to save them. Customer service tracked satisfaction scores, but those didn't predict churn until customers had already mentally checked out. The company needed an early warning system that could identify at-risk members 60-90 days before they'd typically leave.
Implemented ML churn model using random forest and gradient boosting that created a "member health score" based on behavioral signals: product usage changes, support interaction sentiment, digital engagement patterns, and life event triggers. The model scored 5M members daily, flagging those whose behavior indicated quiet dissatisfaction—not just those who were loud about leaving.
Built intervention playbooks tailored to different member segments: proactive rate adjustments for rate-sensitive members, enhanced digital tools for tech-savvy users, dedicated relationship managers for high-value accounts.
Also launched digital financial-readiness tools reaching 5M members, raising monthly active usage 40%.
Saved $12M in the first year through prevented churn. The model identified at-risk members with 78% accuracy 60+ days before they'd typically leave—giving relationship managers time to intervene meaningfully. Became a corporate best practice and was expanded to other product lines.
Prediction accuracy doesn't matter if you can't change the outcome. Our first model had 92% accuracy but only saved $2M because we were catching people too late. The breakthrough came when we shifted focus from "who will leave" to "who's quietly unhappy"—catching them early enough that interventions actually worked.
Mortgage servicers were paying billions in compliance penalties not because they were ignoring regulations, but because their systems couldn't keep pace with regulatory changes. By the time a new CFPB guideline was coded, tested, and deployed (6-12 weeks), two more updates had dropped. Quarterly compliance audits found violations after thousands of transactions had already been processed incorrectly.
Led development and operationalization of Control Tower/Sentry360™—an AI-driven compliance monitoring and analytics SaaS platform that fundamentally changed how servicers approach compliance.
Key innovations:
Collaborated with engineering teams using Git-based repositories within Azure DevOps to manage version control, CI/CD, and feature releases.
Drove 5.2x ROI through launch of the platform. Improved client satisfaction scores and processing efficiency by catching violations at 10 transactions instead of 10,000. Reduced time-to-compliance from 6-8 weeks to under 48 hours for rule changes. The platform processed millions of transactions daily across multiple servicers, becoming the industry standard for compliance monitoring.
Compliance teams don't want more alerts—they want fewer problems. The shift from "check everything quarterly and fix what's broken" to "monitor constantly and never let it break" was transformational. Also learned that configurability beats customization: let users adjust parameters themselves rather than waiting for engineering to hard-code every change.
Prior to these roles, I held engineering and product positions in fintech startups, building eCommerce platforms and digital marketing systems for financial services companies. This included product management work at USAA (2000-2009) leading $150MM sports marketing partnerships and developing marketing performance dashboards that enhanced decision-making efficiency by 20%.
Full career details available in my resume.
Whether it's compliance violations, customer churn, or low-intent calls—I build systems that catch problems early when they're still fixable, not after the damage is done.
Age, income, and location tell you who someone is. Behavior tells you what they'll do next. I've consistently proven that behavioral models outperform demographic segmentation.
A 95% accurate model that doesn't change behavior is worthless. I measure success by business impact—retained customers, prevented violations, saved costs—not model performance metrics.
Regulations change. Customer behavior evolves. Markets shift. I build systems that learn and adapt continuously, not static solutions that become obsolete in 6 months.
Whether you're building your first AI product or scaling an existing platform, I bring hands-on experience solving hard problems in regulated industries.