Citi · Private Bank · 2020–21
KYC AI automation
Associates at Citi Private Bank manually reviewed a prospective client's background and ongoing activity as part of KYC — redundant, effort-heavy work that was also error-prone, with real risk exposure if a client's risk rating was assessed incorrectly. Monica built an AI bot that automated the background review and cut manual effort by 70%.
It all started when
As part of KYC, associates manually reviewed a potential client's background and ongoing activity. The work was redundant and effort-heavy, and — because it was manual — prone to human error. A wrongly assessed risk rating created real risk exposure for the bank, so the stakes on each review were high.
What Monica built
An AI bot that automated the background review by combining several techniques into one pipeline:
- Web scraping to gather client-related information from across the web.
- Sentiment analysis on the gathered content.
- Clustering and classification algorithms to predict a client's risk rating.
Crucially, Monica integrated the model into associates' existing workflow via API calls, so it slotted into their current process rather than forcing them to learn a new tool.
Monica designed and built the automation end-to-end, balancing two constraints at once: mitigating the bank's risk exposure, and keeping the tool seamless and easy for associates to adopt.
Impact
- Reduced manual banker/associate review effort by 70%.
Major Takeaways
- Designing AI for high-stakes, compliance-sensitive processes is different — the cost of a wrong prediction (a mis-rated client) is far higher than in typical product contexts.
- Adoption depends as much on ease of integration into existing workflows (via API) as on model accuracy — a strong model that disrupts the process risks low usage.
- A single KYC automation problem spans the full stack — data collection (scraping), NLP/sentiment analysis, and predictive classification.