Agentic AI investment emails for a private bank

It all started when

Relationship managers (RMs) at the private bank were the primary channel for reaching high-net-worth clients — but the content they sent was unstructured and inconsistent. Bulk emails largely went unopened, and engagement stayed low. The bank needed content that was genuinely personalised and timely, without adding hours of manual work per RM.

What Monica built

A client-communication platform that generated personalised content — trade-idea emails and investment pitchbooks — triggered by contextual nudges such as a large cash inflow, a P&L move over 10%, or an investment nearing maturity. Monica led end-to-end delivery across design, data science, and engineering.

• An agentic AI assistant RMs could prompt directly to generate and send content on demand.
• RAG with Pinecone as the vector database, grounding every content block in source documents (CIO notes, market data), with a cleanup workflow that unified the generated pieces into one polished email.
• A two-step review pipeline: an AI agent first checked financial accuracy against source docs, then the RM signed off as human-in-the-loop — with a long-term path to shift verification fully to AI.
• Owned prompt design — shaping tone and content quality to a standard appropriate for high-net-worth clients.
• Designed the human-in-the-loop evaluation workflow so output met the compliance and quality bar before anything reached a client.
• Solved the cost-at-scale problem — per-email cost grew exponentially with volume and iterative prompting — by decoupling generation: the CIO-derived body was generated once per content refresh and only high-impact personalised elements (like the subject line) per recipient, cutting cost 80%+ and making the feature viable at scale.

Key Stakeholders collaborated with

Client-facing, high-stakes AI output meant three senior functions had to stay aligned, each moving on a different rhythm:

• Weekly standups with the Head of Risk to walk through prompt outputs and confirm regulatory alignment.
• Weekly standups with the Head of Marketing to align email design and disclaimers with bank policy.
• Bi-weekly check-ins with the Chief Investment Officer and his direct reports to validate financial accuracy and tone.
• Process-discovery workshops with RMs up front, then a progressive, feedback-driven build cadence rather than standard sprint demos.

Impact

• 10% revenue increase attributable to the feature.
• 6× increase in conversions to RM conversations.
• 20% productivity uplift for RMs.
• A behaviour shift from heavy editing of AI drafts to minor tweaks — and the RMs who were initially most sceptical became the strongest internal advocates.

Major Takeaways

• Enterprise AI is constrained by more than the model — economics, regulation, and user behaviour all shape what's actually shippable in production.
• Client-facing AI demands a higher bar than internal tooling: a weak investment email doesn't just get ignored, it erodes client trust.
• Managing Risk, Marketing, and the CIO office simultaneously meant tailoring cadence and communication to each function's priorities.
• The real skill was balancing client requirements, unit economics, compliance, and adoption at once — not optimising any single dimension in isolation.