← Selected Work
Corporate Bank · USABanking & FinancialBuild · MVP2023

An AI co-pilot for corporate banking relationship managers.

A multi-agent assistant beside every relationship manager — surfacing cross-sell, retention and risk, supporting live calls in real time, and drafting proposals in the RM's own voice.

RM live-call assistant
Client
Private · $7B rev · 19k staff
Duration
6 months (agent phase)
Scope
MVP
Stack
OpenAI · LangGraph
Architecture
Multi-agent (3)
Region
USA
The challenge

Relationship managers for corporate banking customers were juggling cross-sell, retention and risk across large books of business — with the signals they needed scattered across systems. Live client calls were hard to support in the moment, and proposals took days to draft.

Our approach

We designed a multi-agent assistant that works alongside the RM rather than replacing them: one agent for performance and customer-activity insight, one for live-call support, and one for proposal drafting — all grounded in the bank's own data.

What we built

AI insights agent — tracks the RM's performance, KPIs and customer activity. Live call assist — analyses live transcription, surfaces the right product and information mid-call, runs sentiment analysis, and tracks opening questions. Proposal draft assistant — auto-drafts an email-format proposal from the conversation, in a built-in AI editor.

Tech stack
Models & agents
OpenAILangGraphMulti-agent orchestration
Capabilities
Live transcription analysisSentiment analysisRAG over product dataAI document editor
Application
Web appReal-time call UI
What we delivered
  • A 3-agent RM assistant (insights · live-call · proposal)
  • Real-time live-call support with relevant-product surfacing
  • Call sentiment analysis & opening-question tracking
  • An AI proposal editor that drafts from the conversation
  • Delivered as an MVP
Timeline
Phase 1
Discovery & agent design
6 months
AI agent phase — build to MVP
MVP
In RMs' hands for live calls
At a glance
3
Coordinated AI agents
Live
In-call support & sentiment
6mo
To working MVP
Outcomes
3
RM agents in one assistant
Live
Call support & drafting
1
Co-pilot, not a replacement
Next case →AI knowledge retrieval for a customer support centre.

Have something
that needs to ship?

Tell us the problem. We'll tell you honestly what makes sense — even if that's not us.

Start a conversation