Design a North agent that helps relationship managers at a Canadian bank prepare for client meetings.

Introduction

Cohere's North is an agentic AI workspace that enterprises run inside their own cloud or data centre. It connects to internal systems such as email, CRM, document stores and data warehouses, and lets employees chat with, search across and automate work on top of company data, using Cohere's Command, Embed and Rerank models underneath. Banks are among Cohere's most important customers because they care about exactly what North promises: private deployment, access controls and citations back to source.

The prompt asks us to design a North agent that helps relationship managers (RMs) at a large Canadian bank prepare for client meetings. An RM in commercial or wealth banking may own 80 to 200 client relationships. Before each meeting they pull together account history, recent transactions, open service tickets, product holdings, news about the client's business and the bank's current offers. Today that preparation is manual, spread across six or seven systems, and often skipped. I'll clarify the problem, explain why Cohere should build this, pick a user segment, find the sharpest pain points, and then design and measure a solution.

Step 1

Clarify Questions (3 minutes)

• Which RMs are we designing for: retail wealth, commercial banking, or both?

• Why it matters: A commercial RM covering mid-market companies needs financials and credit exposure; a wealth RM needs portfolio and life-event context.
• Hypothetical answer: Commercial banking RMs serving companies with CAD 5M to 100M in revenue.
• Impact: We'll design around credit, cash management and company news rather than investment portfolios.

• Is North already deployed at the bank, and which systems are connected?

• Why it matters: An agent is only as good as the data it can reach.
• Hypothetical answer: North is live in the bank's private cloud with SharePoint, Outlook and the CRM connected. Core banking data is available through a read-only API.
• Impact: We can assume retrieval over documents and CRM, but need a connector plan for transactional data.

• What constraints does compliance place on generated content?

• Why it matters: In regulated banking, anything shown to a client may count as advice or a marketing communication.
• Hypothetical answer: Outputs are internal only, must cite sources, and cannot recommend specific products without a human decision.
• Impact: The agent prepares and suggests, but the RM stays the decision-maker. Citations are mandatory.

• What does success look like for the bank sponsor?

• Why it matters: It decides the North Star of the design.
• Hypothetical answer: More client meetings per RM per week and higher product penetration per client, without new compliance incidents.
• Impact: We'll measure time saved and meeting quality, not just usage.

Propose the Goal

Help commercial RMs walk into every client meeting fully prepared in under ten minutes, using a North agent that assembles a cited, accurate client brief from the bank's own systems, so the bank deepens relationships and Cohere proves North's value in a flagship financial-services account.

Define the Scope

In scope: pre-meeting preparation for scheduled meetings, a cited client brief, suggested talking points, and follow-up notes after the meeting. Out of scope: client-facing chat, automated product recommendations sent to clients, and credit decisioning.