Kenya Bank: Microsoft Copilot Chat & Voice Agent (IVR)
Banking & Financial Services
Copilot Developer — solution owner, design through deployment
Microsoft Copilot Studio, Power Automate, Azure AI Services, Power BI
Dataverse, SharePoint and Website Knowledge Base, Core Banking REST APIs
30-Second Pitch
“A bank's support line was 100% human and closed after hours. I built a Copilot Voice and Chat agent that plugs directly into their core banking APIs, so customers can check a balance or raise a request without waiting for a human — and leadership gets a live Power BI view of how well it's working.”
The Situation
A Kenyan bank's tier-1 customer support ran entirely through a legacy IVR menu tree and human chat agents. Customers waited on hold for routine account questions, the desk was closed outside business hours, and every simple balance check or transaction lookup consumed an agent's time that could have gone to complex cases.
The Challenge
The bank needed a self-service channel that could handle real account-specific queries — not a scripted FAQ bot — securely, over both voice and chat, without adding contact-centre headcount. It also had to plug into live core banking systems, so answers reflected real account data, and it needed to fit inside the bank's existing supervisor and SLA workflows rather than bypass them.
My Approach
I owned the solution end-to-end and split delivery into two coordinated tracks — a Voice/IVR agent and a Chat agent — built on a shared Copilot Studio foundation so both channels stayed consistent in tone, escalation logic, and data access.
Designed natural-language intent routing for the Voice Agent (IVR) so callers could state what they needed in their own words instead of navigating a phone-tree, with identity verification built in before any account data was surfaced.
Built the Chat Agent using Generative AI Topics and multi-turn conversation flows, grounded in a SharePoint Knowledge Base so answers stayed accurate to the bank's actual policies and products.
Designed a clean handoff to human agents for anything the bot couldn't resolve, carrying full conversation context across so customers never had to repeat themselves.
Connected both agents to Dataverse and the bank's Core Banking REST APIs via Power Automate, so balance checks, transaction lookups, and service-request creation happen live, inside the conversation.
Configured Copilot Service Admin Center and Service Hub — channel assignment across voice and chat, SLA rules, and supervisor dashboards — across 3 enterprise contact-centre deployments, so the new agents slotted into the bank's existing operational workflow rather than working around it.
Layered Power BI reporting on top of Copilot's native analytics, giving leadership a single view of resolution rate, escalation volume, and CSAT trend across both channels.
Result & Impact
Targeted 40% deflection of tier-1 support calls away from human agents.
Delivered 24/7 self-service account resolution with no additional headcount.
Gave contact-centre leadership live visibility into agent performance for the first time, via the connected Power BI layer.
Proved out a repeatable Admin Center / Service Hub configuration pattern, since reused across 3 deployments.