FNB Zambia achieves 80% automation in handling customer queries

Automating repetitive queries at scale

The majority of queries reaching FNB Zambia's Facebook channel were routine – account information, branch details, product FAQs, and service procedures. Each required an agent to respond individually, a model that worked at lower volumes but became a bottleneck as digital engagement grew.

Proto trained the AI agent on the bank's existing query history from its social channels, building a strong baseline for intent recognition. The agent was then refined continuously using real-time conversation data, improving accuracy as new query patterns emerged.

Within the first week of going live, the AI agent handled 4,366 queries – a 44% increase in automated self-serve resolution. That proportion continued to grow, eventually reaching an 80% automation rate across all incoming interactions. The bank now handles an average of more than 13,000 interactions per month, with agents concentrating on the complex, sensitive cases that genuinely require human judgement.

24/7 coverage outside office hours

Before deployment, customers reaching out outside standard working hours had no path to resolution until the following business day. The AI agent changed this – operating continuously and resolving queries at any hour regardless of agent availability.

43% of all interactions are handled outside standard working hours, representing a significant share of the bank's monthly volume that would previously have gone unanswered until the next business day.

Live agent escalation – with full context

Not every query should be resolved automatically. For complex account matters, complaints, and cases governed by bank policy, the AI agent escalates to the contact centre through Proto's unified inbox. Agents receive the full conversation history before responding, removing the need for customers to repeat themselves.

For queries arriving outside office hours, designated standby agents are notified by email, ensuring urgent cases receive a timely response without a full team on call. The design – automation handling the predictable, agents handling the exceptional – allowed a small team to maintain the bank's service standards at a volume that would previously have required significantly more resource.

Security for banking on social

Running an AI agent on a public social channel for a regulated financial institution carries specific data protection obligations. For FNB Zambia's deployment on Facebook Messenger and webchat, Proto's platform provided three controls directly relevant to the bank's context.

All data exchanged between customers and the AI agent is encrypted in transit via TLS/SSL – covering both the Messenger channel and webchat, and ensuring that conversation content cannot be intercepted between the customer and Proto's systems.

Access to conversation histories and customer data within the platform is governed by role-based permissions. Only authorised agents and administrators could view or act on customer conversations, with access scoped to each user's role — preventing exposure of sensitive information to the broader team.

Proto's platform is independently certified to SOC 2 and ISO 27001 standards. For a bank deploying AI on a consumer-facing social channel, these certifications provided assurance that the underlying infrastructure met the security and data handling requirements expected of a regulated financial services partner.