Agentifying inbound support and debt collection for a leading Saudi fintech

A fast-growing Saudi investment platform handed its inbound support line and its early-stage collection calls to Sarj voice agents. Within the first month, peak call queues were 70% shorter, and the agents now handle more than 150,000 investor calls a month in the dialects customers actually speak.

The platform serves retail investors across the Kingdom. As its customer base grew, the support line became the place where growth showed up as friction: long holds at market open, repeat calls about the same transfers, and a collections team spending most of its day on reminders rather than on the accounts that needed a conversation.

The challenge

Call volume tracked the market. Every morning at the open, and every time a product launched, the queue climbed past what the support team could staff for. Callers waited, hung up and called again, which made the queue longer still.

Most of those calls were routine: balance and portfolio questions, the status of a deposit or withdrawal, identity checks before an account change. They needed an accurate answer from the platform's own systems, not a person.

The previous IVR could not get callers there. It understood Modern Standard Arabic menus, but most customers speak Najdi or Hijazi Arabic and switch into English for product names. Callers pressed zero to reach an agent, and the automation saved almost nothing.

Collections had the same shape. Early-stage reminders followed a script, but each call still took an agent's time, and the team could not reach every account in the window when a reminder works best.

Any replacement also had to meet the bar of a regulated financial business: customer audio and data hosted in the Kingdom, every action logged, and a deployment aligned with SAMA and NCA requirements from day one.

Why Sarj

The platform evaluated Sarj against general-purpose voice AI and its existing contact-center vendor. Three things decided it:

• Dialect first. Sarj's speech models are built for Saudi dialects and the code-switching that comes with them, so callers could speak the way they normally do.
• Actions, not answers. Sarj agents read from and write to the systems behind the conversation, so a caller can check a transfer, verify their identity and update a record on one call.
• Sovereign by default. Sarj runs on sovereign cloud infrastructure in Saudi Arabia and is deployed to meet SAMA and NCA requirements, which kept the compliance review short.

The solution

Sarj connected to the platform's existing telephony through its low-latency SDK, so calls routed to the agent while the current contact center kept handling everything else. The rollout ran in three stages.

1. Inbound support in dialect

The voice agent took over the most common inbound journeys first: portfolio and balance questions, deposit and withdrawal status, and account changes behind an identity check. Responses stay under 120 milliseconds end to end, so the conversation keeps a natural pace even when a caller interrupts.

2. Live actions in the back office

The agent was connected to the platform's CRM and core systems. It verifies identity, retrieves the transaction a caller is asking about, and logs each call with its outcome, so a human agent who picks up an escalation already has the full context.

3. Early-stage collections

Reminder calls moved to outbound voice agents that follow the platform's approved scripts, confirm a payment date, and send a payment link while the customer is on the call. Accounts that need judgment go straight to the collections team.

Results

Within the first month in production:

• 70% reduction in peak call queue. The morning spike at market open no longer backs up the line.
• 150,000+ investor calls handled each month by the voice agent across support and collections.
• 85% first-contact resolution. Most routine calls are resolved end to end without a transfer.
• Every call logged with its transcript, actions and outcome, and all audio and data kept in the Kingdom.

Conclusion

The platform did not replace its contact center. It gave the routine work to agents that speak its customers' dialects and can act in its own systems, and gave its people the conversations that need them.

The next phase extends the same agents to new-customer onboarding calls and to in-app voice, using the same dialect models and integrations.