How a GCC Telco Cut Misrouted Calls by Fixing Arabic IVR Speech Recognition

A GCC telecom operator reduced IVR intent fallback rates and misrouted calls by replacing generic Arabic speech recognition with Munsit's Gulf dialect speech-to-text. The deployment improved call routing accuracy, lowered agent workload, enhanced customer satisfaction, and unlocked valuable transcript-based operational insights.

The Challenge

IVR systems are the first point of contact in any telco's customer operation. When they work, they route callers quickly and reduce pressure on human agents. When they fail to understand what a customer is saying, callers repeat themselves, get sent to the wrong queue, or end up with a live agent anyway, creating the exact cost and friction the IVR was meant to eliminate.

A GCC telecom operator had built its Arabic IVR on a third-party ASR platform that worked reasonably well for the fixed phrases it was trained on. But as the company's product range grew and customer questions became more varied, recognition quality started slipping. Callers were speaking in natural language, using Gulf dialect, mentioning specific product names, and mixing Arabic and English in the same sentence, none of which the underlying model handled reliably.

The operations team tracked what they called an "intent fallback rate": the share of Arabic calls where the IVR couldn't determine what the caller wanted and sent them to a general agent queue. Over two years, that number had been climbing. By the time a replacement evaluation started, roughly one in four Arabic calls was bypassing the IVR entirely. Each of those calls required a live agent to start from scratch, adding to handle time and hurting first-call resolution rates

Evaluation: Testing Munsit STT Against the Existing System

The telco's technical team ran a six-week parallel evaluation. A portion of incoming Arabic calls was routed through Munsit STT while the existing Arabic ASR handled the rest. Both streams fed into the same intent classification layer, and outcomes were measured against human-labeled transcripts from the same call sample.

Munsit STT delivered higher intent detection accuracy across the board. The biggest improvement came on Arabic dialect natural language inputs, the exact pattern that had been causing the most fallback failures. The evaluation also tested calls where customers switched between Arabic and English mid-sentence, typically when naming a product or describing a technical issue. Munsit handled these code-switching calls more reliably, maintaining recognition on the Arabic portions even when English terms appeared inline.

Implementation: Replacing the ASR Layer

Following the evaluation, the telco began a phased rollout, starting with its busiest service line: prepaid plan management. This line had the highest call volume and the worst fallback rate, making it the highest-impact place to start.

The Munsit Speech-to-text API was integrated directly into the IVR platform, replacing the recognition layer while leaving everything else, intent classification, routing logic, and agent queues untouched. The integration ran through the IVR vendor's API layer, with CNTXT AI's technical team supporting the process over three weeks. The main items requiring calibration were the audio encoding format and the streaming configuration. Once those were resolved, the integration was stable.

What Changed

Fewer misrouted calls, better customer experience, and a new data source from a single infrastructure change.

After deployment, the intent fallback rate on the prepaid line dropped noticeably. Calls that had been landing in the general queue were now being correctly identified and sent to the right self-service path or specialist queue. Average handle time for agent-assisted calls from this line also fell because agents were receiving better-qualified calls rather than unrouted general inquiries.

Post-call satisfaction scores for the IVR experience improved on the affected service line. In survey responses, the most common theme was simply that the system understood callers, something that had been a recurring complaint before.

There was also a benefit the operations team hadn't fully anticipated:

• Structured transcripts as a data asset. The previous system produced only binary routing decisions with no text record. Munsit produced full transcripts, giving the operations team a new way to analyze what customers were actually asking, including emerging inquiry types, product confusion patterns, and seasonal volume shifts.

The telco is currently running the same evaluation on two additional service lines, with a roadmap for full IVR migration over the following two quarters.

Result

Arabic IVR performance has a ceiling set by the quality of the ASR underneath it. No amount of tuning the routing logic or intent classification helps if the system can't accurately transcribe what callers say in Gulf dialect or across code-switched speech.

Replacing the ASR layer with Munsit STT is a targeted infrastructure change; it doesn't require rebuilding the IVR. The results are measurable from deployment and extend further as the change rolls out across more service lines.