Most of us harbor foreign accent bias. It’s human nature. Linquistic intonations that don’t match our own can trigger a rush of emotions: Is the contact center associate qualified? Truthful? Will this take a long time? Do I have to work twice as hard to make myself understood? Can I be transferred to a native English speaker, please?

Associates experience stress, too. They need to concentrate harder to understand customers — and respond with clarity and empathy. The added cognitive load impacts associates’ ability to multitask, to keep a conversation flowing naturally, write case notes, and navigate the knowledgebase to locate resource materials.

The challenge

Contact center associates working offshore in India were well-trained, proficient in English, and comfortable with the idioms of their customers. Call times were extended, however, because information sometimes needed to be repeated or clarified.

Further complicating matters was the nature of customer support: technology help desk, which sees a disproportionate share of agitated callers. At the outset of a call, customers were often frustrated, embarrassed, or resented having to ask for help when they were unable to troubleshoot on their own.

The client wanted to simplify interactions, speed resolution, and provide the best experience for customers calling into the global help desk.

Our solution

TTEC deployed its voice enhancement solution that softened accents of non-native speakers dynamically, in real time, during live customer interactions. The AI-enhanced tool did not eliminate accents — and customers knew they were calling an offshore site — but it modified audio on the caller’s end to add clarity, definition, and a little bit of space between spoken words if an associate tended to speak rapidly. 

We conducted a short-term pilot test to collect key performance indicators for a control group of associates handling relatively simple inbound calls, without AI technology, and also KPIs for a test group using the accent-softening tool for handling more complex customer calls.

The results

Overall average handle time (AHT) among associates using the AI tool dropped 15% (3.5 minutes), an unexpected result given that this test group handled complex calls that involved more questions and added time when help desk associates needed to login to users’ computers to troubleshoot problems.

Silent time, or “dead air,” dropped a dramatic 58% among the test group, indicating fewer pauses in conversations that could signal poor communications. Using a large language model (LLM) to analyze recorded interactions revealed fewer instances of information being repeated, fewer questions and interruptions, and reduced ambiguity. This was another counterintuitive finding because the test group fielded calls more complex than the control group, requiring more clicks through the knowledgebase to locate resource materials.

Ambiguity, defined as instances when associates or customers take extra time to clarify a point or ask a followup question, dropped 7%.

From an internal standpoint, speech analytics technology performed far better evaluating conversations involving the AI accent-softening tool than it did with recordings not using the tool.