A Twi-speaking caller who exits an English-only IVR before reaching a person has taken their question, purchase, payment, or support request somewhere else. Treat that drop-off as a revenue event you can measure, not as a secondary language preference.
On September 23, 1999, NASA lost contact with the Mars Climate Orbiter as it approached Mars. The investigation found that one part of the mission used pound-seconds while another expected newton-seconds. The spacecraft was lost because the systems did not share the same operational language. NASA documented the failure in the Mars Climate Orbiter Mishap Investigation Board Phase I Report.
A customer calling from Kumasi, Accra, or elsewhere in Ghana faces a much smaller version of that mismatch. They hear an English menu, hesitate at the first instruction, and end the call before a staff member can help. The business may record a short call. It may see no completed payment or support case. What it misses is the reason: the caller met a system that did not recognize the language they were ready to use.
An IVR is part of the sales and service journey
A phone menu is often treated as plumbing. Press one for billing. Press two for support. Wait for an agent.
For a caller, it is the front door. If that door speaks only English and the caller is more comfortable in Twi, the business has placed a translation task ahead of the actual task. Someone calling to confirm an order should not have to decode menu options before they can ask where it is. Someone trying to resolve a payment issue should not need enough confidence in spoken English to navigate the first thirty seconds.
That friction matters most when the caller already has a reason to leave. A delayed delivery, an unclear charge, a missed appointment, or a campaign callback comes with limited patience. Every extra step gives the caller an easy reason to hang up.
The result is easy to misclassify. Teams may label it abandonment, low-intent traffic, or a failed campaign contact. Yet some of those calls began with intent. The system lost it before the business could identify the person’s need.
Language choice reveals where the call breaks
The Mars Climate Orbiter failure was not caused by a lack of engineering effort. It came from an unhandled mismatch between connected systems. An English-only voice flow can create the same kind of blind spot: the business hears a call start, while the customer hears an obstacle.
Call data should distinguish between a call that rang, a call that connected, a caller who heard the opening message, and a caller who reached a useful outcome. Without those stages, a short Twi-speaking call can disappear into a broad “started” total.
Review the first moments of your inbound flow. Look for callers who disconnect immediately after the greeting or menu. Compare those calls with completed interactions, callbacks, and human-agent transfers. Ask support staff which questions customers repeat after finally getting through. These are practical clues that the entry point is doing too much of the work.
Twi-English switching also matters. A caller may begin in English, then move into Twi when explaining the real issue. A system that recognizes only one language can turn a routine support conversation into repetition and frustration. What Happens When Your Voice Agent Cannot Understand yɛfrɛ me? explores what that failure can mean in a live interaction.
Build the first minute around the caller
Start with a short greeting that makes the available language paths clear. Keep the opening focused on the caller’s job: payment, delivery, account help, booking, or campaign response. Do not place a long company introduction before the caller can state why they called.
Then test the flow with real Twi and English utterances, including switching between the two. A language option that exists on a script but fails when callers speak naturally does not solve the problem. Test recognition, confirmation prompts, transfers, and fallback paths.
Asenda Talk is built for teams evaluating Twi and English voice interactions. Its Twi speech recognition and synthesis are fine-tuned in-house, while agents can be configured with a persona, first message, and voice. The platform also records telephony lifecycle events and call truth, so operators can inspect where a call progressed or stopped.
It remains in early access. The assistant runtime is Vapi-orchestrated, and outbound calling remains gated behind an explicit telephony-provider decision that is not yet live. Teams should evaluate the current product for the capabilities available today, rather than assume feature parity with established voice-agent platforms.
Measure the revenue opportunity before scaling calls
Do not begin by asking how many calls an agent can make. Begin by asking how many callers fail before they can explain what they need.
Track language selection or language detection where appropriate. Track first-message completion, successful intent capture, transfer to a human, opt-out requests, and final call outcome. Keep consent and opt-out records attached to every call, especially when a customer asks to stop receiving contact. What Happens When a Caller Switches to Twi, Then Says Stop Calling Me? covers why language handling and call controls must work together.
NASA could not recover the Mars Climate Orbiter after the mismatch had shaped its trajectory. A business has a more repairable problem. Listen to the first minute of real customer calls, identify where language becomes a barrier, and give callers a path they can use before the opportunity leaves the line.
Comments
No comments yet.