An English-only voice agent can complete the dial and still fail a campaign because a call only works when people can understand, answer, and stay in the language they choose. For Ghanaian audiences who move between Twi and English, campaign teams should evaluate complete conversational turns, including language switches, refusals, and escalation, before treating call volume as success.
In April 1970, Apollo 13 had a problem that was small in shape and enormous in consequence. After the spacecraft’s oxygen tank explosion, Jim Lovell, Jack Swigert, and Fred Haise needed to use the lunar module as a lifeboat. Its carbon-dioxide system created a mismatch: the command module used square filters, while the lunar module accepted round ones.
NASA engineers had to make the available parts work together before carbon dioxide became a greater danger. The improvised adapter is documented in NASA’s Apollo 13 history. The mission did not need another square filter. It needed a way for the crew’s existing equipment to connect and function under pressure.
That is the useful comparison for a campaign team on a Monday morning. The English-only agent may have the phone number, the script, the campaign list, and a successful connection. Yet when a recipient answers in Twi, mixes Twi and English, or asks a question in the language they use at home or work, the campaign has reached its compatibility problem.
A completed call is not a completed conversation
A calling dashboard can report that an agent placed every planned call. It can show connected calls, durations, and completed scripts. Those signals matter, but they do not establish that the audience received the message or could act on it.
Imagine a campaign agent opening in English and asking a recipient to confirm an appointment. The recipient answers in Twi, perhaps with a clarification, a concern, or a request to repeat the detail. If the agent cannot recognize that turn accurately, respond naturally, or hand the call to a person with enough context, the campaign has already lost the important part of the interaction.
This is not a claim that every Ghanaian customer wants every call in Twi. Many conversations will continue in English. The operational requirement is broader: the agent must handle the language the caller actually uses, and the team must know when it did not.
That requires evaluating more than a polished first response. Test the parts of a call where meaning can drift:
- A customer switches to Twi after an English greeting.
- A caller asks for a detail to be repeated in a different language.
- A refusal or opt-out is spoken naturally rather than delivered in the campaign’s preferred wording.
- The agent cannot answer a question and needs to route the caller to a human owner.
Asenda Talk is being built for this problem with native Twi speech recognition and synthesis fine-tuned in-house, alongside English-capable agent workflows. It is an early-access platform, and teams should evaluate its speech quality against their own call types before relying on it for live customer communication.
The language test belongs in campaign acceptance criteria
Teams often test whether an agent can place a call, deliver an opening, and capture a simple response. Those checks are necessary, but they leave out the moment that determines whether a recipient feels heard.
Write language behavior into the campaign’s acceptance criteria. Define the target opening language, the likely switch points, the responses that need review, and the conditions for handoff. Use real phrases from consented test participants where possible. A clean transcript of a short prompt is useful evidence. It does not prove that the agent can handle a billing question, repayment query, support request, or a change of mind halfway through a sentence.
For a deeper example of the handoff problem, see A Twi Switch Risks Lost Context. The Agent Must Know When to Escalate.
Language evaluation should also include what the agent records after the call. If a customer declines, opts out, or asks for follow-up, the team needs an audit trail that preserves the outcome and the reason it matters. A vague label such as “call completed” cannot tell an operator whether to retry, escalate, or stop contacting that person.
Build the operating controls before live outbound calls
Natural conversation is only one part of responsible voice automation. Campaign teams also need consent handling, opt-out capture, call-truth tracking, and clear ownership when an agent cannot proceed.
Asenda Talk includes consent, opt-out, and audit-trail records for every call, plus lifecycle webhooks that track what happened across the call process. Its per-minute billing is metered, with an operator-controlled gate before real-money use. These controls help teams test deliberately instead of discovering a language failure after a large call batch has run.
Outbound calling itself remains gated behind an explicit telephony-provider decision that is not live today. That limitation should shape planning. Teams can configure agents, evaluate voice behavior, and prepare their review process now, while keeping live outbound commitments separate from what the platform currently supports. What Happens When Your Outbound Agent Has No Telephony Provider? explains why that distinction matters.
Treat the first unclear response as useful evidence
Apollo 13’s crew could not solve the filter problem by pretending the parts fit. The mismatch had to be identified, worked around, and checked under the conditions they actually faced.
Campaign teams should apply the same discipline to the first call where an English flow meets a Twi-speaking customer. Preserve the relevant transcript and outcome. Review whether the agent understood the request, whether its response remained accurate, whether consent or opt-out status was captured, and whether a human had enough information to continue the conversation.
A campaign that reaches people in the language they use has a better foundation than one that merely reaches their phones.
Comments
No comments yet.