The Provider Decision Kojo Didn’t Have, and What It Almost Cost the Campaign

Asenda Talk cannot place an outbound Twi campaign today because its telephony-provider decision has not been made live. Teams can still build and evaluate the agent, its Twi and English conversations, consent language, call records, and billing controls before connecting it to a live calling operation.

At 8:40 on Monday morning, Kojo, a composite marketing lead in Accra, had a spreadsheet open beside a cold cup of tea and a launch brief waiting for approval. His team had planned to call a consented list about a community programme, beginning in Twi and moving to English when callers preferred it. The list was time-sensitive. If the calls did not begin that week, the campaign window would close before the team could learn whether the message worked.

Then he reached the operational limit: outbound telephony was not live.

That pause was frustrating, but it prevented a worse mistake. A campaign should not move from a promising demo to real calls until the provider path, call delivery, consent handling, opt-out process, and real-money controls are ready for the operator running it. A natural Twi voice alone does not make an outbound campaign safe to launch.

Build the conversation while the calling decision stays open

Kojo’s first instinct was to wait. Instead, the team used the morning to build the part that should survive any telephony decision: the agent itself.

They configured a persona, first message, and voice in Asenda Talk. Then they wrote the opening as a real conversation starter, rather than a script designed to push through every call. The agent needed to explain why it had reached out, identify the organisation behind the call, make a clear path to opt out, and respond when someone answered in Twi, English, or both.

That work matters because language switching can change the meaning of a caller’s request. A caller may begin in English, give a key constraint in Twi, then return to English for a final question. The agent needs to preserve that context through the whole exchange. Twi and English Code-Switching: Why Voice Agents Must Preserve Context explores why that detail belongs in the design and evaluation process.

Asenda Talk’s Twi speech recognition and synthesis are fine-tuned in-house. The team could use that capability to test the actual language experience they wanted to offer, without pretending that a completed test meant the campaign was ready for live outbound calling.

Treat the pause as an evaluation sprint

By midday, Kojo had replaced the launch checklist with an evaluation plan. Each test conversation had a purpose.

Could the agent understand the intended meaning when a caller switched languages? Could it deliver the disclosure clearly? Did it recognise an opt-out request and leave a record that an operator could audit? Did the first message sound like a respectful introduction, or like an automated interruption?

These are practical questions, not polish work. They shape whether a future campaign deserves to go live.

Asenda Talk includes a telephony lifecycle webhook pipeline with call-truth tracking, along with consent, opt-out, and audit records for every call. In early access, those capabilities give a team a concrete model for the evidence it should expect from a live programme. The operator-controlled real-money gate also makes the boundary clear: metered per-minute billing should remain under operator control, rather than becoming an accidental consequence of a configuration change.

Kojo asked one person on the team to review each test as if they were resolving a complaint. Could they tell what the agent said, what the caller requested, and what happened next? That question changed the notes they kept. Vague labels such as “call completed” stopped being enough.

Define the live-call gate before the campaign returns

The campaign had not failed. It had changed shape.

Before anyone scheduled a new launch date, Kojo’s team wrote down the conditions that would have to be true: a live telephony-provider decision, approved consent and disclosure language, a tested opt-out route, a process for reviewing call records, and an operator who understood when per-minute billing could begin. They also identified cases that should go to a human rather than an agent.

That list gave the team something better than a hopeful date. It gave them a release gate.

This is especially important for outreach where callers may question why they were contacted or ask to stop future contact. What Evidence Do You Need Before Calling Again After a Twi Opt-Out Complaint? sets out the kind of evidence a team should be ready to review before it tries again.

Leave Monday with a usable asset

Late that afternoon, Kojo no longer had an outbound campaign ready to send. He had a tested agent brief, revised consent wording, a set of Twi and English evaluation calls, and a short list of failure cases that needed human handling.

The spreadsheet stayed open, but the row labelled “launch” became “provider decision pending.” Beside it, the team added the work they could complete now: improve the opening, test code-switching, verify opt-out records, and decide what evidence would be required before a real person received a call.

That is the right use of an early-access platform. Build the voice experience carefully, inspect what happened in each test, and keep outbound calling paused until the operational path is genuinely live.