The night-before checklist for a Twi outbound campaign

Do not launch a Twi outbound campaign until you can prove consent, apply suppression rules before dialing, and reconstruct every test call from its logs. A complete spreadsheet is only an input; the campaign is ready when those three controls work together and the telephony provider has explicit approval.

Confirm the prerequisites before reviewing the call list

Start with the systems around the spreadsheet. You need an approved campaign purpose, a documented source for each contact, a current suppression list, named owners for launch and pause decisions, and access to test records.

You also need a live telephony-provider decision. Asenda Talk is in active early access, and outbound calling remains behind an operator-controlled real-money gate. The platform can configure agents, orchestrate assistant runtime through Vapi, track telephony lifecycle events, and meter usage per minute. Those capabilities do not authorize live dialing by themselves.

If the provider or real-money gate has not been approved, keep the campaign paused. Run platform-level checks where possible, document the blocked tests, and record who must resolve them. What happens when a voice agent is ready but its phone provider is not approved? explains why technical readiness and launch authorization must remain separate.

Match every callable row to consent evidence

Review the list row by row. Each callable number should have evidence showing who consented, what they agreed to receive, how consent was captured, and when it was recorded. A column containing “yes” is weak evidence unless it points to a source that a reviewer can inspect.

Use stable references. For example, a row might contain a consent record ID tied to a form submission or an approved system of record. Keep passwords, tokens, and reviewer credentials out of the spreadsheet and campaign notes.

Flag these rows before upload:

• The phone number has no consent reference.
• Consent covers service updates, but the planned call asks for a donation or promotes an offer.
• The source record cannot be opened by the reviewer.
• The contact withdrew consent after the list was exported.
• The number was copied from an earlier campaign without a fresh eligibility check.

One unresolved row should not disappear inside a large list. Exclude it or pause the batch containing it until the evidence is available.

Test suppression before testing conversation quality

Suppression must happen before a call reaches the agent. Build a small test set containing an eligible contact, an opted-out contact, a duplicate number, and a number that appears on the current suppression list.

Run the eligibility process and inspect the result. The eligible record should proceed to the authorized test stage. Every suppressed record should stop with a clear reason, timestamp, and rule reference.

Then test opt-out handling inside the conversation. Use direct phrases in both Twi and English, plus a natural language switch. The agent should recognize the request, end or redirect the interaction according to policy, and update the suppression state before any later campaign run.

Do not rely only on the transcript. A transcript can look correct while the downstream action is wrong. This matters when a caller begins an instruction in English and completes it in Twi. See Bilingual Voice Agents: Why a Correct Transcript Can Still Trigger the Wrong Action.

Verify the Twi agent configuration

Open the exact agent version planned for the campaign. Check the persona, first message, selected voice, campaign instructions, and permitted actions. Read names, account references, dates, amounts, and opt-out language aloud in Twi and English.

Asenda Talk uses native Twi speech recognition and synthesis fine-tuned in-house. Treat that as a capability to evaluate with your campaign vocabulary, accents, and language transitions. It does not remove the need for listening tests.

Save the tested configuration version. If anyone changes the first message, voice, prompt, or action rules afterward, rerun the relevant checks. A successful test of yesterday’s configuration does not validate tonight’s edit.

Reconstruct one test call from start to finish

Place an authorized test call only after the provider and billing gates permit it. Then follow the call through every recorded stage: eligibility decision, dial attempt, connection, agent runtime, language input, transcript, opt-out event if triggered, call end, final status, duration, and metered charge.

The call identifier should connect those records without manual guesswork. Confirm that retries, failed connections, and short calls receive accurate final states. A dashboard showing “completed” is insufficient if the provider reports a failure or the lifecycle events stop halfway through.

Check access controls while reviewing logs. Secrets should remain write-only and masked, with the correct environment selected. Campaign staff need enough access to investigate a call, but they should never copy provider credentials into tickets, spreadsheets, or shared notes.

Record the launch decision

Create a short launch record with the list version, agent version, suppression-list timestamp, test call IDs, consent sample results, provider approval status, billing-gate status, and the names of the people authorized to launch or pause.

Use three outcomes: approved, blocked, or approved with documented exclusions. Avoid “probably ready.”

Your next action tonight is concrete: select one eligible row, one opted-out row, one suppressed row, and one bilingual opt-out scenario. Trace each through the system. If any result lacks testable evidence, leave outbound dialing paused and assign the missing gate to a named owner.