How to Prepare an AI Voice Campaign: Pre-Launch Checklist

Most problems with an AI voice campaign begin before the first call.

The contact list mixes several audiences. Names are stored in the wrong fields. Different GEOs run through one script. The offer sounds clear internally, but takes a minute to explain. A user agrees during the call and then has to search for the promised next step.

This checklist helps catch those problems before they affect conversations and conversion.

Define One Action

Start with the action the user should complete after the call.

Examples include:

• booking a meeting;
• completing registration;
• making a payment;
• returning to an account;
• confirming an appointment;
• renewing a subscription;
• activating an offer;
• uploading a required document.

“Improve engagement” is not a campaign action. It is too broad to guide the conversation, follow-up, or tracking.

A clear action answers three questions:

• What should the user do?
• Where should they do it?
• How will the campaign know it happened?

The campaign should have one primary conversion goal. Secondary outcomes can still be recorded, but the call should not ask the user to complete several unrelated actions.

Check before launch

• Is one primary action defined?
• Can the user complete it immediately?
• Can the company see when it happens?
• Does every call path lead towards the same result?
• Is the conversion event understood by everyone involved?

Keep One Audience per Campaign

A contact list is not always one audience.

New registrations, old leads, failed payments, inactive customers, and previous buyers may all exist in the same database. They should not receive the same call.

Each audience has a different context:

• what the user already knows;
• what happened before the call;
• why the company is contacting them;
• which questions they may have;
• what the company can offer;
• what should happen next.

Mixing several use cases creates a script that needs to explain everything and fits no one.

A practical rule:

If two users need a different reason for the call, they belong in different campaigns.

Check before launch

• Does every contact match the same use case?
• Are new and existing users separated?
• Are active and inactive users separated?
• Are different offers assigned to separate segments?
• Have opt-outs and suppression lists been applied?
• Is the segment recent enough for the campaign goal?
• Is there a baseline result for this audience?

Separate GEOs and Languages

Two users can speak the same language and still need different campaign setups.

GEO affects more than translation. It can change:

• local vocabulary;
• pronunciation;
• time zone;
• currency;
• phone-number format;
• available payment methods;
• contact hours;
• offer conditions;
• regulatory requirements;
• the page or product shown after the call.

A campaign for Portugal should not use the same voice, wording, offer, and conversion path as a campaign for Brazil, only because both audiences speak Portuguese.

Check before launch

• Is each GEO placed in a separate segment?
• Is the local language variant correct?
• Are calls scheduled in the user’s local time?
• Are currency and offer terms localised?
• Are the right payment methods available?
• Does the follow-up open the correct local page?
• Are local compliance requirements reflected in the campaign?

Clean the Contact Data

An AI agent can only use the information it receives.

A list may look complete while still containing problems that break personalisation or send the wrong user into the campaign:

• full names stored in one field;
• missing country codes;
• duplicate phone numbers;
• symbols inside name fields;
• placeholder values such as NULL;
• mixed languages;
• outdated account statuses;
• inconsistent date formats;
• missing user or campaign IDs.

These issues should be fixed before the contact list enters the calling flow.

Check before launch

• Are first name and last name stored separately?
• Are phone numbers in one consistent format?
• Does every number include a country code?
• Have duplicate records been removed?
• Are empty and placeholder values excluded?
• Is the preferred language available?
• Does every contact have a unique internal ID?
• Are the required personalisation fields complete?
• Is the user’s current status available?

Check Names Before the First Call

Name quality deserves a separate check because it affects the opening seconds of the conversation.

Bad examples include:

• JOHN_SMITH_123;
• Mr John;
• John / Smith;
• NULL;
• a company name inside the first-name field;
• first and last names joined without spacing;
• a name transliterated for the wrong language.

The name may look acceptable in a spreadsheet, but sound wrong when spoken aloud.

For multilingual campaigns, pronunciation also depends on the alphabet, language variant, and voice used in the campaign.

Check before launch

• Does each name sound natural when read aloud?
• Are titles removed from the first-name field?
• Are names written in the correct alphabet?
• Is transliteration consistent?
• Are company names kept out of personal-name fields?
• Have difficult names been tested with the selected voice?
• Is there a fallback when the name field is empty or unreliable?

Make the Offer Easy to Explain

The offer should be clear after one or two sentences.

If it requires a long internal explanation, the user will struggle to understand it during a call.

A clear offer answers:

• What does the user receive?
• What do they need to do?
• Why is the company contacting them now?
• When does the offer expire?
• Are there important conditions?
• Where can the action be completed?

The call, follow-up message, landing page, and product should describe the same offer.

A call should not promise one thing while the link opens a page with different wording or conditions.

Check before launch

• Can the offer be explained in one sentence?
• Is the user’s required action clear?
• Are the main conditions easy to understand?
• Does the follow-up message repeat the same promise?
• Does the product show the same offer?
• Can the agent answer the expected questions?
• Will the offer remain active throughout the campaign?

Test the Conversation, Not Only the Script

A written script can look correct and still fail in a real conversation.

Users interrupt. They answer with one word. They ask unrelated questions. They misunderstand the offer. They request another call. They say they are not interested and then explain the actual problem.

The campaign needs conversation logic, not only approved copy.

Test:

• the opening line;
• the reason for the call;
• short and unclear answers;
• interruptions;
• objections;
• clarification questions;
• requests to call later;
• wrong-person responses;
• opt-outs;
• silence;
• call endings;
• handoffs;
• follow-up triggers.

The agent should also know when to stop. Repeating the same question after a clear refusal damages the conversation.

Check before launch

• Does the opening explain the reason for the call?
• Can the agent handle interruptions?
• Are the main questions and objections covered?
• Can the agent clarify the offer without repeating the script?
• Does the agent recognise a clear refusal?
• Can the user request another time?
• Is opt-out logic defined?
• Are escalation and handoff rules clear?
• Does each conversation end with a clear next step?

Set the Follow-Up Logic

Not every contact should receive the same message after the call.

The follow-up should depend on what happened:

• no answer;
• call completed;
• user asked for a link;
• user requested another call;
• user showed interest but did not convert;
• number was invalid;
• user declined;
• user opted out.

A generic SMS sent after every call ignores the context of the conversation.

The sequence also needs a stop condition. Once the user converts, opts out, or becomes ineligible, the campaign should stop contacting them.

Check before launch

• Who receives a follow-up message?
• When is it sent?
• Does the message reflect the call outcome?
• Does it include the exact next step?
• How many contact attempts are allowed?
• When does the sequence stop?
• Does it overlap with existing CRM communication?
• Are converted users removed from the sequence?
• Are opt-outs passed back immediately?
• Are callback requests recorded?

Check the User Journey After the Call

A user saying “yes” during the call does not mean the campaign has converted them.

The next step may still fail because:

• the link opens a general page;
• the user has to log in again;
• the promised offer is difficult to find;
• a form asks for information the company already has;
• the page does not work correctly on mobile;
• the action requires several screens and confirmations;
• the offer shown in the product differs from what was explained during the call.

Open the full journey as a user would experience it.

Start with the message or link sent after the conversation and complete the action on both desktop and mobile. Check every page, field, redirect, login step, and confirmation.

The user should arrive directly at the action discussed during the call.

A user should not have to search for what the call just promised.

Check before launch

• Does the follow-up link open the correct page?
• Is the user recognised or logged in where possible?
• Is the promised offer already visible?
• Does the product wording match the call and message?
• Are unnecessary fields and clicks removed?
• Does the journey work on mobile?
• Is the completed action clearly confirmed?
• Can the user return to the same step after closing the page?
• Is support available if the action fails?

Check the Data Flow Before Launch

The campaign needs enough user context to run the right conversation and a reliable way to return the result.

Before launch, confirm which data is passed into the campaign:

• user ID;
• segment;
• GEO;
• language;
• account or lifecycle status;
• offer eligibility;
• fields required for personalisation.

The data also needs to move in the other direction.

Call outcomes, callback requests, opt-outs, and completed actions should return to the CRM or internal analytics system and remain connected to the same user ID.

Without this feedback loop, converted users may stay in the sequence, opt-outs may receive another message, and the company may not be able to connect outreach to the final result.

The internal report should connect the original audience, conversation outcomes, completed actions, and generated value. Answer rate and call duration show whether communication happened, but not whether the campaign changed the result.

Check before launch

• Does every user have a unique internal ID?
• Are segment, GEO, language, and eligibility passed correctly?
• Are converted and opted-out users excluded?
• Do call outcomes return to the internal system?
• Are callback requests and opt-outs recorded?
• Is the conversion event clearly defined?
• Can the completed action be matched to the contacted user?
• Can revenue or completed value be included in reporting?
• Is the attribution window agreed?
• Can the internal team see the final campaign result?

Run a Test Segment

Do not start with the full database.

A small test segment can reveal:

• incorrect pronunciation;
• missing fields;
• outdated account statuses;
• broken links;
• unclear offers;
• unexpected objections;
• missing campaign outcomes;
• follow-up delays;
• conflicts with other communication.

The test should contain enough contacts to produce real conversations, not only internal calls between colleagues.

Review recordings, outcomes, user journeys, and completed actions before scaling.

Check before launch

• Have internal test calls been completed?
• Has a real test segment been selected?
• Are different user outcomes represented?
• Have links and conversion events been checked?
• Have real conversations been reviewed?
• Has the conversation logic been updated after testing?
• Has the internal report been checked?
• Is there a clear decision on when to scale?

Confirm Who Covers Each Part of the Campaign

The client should not need to build or supervise every technical layer of an AI voice campaign.

Before launch, define which parts are handled by the campaign provider and which remain with the internal team.

Bswan manages the execution layer, including:

• AI voice setup;
• scripts and conversation logic;
• telephony infrastructure;
• messaging providers;
• delivery routes;
• retry logic;
• call QA;
• delivery monitoring;
• campaign optimisation;
• technical campaign tracking.

The client provides:

• the audience;
• the offer;
• business rules;
• compliance and eligibility requirements;
• campaign approvals;
• product access;
• the data required to measure the result.

The split should be clear before implementation begins.

Check before launch

• Who prepares and updates the audience?
• Who approves the offer and conversation?
• Who manages campaign execution?
• Who can pause or change the campaign?
• Who checks the user journey?
• Who validates conversion data?
• Who reviews the final business result?
• Who makes changes when conversion drops?

AI Voice Campaign Pre-Launch Checklist

Goal

• One primary user action is defined
• The action can be tracked
• Every conversation path supports the same goal
• The conversion event is understood by everyone involved

Audience

• One use case is assigned to the campaign
• New, active, and inactive users are separated
• Different offers are assigned to separate segments
• Suppression and opt-out lists are applied
• The segment has a clear baseline
• The audience is recent enough for the campaign goal

GEO and language

• GEOs are separated
• Language variants are correct
• Local time zones are applied
• Currency and offer terms are localised
• Local payment methods are available
• Follow-up links open the correct local product

Contact data

• First and last names are separated
• Phone numbers include country codes
• Duplicates are removed
• Placeholder values are excluded
• Every user has a unique internal ID
• Preferred language and GEO are available
• Account and eligibility statuses are current
• Personalisation fields are complete

Offer and conversation

• The offer can be explained in one sentence
• Conditions are clear
• Call, message, and product wording match
• The opening line has been tested
• Main questions and objections are covered
• Interruptions and unclear answers are handled
• Opt-out and callback logic is defined
• The agent knows when to end the call

Follow-up

• Messages depend on the call outcome
• The exact next step is included
• Retry timing is defined
• Converted users leave the sequence
• Opted-out users are suppressed
• Existing CRM communication does not overlap

User journey

• The follow-up link opens the exact next step
• The offer shown in the product matches the call
• The user is recognised where possible
• The journey works on mobile
• Unnecessary fields and clicks are removed
• The completed action is clearly confirmed

Data and reporting

• Segment, GEO, language, and eligibility are passed
• Converted and opted-out users are excluded
• Campaign outcomes return to the internal system
• Callback requests and opt-outs update the user record
• The conversion event is clearly defined
• Conversion can be matched to the contacted user
• Revenue or completed value can be included
• The attribution window is agreed
• Reporting connects outreach to the final result

Testing and ownership

• Internal test calls are complete
• A real test segment has been launched
• Real conversations have been reviewed
• Links and conversion events have been checked
• Responsibilities between the client and provider are clear
• One person owns the final campaign result