Select the Right Persona for Your Agent
Selecting the right persona for your agent shapes how every call with your customers goes, so it's worth selecting deliberately rather than by first impression.
Required Editions
| Available in: Lightning Experience |
| Available in: Enterprise, Unlimited, and Developer Editions with Foundations or Agentforce 1 Editions, and Salesforce Voice add-ons. |
A persona is the voice that your agent speaks with. It covers everything that goes into how your agent sounds, including its tone, style, age, gender, and accent.
You select a persona in the Voice Settings page of the Agentforce Builder. Set your agent's default language first, though: each voice agent is monolingual, and its default language filters the personas available to you. For the full configuration steps, see Customize the Persona of Your Agent.
Getting the persona right comes down to three things:
- Know your context. Your use case and audience drive concrete choices, like speech rate, accent, and how much a persona elaborates. There's no single best persona, only the one that fits your customers' situations.
- Screen for quality. Select a voice that represents your brand and feels like a natural partner to your customers. Keep the voice separate from the agent's instructions: fix issues like long intros or repetitive apologies by refining your instructions, not by changing the persona.
- Test and fine-tune. Test a persona against your real use case, not just one sample line. If the default doesn't fit, fine-tune one setting at a time and re-test after each change.
However you select, test the persona before you rely on it. A voice can sound right on a single sample line and still change in a real, multi-turn conversation, with pauses, interruptions, and corrections. Always audition your choice in a full interaction, not just a quick sample, before you commit to it.
Know Your Context
Your use case and audience drive these choices. There's no universally best persona, only the one that best fits the situations, your use case, and audience.
| Consider... | Select a persona that... |
|---|---|
| Your use case is complex, or your audience is unfamiliar with your product | Defaults to a moderate, unhurried pace |
| Your use case is simple or transactional and your audience is experienced | Defaults to a brisker pace |
| Most of your callers are in one region | Matches that region's accent and pronunciation |
| Your audience spans many regions | Uses neutral, widely understood pronunciation, and avoids regional idioms |
| Your use case is quick and transactional | Tends towards concise responses |
| Your use case is complex or high-touch | Elaborates on answers when a customer needs more detail |
This table covers the choices that your use case drives directly. You judge a persona's other qualities, like its overall character and how well its tone fits your brand, by ear rather than by rule.
A persona's default pace and expressiveness are a starting point, not fixed. Select a persona mainly for the qualities you can't change later, like its accent and character. You can fine-tune pace and expressiveness afterward.
Your language choice also affects which voice settings are available. English supports all voice settings. Support for other languages varies, and some are still in beta with limited support for advanced settings like the Pronunciation Dictionary. If you're using a Beta language, test carefully before deploying to production. Keep in mind that each language also has its own default characteristics, such as pace and delivery, so factor your language into your persona choice. For the current list of supported languages, see Agentforce Voice Considerations.
Screen for Quality
When you audition a persona, listen for how it sounds, regardless of your use case. Two qualities are worth paying attention to:
- Articulation — the voice is clear and easy to follow, with emphasis that feels appropriate, not over-emotive, or flat and unvarying.
- Brand alignment — the tone is inviting and professional and fits your brand, neither overly familiar nor robotic.
Both are qualities of the voice itself, so if one feels off, screening a different persona is usually the fix. Some qualities also come from your agent's instructions rather than the voice. If specific words, such as brand or product names, are mispronounced, add them to the Pronunciation Dictionary when you fine-tune.
Test and Fine-Tune
Start by testing a persona's default settings. In Voice Settings, enter a sample message and use Play to hear how the persona says it. Use a message that your agent would say, including the words, product names, and phrases common in your business. Try a few different phrases rather than just one. The more realistic the sample, the more you learn. A persona can sound polished on one demo question and still turn robotic after it's handling a real, multi-turn conversation with pauses, corrections, and interruptions. Test it through a full, realistic interaction before deciding.
Play vs. Preview: The Play control in Voice Settings speaks whatever text that you enter in the message box, so you can hear the persona's voice on your own sample lines. It plays only that text. It doesn't run your agent. To hear the persona inside a real, simulated conversation, with your agent's instructions and subagents in play, use Preview at the agent level to run a full interaction. Because a persona can hold up on a single-line but change under real conversational load, check it both ways: use Play for quick voice comparisons, and Preview to test it through a full interaction.
If the default doesn't completely fit, fine-tune it with these settings.
- Similarity — controls how closely the generated speech matches the original persona.
Two additional tools help with specific pronunciation and recognition issues. Use them only after testing surfaces a real problem:
- Pronunciation Dictionary — add words or phrases unique to your org that are likely to be mispronounced, with a specified pronunciation (CMU for English, IPA for non-English speech).
- Keyword Boosting — add up to 100 single terms or phrases (uncommon words, proper nouns, product names, industry-specific terms) to improve the model's recognition and transcription accuracy.
For guidance on how to use these tools to handle specific recognition challenges, such as regional pronunciations and place names, see the Agentforce Voice Best Practices guide.
Change one setting at a time and re-test after each change. This approach makes sure that you identify what truly made a difference, instead of layering multiple adjustments and losing track of what is effective.
Switching to a different persona
If testing leads you to switch personas, treat the new persona as a fresh selection rather than a small adjustment. When you switch, the voice tuning settings automatically reset to the new persona's defaults, so any tuning you did for the previous voice doesn't carry over. That's by design: a different voice has its own default pace, delivery, and articulation. Start over with the same process: listen to the new persona's defaults, screen it for quality, test it against a full, realistic conversation, and only then fine-tune.
