Functions
Last updated: May 13, 2026
Here you will find all the information about the technical features available to you.
Creativity
With the creativity setting (often called "temperature") in an AI model, you determine how free and original the AI's responses will be.
Low creativity (strictly follows the prompt): The responses are focused, predictable, and closely follow known patterns – good for facts and clear instructions.
High creativity (very creative): The responses become more unusual, versatile, and sometimes surprising – ideal for brainstorming, creative texts, or when new ideas are needed.
With this slider, you can control whether the model acts more conservatively or creatively.
When each creativity setting is useful:
Low creativity:
When the AI should respond reliably, consistently, and precisely
Ideal for support requests, appointment scheduling, standard processes, or sensitive information (e.g., prices, opening hours)
Advantage: Responses stay close to the given script and there are no surprises
When the AI must strictly follow complex instructions or multi-step procedures without deviation - set creativity to 0 for maximum prompt adherence
Medium creativity:
Good for friendly but structured conversations – such as follow-up questions, explanations, or some small talk
The AI can react individually but remains safely within limits
High creativity:
For creative tasks like idea generation, marketing, casual conversations, or when the dialogue should feel particularly human and flexible
Useful when the AI should respond spontaneously to unusual input
Caution: Responses can sometimes be unpredictable or less appropriate
Precise Information Processing
With the "precise information processing" feature, you can specifically define certain data such as addresses, phone numbers, or customer numbers. The AI phone assistant activates a particularly precise (but slightly slower) model for this. This ensures important information is recognized and processed more reliably.
What does precise information processing enable?
Important information such as email addresses, phone numbers, or customer numbers is recognized without errors
A more powerful model is activated during the conversation
Information is correctly stored and errors in post-processing are avoided
Advantages:
Higher accuracy for critical data
Less manual rework thanks to correct extraction
Ideal for processes with further processing, e.g., CRM systems or follow-ups
How to set up precise information processing:
Select the AI assistant you want to configure in your Fonio account
Under the "Technical" tab, find the "Precise Information Processing" feature
Add new information in example formats, for example:
Email address → Example:
Phone number → Example: +49 176 12345678
Customer number → Example: 1234
Save changes
Advanced Prompt Protocols for Better Accuracy
To further improve recognition accuracy for phone numbers and email addresses, you can add specific behavioral protocols to your assistant's prompt. These protocols guide the AI on how to handle pauses, when to interrupt, and how to confirm information.
Protocol for Phone Number Capture
### BEHAVIORAL RULES FOR DATA CAPTURE (PHONE NUMBERS)
1. PATIENT LISTENING: Be extremely patient when the customer provides a phone number or other digits. People often make natural pauses (thinking pauses) when reading numbers.
2. INTERRUPTION STOP: NEVER interrupt the customer while they're stating a number. Even if there is a 1-2 second pause, actively wait.
3. END RECOGNITION: Only assume the number is complete when:
- The customer explicitly says they're finished (e.g., "That's it" or "Did you get that?")
- There's an unusually long silence of more than 3 seconds
4. BACKCHANNELING: While the customer is speaking, do NOT use confirmation words like "Yes," "Okay," or "Mhm," as this can disrupt the customer's speech recognition and interrupt their flow. Stay silent until the entire number has been stated.
5. CONFIRMATION: Once the number is fully captured, repeat it slowly and clearly in digit groups (e.g., "0176... 123... 45... 67") to ensure correctness.
### GOAL
A smooth, interruption-free capture of contact information without premature interjections.Protocol for Email Address Capture
### EMAIL CAPTURE PROTOCOL
1. ACTIVE SPELLING REQUEST: When asking for the email address, actively guide the customer. Say: "Could you provide your email address and spell it out for security?"
2. PHONETIC ASSISTANCE:
- When the customer spells, remain absolutely silent (no "Mhm" or "Yes")
- If you don't understand a letter, ask specifically using the phonetic alphabet (e.g., "Did you mean B as in Bravo or P as in Papa?")
3. SPECIAL CHARACTER TRANSCRIPTION:
- "At" or "at sign" = @
- "Dot" or "period" = .
- "Hyphen" or "dash" = -
- "Underscore" = _
4. CONFIRMATION LOOP (ESSENTIAL):
Once the customer finishes, ALWAYS repeat the email address for confirmation. Spell out critical parts yourself or use words:
"I've noted: 'm' as in Mike, 'u' as in Uniform, 's' as in Sierra, 't' as in Tango, then the at sign, and then gmail dot com. Is that correct?"
5. ERROR FALLBACK:
If after the second attempt the email is still not correctly captured, say politely:
"I want to make sure the confirmation reaches you. I'll send you an SMS to this number, and you can simply reply with your email address."
### GOAL
Maximum precision through controlled conversation guidance and explicit confirmation.Note: These protocols work best when combined with the Precise Information Processing feature enabled for the relevant data types (phone numbers, email addresses, names).
Testing precise information processing:
Use the "Test Audio" function in the AI assistant
Conduct a sample conversation by stating the desired information
The assistant takes a bit more time but recognizes the data more reliably
In the call history, you can then see which variables were recognized
Call Forwarding
Coming soon...
DTMF / Tone Dialing
Coming soon...