What happens when a customer starts a conversation?
The widget sits in the bottom-right corner of your site. A customer types what they need help with, allows the microphone and starts talking. Before the call begins, the widget tells them they’re speaking with AI, that a transcript will be saved and that the call lasts up to 10 minutes.
During the call they can read live captions, mute the microphone or hang up. When it ends, they can mark the conversation as helpful or start a new one.
Where does voice support help SaaS customers most?
Voice suits questions people would rather explain than type, especially when they’re partway through a task. Start with one area where your team already has good written answers.
- Setup help during onboarding, when a new user explains what they’ve connected so far and where they got stuck.
- Feature questions, such as how something works or where to find a setting.
- Troubleshooting that your docs already cover, talked through step by step.
- How-to questions from your help centre, for customers who would rather ask than search.
How do answers stay grounded in approved knowledge?
Your team adds FAQs and notes, or imports selected .md, .mdx and .txt files from GitHub, and approves each source before the agent can use it. Adding a source never approves it, and every GitHub re-sync needs approval again, so changed docs don’t reach customers until someone has checked them.
During a call, the agent searches only the approved sources for its channel. If they don’t answer the question, it’s told to say it doesn’t know and point the customer to your website’s usual contact options. If you’ve turned on human handover, an early release, it offers to connect them to your team instead. It’s also told to treat source text as information and never to follow instructions found in it, and the topic a customer types is labelled as untrusted.
Approval controls what the agent can use. It doesn’t guarantee that every answer is right, so test with real questions and read the transcripts before you switch on customer calls.
What will customers hear?
You choose the agent’s name, a greeting of up to 500 characters and instructions of up to 4,000. The agent tells callers it’s an AI and greets them in English. There’s no language setting.
Pick one of two voices: Marin, which is warm and expressive, or Cedar, which is calm and grounded. You also set the button label and colour, so the widget fits your site.
What can’t the support agent do?
These limits apply to every customer workspace today. Keep your usual contact routes on the page so customers can still reach your team.
- It can’t look up or change a customer’s account, orders or subscription, and it doesn’t check who is calling.
- It passes a caller to your team only if you turn on human handover, an early release. Your team verifies a UK or US phone number in Settings, the person who answers hears a private briefing first, and customer routing is still being tested.
- It works in the browser only, with no phone number, text chat, SMS or email channel.
- It doesn’t connect to your help desk or CRM. You read conversations in your dashboard or through the REST API and MCP.
- Each call stops at 10 minutes, and a workspace takes up to 3 calls at once and 12 new calls a minute.
What does your team see after each call?
Each conversation is saved with its topic, status, length and a two-sided text transcript, marked as support or sales. No call audio is recorded. Search by topic, filter by channel and status, and open the transcript to read what the customer asked, as transcribed, and what the agent said.
When a customer asks, the agent can record a bug report, feature request or other feedback on the conversation, without promising a fix. Use what you read to add missing answers and improve the docs your agent relies on.
The Insights tab also groups recurring questions, reported issues and themes from up to your latest 30 completed customer calls for each agent in the last 30 days, with quotes from the transcripts.
Start with one part of your help centre.
Bring a website, a small set of product documentation and a few representative questions. A pilot lets you check the answers and the caller experience before you widen the scope, and its first 50 conversations are free, including your own tests.
- Choose a product or help-centre section.
- Approve the sources the agent may use.
- Test the questions customers actually ask.
- Review transcripts before enabling wider access.
Common questions
Will the AI agent replace our support team?
No. Kawl Now handles product questions that your approved docs already answer. Your team keeps account issues, bugs that need investigating and anything the agent can’t answer, and the transcripts show you which questions customers bring to it.
Can the support agent look up a customer’s account?
No. Kawl Now answers from the product knowledge you approve and can’t see customer accounts, orders or subscriptions. Identity checks and account-specific support are outside the pilot’s scope, and its default instructions tell it never to claim it has changed an account.
Which languages and voices does Kawl Now use?
Kawl Now agents greet callers in English, and there’s no language setting, so plan for English-speaking customers. Each agent uses one of two voices: Marin, warm and expressive, or Cedar, calm and grounded. Custom and cloned voices aren’t available.
What happens when the agent doesn’t know an answer?
A Kawl Now agent is told to say it doesn’t know. With human handover off, it suggests your website’s usual contact options, such as your help centre or email. If you’ve turned on handover, an early release, it offers to connect the caller to your team instead. Either way the conversation is saved, so your team can add the missing answer to your approved knowledge.
Can customers report a bug during a call?
Yes. When a customer asks, the Kawl Now agent can record a bug report, feature request or other feedback on the conversation. It’s told not to promise a fix, and you’ll find the note with the transcript in your dashboard.