Contact center

Put bots on the queue beside your team.

A bot answers calls, chats, texts, and emails from the queues you give it, works from your playbook and knowledge base, and brings in a person when it needs judgment.

It works inside the limits you set, and your managers review it the way they review the rest of the team.

Start free Talk to us

100+ countries · Cloud, hybrid, or onsite · 1 to 1,000 seats

Task list with the charged-twice email completed by the Billing Bot and the full refund and cancellation request, tagged Needs a person, connected to Marcus — The billing bot finished the charged-twice email. The refund and cancellation request went to a person.

Shared queues

Bots and agents share queues and routing rules.

Each bot is a worker in the TaskRouter, with skills, languages, capacity, and a priority, so routing decides whether the next task goes to a bot or a person. Managers watch bots live, score them against the playbook, and see their handle time on the wallboard with the rest of the shift.

Team view with the Billing Bot on task beside Marcus Oyelaran, who's working the full refund and cancellation request — The billing bot is on the team list, on task. Marcus, on the same queue, has the refund request.

Offered work.
A bot gets work as an offer, by the skill, priority, and capacity rules everyone else gets.
Skills pick the worker.
If a task needs a skill the bot doesn't have, it goes to someone who does.
Own a queue, or share one.
Run the after-hours queue on bots, or add one to a busy daytime queue.
Asks for help.
A bot can invite a person into a live task, just like an agent can.

One email, start to finish

“I was charged twice this month.”

Here's how one billing email goes, from the bot's first read to the person who finishes it.

It lands on the billing queue. At 10:02 Daniel Reyes at Harbor Analytics writes in, the email turns into a task, and the billing bot gets the offer, the way anyone on billing would.
The bot checks and credits. It reads the account and both charges, confirms the duplicate, and issues the $149 credit through the payment step, within the limit its permission set allows. Then it replies with the confirmation.
A receipt is still the bot's job. At 10:31 Daniel asks for a receipt for his accountant, and a minute later the bot sends it.
Then he asks for something a bot shouldn't settle. At 10:45 he wants everything he's paid this year refunded and the account cancelled. Full refunds and cancellations sit outside what you let bots do, so the AI marks the task as needing a person and routes it to the billing team with the thread, the credit, the receipt, and the bot's notes attached. Marcus Oyelaran has it a minute later.
Nobody starts over. Marcus reads what Daniel wrote, what the bot did, and what's still open, so Daniel doesn't explain the double charge again. Later, the manager reviews the bot's part on the same scorecard as the team's.

The billing task timeline: the refund and cancellation email, the AI marking it Needs a person and routing it to billing, the assignment to Marcus Oyelaran, and the bot's receipt and confirmation below — The refund request, marked as needing a person and routed to billing. Marcus gets it with the bot's credit, receipt, and notes on the same task.

AI Workflows

Build a bot on the workflow canvas.

The Bot module sits in a workflow with three exits. Resolved means the bot finished the job. Escalate transfers the task to a queue or a named person by skill, or opens a task for the on-call team. Error is the fallback you design.

Bot module on the workflow canvas with Resolved, Escalate, and Error exits — Three exits from the bot step: Resolved when the credit is issued and confirmed, Escalate when a person takes the task, and Error for the fallback you design.

Playbooks and scripts

The words and goals you approved.

Agent goals steer the bot's side of the conversation, and scripts set how it opens, confirms details, and wraps up.

Knowledge

Answers from your articles.

The bot answers from your knowledge base articles, so when you update one, its answers change too.

MCP tools

Look it up, then act.

Mid-conversation, a bot can pull account details and recent charges, issue a credit, or collect a payment. On another queue, a bot confirms appointments the same way.

Versions and sessions

Drafts stay drafts.

Edits reach customers only after someone with publish rights publishes them, and every session records the path the conversation took.

Managing a bot

How a manager watches, scores, and changes a bot.

Managers run bots like people on the floor: watched live, scored, and reviewed. A bot's numbers sit on the wallboard beside the rest of the team.

Monitor. Open a live bot conversation from the team view, follow along, and leave feedback right in it.
Score. Its conversations get the playbook rubric and QA scorecard your agents work to, and the manager can override any score.
Change. Feedback turns into edits to the bot's playbook, scripts, knowledge, or tools, and goes live as a new published version. The bot doesn't rewrite itself.
Step in. When a conversation needs a human, the manager takes the task over from the bot and finishes it.

The Duplicate charge playbook with all five goals met, including Hand off refund requests, and the script step that hands the refund request to billing — The Duplicate charge goals a review scores against, handing off refund requests included. The script step the bot followed to pass it on.

You stay in control

You decide what a bot can see, say, and do.

A bot usually gets a narrower scope than an agent: a few queues, a few tools, a few record types. Its credit limit is one of those settings: a credit inside the limit goes through, and one over it goes to a person before any money moves. Some requests you can keep off its plate entirely, so full refunds and cancellations always go to a person. It all runs under the same permission sets and security settings as your team, so there's no second admin console to keep in step.

Permission sets.
Who can edit and publish a bot, and the limits it works within, like the largest credit it can issue and the requests it hands to a person.
Record access.
Record-type permissions and workflow policies decide which records a bot can read and change.
Media policy.
You decide what gets recorded on the conversations a bot works.
A trail you can read.
Workflow sessions keep each bot conversation's run history, and a bot's record changes are logged alongside your team's.

Questions

What buyers ask before they hire a bot.

Which work can an AI agent take from a queue?

Voice, chat, SMS, and email tasks on the queues you assign it. The TaskRouter offers it work by skill, language, capacity, and priority, just as it does for your agents.

What if a bot can't resolve an issue?

It hands the task to a person. Say a customer the bot just credited writes back asking for a full refund and to cancel the account. The bot doesn't try to settle that: the AI reads it as a request that needs a person and routes it to your billing team, with the conversation, the credit, the receipt, and the bot's notes attached, so the customer doesn't start over.

What does a manager's feedback change?

The bot's playbook, scripts, or knowledge, or the tools and permissions that set what it can do. Each change goes out as a new version, and the bot doesn't change itself in between.

What can a bot reach outside Unbound?

The MCP tools connected in its workflow, like an account lookup or payments, plus whatever records its permission set allows. Adding a tool is a workflow change, so customers only see it after someone with publish rights publishes it.

Put a bot on a queue and call it.

Start a 30-day trial with sample data loaded, put a bot on a queue, call the number, and watch it work from the team view. Then point it at your own playbook.

Start free Talk to us