01.
Troubleshooting
The diagnostic conversation where the next question depends on the last answer. The AI walks the customer through, narrows it down, and points to a fix.
Examples: Connectivity issues · Device setup help
Playbooks
Describe the process.
The AI follows it.
Write your process the way you’d brief a new hire. What to listen for, what to ask, how to decide. The AI agent follows it across customer conversations, interpreting your guidance against what each customer actually says. Update the words. The behavior changes the moment you republish.
Trusted by operators running Playbooks at scale
Written by your team. Interpreted in every conversation. Visible end to end.
You write the Playbook in plain language. The agent reads it the way a person would. It adapts to what each customer actually says and stays inside the steps and conditions you wrote. Supervisor scores the result. Traces record the path. Flexibility you can see, end to end.
01
Written Playbook
Plain text. Plain English. Edited like a doc.
# Cancellation
Listen first. Don't offer before you understand the reason.
Ask why. "What's prompting this?" works.
Acknowledge before you respond.
Offer what fits the reason -
price → downgrade. not using → pause.
wrong fit → accept gracefully.
Never offer the same thing twice.
02
The agent interprets, per conversation.
Same Playbook. Different customer. Different words. Same shape.
Maya · price
i need to cancel - too expensive
Totally fair - price is a real reason. Looking at your usage, the Starter plan covers what you actually use at $12 instead of $39. Want to switch?
yeah that works.
Done - moved you to Starter. Same login, same data.
03
Supervisor + Traces
Every interpretation is scored and recorded.
Maya · price
listened · acknowledged · downgrade fit reason
96
Tom · not using
listened · acknowledged · pause fit reason
94
Iris · #38142
skipped: ask why · offered before listening
62
The Editor
Write what should happen.Watch it happen.
The editor is one window. On top, the Playbook your team writes. Below, real conversations running it. The same principles produce different conversations because the agent reads the situation, not a flowchart.
playbooks / retention
TestRepublish
Playbook
# Cancellation
Use when the customer says they want to cancel, downgrade, or stop their subscription.
Listen first.
Don't offer anything before you understand the reason. The save depends on the cause.
Ask why.
Open the question. “What’s prompting this?” works better than a list of options.
Acknowledge.
Reflect what you heard before you respond. The customer should feel understood, not handled.
Offer what fits the reason.
Price → propose a downgrade, an annual plan, or a discount that maps to the gap.
Not using it → propose a pause. Coming back is easier than re-signing up.
Wrong fit, life change, anything else → accept gracefully. A clean exit is a future return.
Never offer the same thing twice.
If they decline, don’t escalate the same offer with a bigger number. Move to the next thing or accept the cancellation.
Live conversations
Maya · #38104
REASON: PRICEi need to cancel my subscription
I can help with that. What’s prompting the cancellation?
honestly it’s gotten too expensive for what I use
Totally fair - price is a real reason and I appreciate you saying so directly.
Looking at your usage, you’re mostly in the editor and a few exports. The Starter plan covers both at $12/mo instead of $39 - would that work better?
wait that actually does cover what I use. let’s do that
Tom · #38127
REASON: NOT USINGwant to cancel - haven’t logged in in like two months
Got it. Is there something specific that pulled you away, or just busy?
just busy. work picked up. might come back to it eventually
That makes sense - and I’d hate for you to re-sign up later. I can pause your subscription for 90 days. No charges, no data lost. Resume whenever, or it’ll cancel automatically.
The processes that don't fit a flowchart.
The conversations where the right next move depends on what the customer just said, not what a script predicted they'd say. Playbooks handle the conversation. When an action is needed, like issuing a refund or filing a claim, the Playbook hands off to a Flow.
02.
Retention conversations
The save call, in chat. Listen first, propose later. The AI hears what's actually wrong before it reaches for an offer.
Examples: Subscription save
03.
Complaints before escalation
The first ten minutes after something has gone wrong. The AI listens, acknowledges, gathers context.
Examples: Initial complaint · Pre-escalation triage
Flows and Playbooks
Flows execute.Playbooks guide.
Two execution models inside the same Agent Studio. The split is about what kind of work the AI is doing in that moment. Most customers run both, on the same agent, in the same conversation.
Behavior
Actions follow exact steps defined in the flow.
Follows principles and adapts them to what the customer actually says.
When to use
Predictable, rule-bound processes - refunds, routing, data collection.
Judgment-heavy conversations - cancellation saves, complaints, complex support.
Requires
Every path mapped in advance.
A description of how to handle the scenario - not a flowchart.
Input
Structured triggers and conditions.
Natural language guidance written by your team.
Where to edit
Flow canvas - drag, connect, configure.
Playbook editor - write, publish, update the words.
Compliance
The same compliance posture, across every agent.
Every agent connected through Agent Connect runs under the same certifications as the rest of the platform. SOC 2, GDPR, DORA, EU AI Act. The audit trail captures every interaction, every routing decision, every cross-agent handoff.
Connecting your existing agents through Agent Connect doesn't introduce a compliance gap. It closes the ones you have.
GDPR
EU AIACT
DORA
GDPR
EU AIACT
DORA
Write the process. Republish in a sentence. Watch every conversation.
Playbooks turn the operational team into the team that ships AI behavior. No flowchart, no engineering ticket, no retraining. Supervisor and Traces make every interpretation visible. The split with Flows is clean: Flows execute, Playbooks guide. Most customers run both.