Nobody can read every conversation
Manual QA covers a sample of AI and team conversations. The rest go unseen until a customer escalates.
Catch the AI answers that break your policies. Then see what your customers are really telling you: the bugs they hit, the docs they can't find, and the features they keep asking for.
Your AI agent can handle more conversations than any team could review. These gaps decide whether you hear about a problem from CraftCX or from a customer.
Manual QA covers a sample of AI and team conversations. The rest go unseen until a customer escalates.
Pricing, policies, and help articles change all the time. AI answers and saved replies that were right last month may not be right today.
Resolution and deflection rates show how much your AI agent handled. Conversation-level quality shows whether customers got the right answer.
CraftCX checks replies from your AI agent, your team, or both against your support policies. Each incident shows the rule, the message that broke it, and the tickets where it happened again.
Starts from the policies you already have
Upload a PDF or Word document, or connect Notion. CraftCX drafts the rules, and you choose which ones to check.
Watches the rules that are costly to break
Refunds approved without the right review, callbacks no one can make, and requests that skip the team that owns them.
Fits how your team reviews
Mark each finding as needs fix, resolved, or false positive, and assign an owner with a due date. Severe and repeated breaks become incidents.
Yes, I've approved the $240 refund. You'll see it in 5–7 business days.
Duplicate charge on annual plan
Good news: I've refunded the $96 upgrade to the card you used.
Charged for the wrong plan
Customers tell support about problems they never submit as formal feedback. CraftCX finds the likely bugs, usability problems, documentation gaps, and feature requests in every conversation you monitor, from your AI agent and your team. It groups them into patterns with the count and trend behind each one, so support and product can agree on what to fix first. When enough customers ask how to export invoices, that is not a support problem. It is your next feature.
Customer patterns and support incidents that CraftCX found in your conversations.
Agent approves refunds without routing them to Billing
The AI agent confirmed refunds over $50 itself instead of routing them to Billing.
Web login loops back to sign-in
Customers say the web app returns them to sign-in after they enter a correct password.
Customers cannot find CSV export
Customers ask where CSV export lives because the help center does not cover it.
Restore or undo completed tasks
Customers ask for a way to restore tasks they marked as done by mistake.
Every update to your help content, policies, or AI agent setup can improve answers or quietly make them worse. AXIS, the CraftCX score for AI answer accuracy, customer effort, and handoff quality, compares each period with the one before. Catch a drop early, and show leadership what improved with evidence instead of volume numbers alone.
Open items to review now
4 new or reopened incidents
Nobody has acknowledged them yet. Most severe: Agent approves refunds without routing them to Billing
2 rising customer signals
Reports are increasing and the signal is active. Most reported: Customers cannot find CSV export (18 reports)
Selected period, compared with the 30 days before it
AXIS score
Conversations evaluated
On-time handoffs
CraftCX connects to Front, Gorgias, HubSpot, or Intercom and reads the conversations you choose to monitor, from your AI agent, your team, or both.
It scores AI answers for accuracy, customer effort, and handoffs, checks replies against your support policies, and groups recurring customer problems.
Each finding links to the conversations behind it, and you can turn it into a task for your team.
How CraftCX turns AI-handled support into evidence your team can use.
Connect your helpdesk and see what your AI agent got right, what needs attention, and the conversations behind each finding.