Where the thinking happens
AMI
AMI is not a bolted-on chatbot. It is an assistant your team can talk to directly and a set of steps any workflow can call — reading the request, searching your knowledge base, and drafting the reply.
Marek reported an SSO redirect loop on Tuesday. Tier 2 took it after a transfer, confirmed the identity provider was returning a stale assertion, and told him a fix would ship this week.
Capabilities
What AMI does
Answer, and know when to stop
AMI can reply using your knowledge base with a tone you set, and hands over to a person after a set number of turns or the moment it is out of its depth.
Search before anyone spends time
A knowledge base lookup with a confidence floor. Above it, the answer can go straight to the customer; below it, the flow takes the human path.
Classify, and read intent
AMI picks the right category from the list you allow, and works out what the customer actually wants — a refund, a bug report, a question.
Suggest a priority
Weighed against the wording, the customer and the impact, with a ceiling you set and the choice to apply it automatically or leave it as a suggestion.
Draft for review, never send blind
A drafted reply, optionally with knowledge base links, waits for the agent. Nothing goes out without approval.
Summarise and read the mood
A recap posted as an internal note so whoever picks the ticket up is current, and a sentiment check that routes unhappy customers before they churn.
Works with
It already knows about the rest of the platform
No connectors, no field mapping, no nightly sync. These share the same records.
In practice
Where AMI is doing the work
- Self-service
Provision a help centre in a few steps, publish it on your own domain, and let search analytics tell you exactly which answers are missing.
- AI-assisted service
AMI works inside your workflows with explicit thresholds: a confidence floor on knowledge, a turn limit before handover, and drafts that wait for approval.