12 Customer Support AI Agents, Live in Production.
12 live agents automate customer support within customer service. 6 run on a schedule, 3 run on demand and 3 run when a message arrives. Each one publishes the inputs it needs, the steps it works through and what it hands back.
-
Account Change Request
Check whether the person asking for an account change is entitled to make it, then draft either the confirmation or the verification request. Nothing is changed and nothing is sent.
-
Chat Transcript Summary
Turn a live-chat log into the two things needed afterwards: a ticket record the next agent can act on, and a clean transcript the customer can be sent.
-
Complaint Tracking
Track every open complaint against the commitments made to that customer — which are past the promised date, which have gone quiet, and which were closed without the customer ever agreeing they were.
-
FAQ Gap Monitor
Check the published FAQ against what agents are actually replying, and find the entries that have quietly gone out of date as well as the questions that were never added.
-
Inquiry Self-Service Deflection
Work out how much of last month's inbound could already have been answered by the help centre, which articles were missing, and which questions should never be deflected at all.
-
Knowledge Article Drafting
Draft a publishable article from a resolved case, checked against the existing knowledge base first so it extends what is there instead of quietly contradicting it.
-
Order Status Response
Match a customer asking where their order is to the right line in the order book, and draft a reply that states only what the record actually confirms.
-
Post-Service Survey Dispatch
Decide which closed tickets should be surveyed and which would make things worse, then draft the invite for each one. Every draft is approved individually — nothing is sent.
-
Query Resolution
Classify an inbound customer message, look up the account, and draft a reply for your approval. Nothing is sent.
-
Recurring Issue Root Cause
Cluster a month of tickets into the underlying faults behind them, and rank the fixes by how many tickets each one would actually stop arriving.
-
Service Inquiry Follow-Up
Find the open inquiries that have gone quiet on our side, and draft the chase for each one from what the thread actually last said. Every draft is approved individually — nothing is sent.
-
Technical Issue Diagnosis
Read the diagnostic export a customer sent in alongside what they reported, and work out what the evidence actually shows — separating what the logs prove from what the symptom merely suggests.
Answering from the record, and saying only what the record confirms
Most support volume is not hard, it is dispersed. Answering “where is my order” means finding the right line in the order book, reading what the fulfilment status actually means, and writing something accurate — three or four minutes of retrieval for a reply of two sentences. The pressure of a queue then produces a specific failure: rather than say what the record confirms, an agent says what it probably means, and a soft reassurance about a delivery date becomes a complaint when it does not hold. Underneath this, the same faults arrive repeatedly. Nobody has the time to cluster three thousand tickets into their root causes, so the team keeps answering the same question expertly instead of stopping it being asked.
These agents do the retrieval and hold the line on accuracy. Query Resolution classifies an inbound message, looks up the account and drafts a reply for approval, sending nothing. Order Status Response matches the customer to the right line in the order book and drafts a reply stating only what the record confirms. Technical Issue Diagnosis reads a diagnostic export alongside what the customer reported and separates what the logs prove from what the symptom merely suggests. Account Change Request checks whether the person asking is entitled to make the change before drafting either the confirmation or the verification request — changing nothing itself. Behind the queue, Recurring Issue Root Cause clusters a month of tickets into underlying faults and ranks fixes by ticket volume prevented, while FAQ Gap Monitor and Inquiry Self-Service Deflection find the articles that have gone out of date, the questions never added, and the questions that should never be deflected at all. Knowledge Article Drafting turns a resolved case into a publishable article checked against the existing knowledge base first, so it extends rather than contradicts it.
What this moves
- Time to a drafted first reply
- The message is classified, the account looked up and the response drafted in one pass, so an agent edits and approves rather than researching from scratch.
- Repeat contacts on the same fault
- A month of tickets is clustered into the underlying faults behind them and fixes ranked by how many tickets each would stop arriving.
- Replies that overstate what is known
- Order and status responses are constrained to what the record actually confirms, which prevents the reassuring guess that becomes a complaint a week later.
Customer Service
How AI agents handle customer support
Drawn from the 12 agents above — what they require, how they run, and what comes back.
What they need
- Analysis settings
- Inbound request
- Chat log
- Output settings
- Complaint register
- Policy
- Published FAQ
- Recent agent replies
What comes back
- What is being asked for
- Entitlement check
- Draft reply
- Account on file
- How the chat ended
- Ticket record
- What was promised in the chat
- Loose ends
How they run
- Triggered by an incoming message
- 3
- Run on demand
- 3
- Runs on a schedule
- 6
- Steps per run
- 3–5
- Credits per run
- 4–8
- Use a knowledge base
- 3
Where customer support fits in customer service
Most tickets are about one customer’s specific order or account. An agent that can read that state actually fixes the problem, instead of replying with a help article the customer already found.
Next Step
Deploying customer support agents
These run as-is against the inputs listed above. Most deployments adapt one — a different source system, a different tolerance, a different approval path. The first call establishes which.