Customer support teams
AI for customer support teams, run as a weekly ritual
For the CX lead who answers for quality, not just volume. The assistant answers routine questions from your help centre, hands everything else to an agent with the thread attached, and your team reviews a sample of conversations every week, so a wrong answer is fixed in the article that caused it.
Who owns it
Who owns the workflow
The support or CX lead owns this workflow, with one named person looking after the knowledge base. Together they decide which topics the assistant may answer, approve changes to the articles and macros it answers from, and chair a short weekly review of sampled conversations. Agents stay in your helpdesk and see each handoff as a ticket with the conversation, the customer’s details and the articles the assistant tried. We build and tune the assistant, set up the review views and run the first reviews with you until the routine belongs to your team.
Jobs to be done
The jobs AI can take off the team
Answer the repeat questions
Delivery times, account access, plan changes and policy questions answered from your help centre, with the article cited so the customer can check the wording themselves.
Hand over with the whole thread
When a question is out of scope, emotional or still unresolved, the conversation becomes a ticket with a summary, the customer’s details and what was already tried, so the agent starts at step two.
Triage the queue before an agent opens it
Email and form tickets are tagged by topic, language and urgency so they land in the right view. Tagging is all it does here; nothing is answered on an agent’s behalf.
Draft replies for agents to edit
For topics the assistant should not answer alone, such as refunds outside policy, it drafts a reply from your macros and the order record. The agent edits, sends and owns the answer.
Log every gap
Each question the assistant could not answer is recorded in the customer’s own words, so the knowledge owner sees which article is missing, out of date or written in terms customers never use.
Stay in step with incidents
During an outage your team switches on an incident message. The assistant gives the approved update, links your status page and routes affected customers to a tagged queue instead of guessing.
Deflection that stuck
Conversations the assistant closed where the customer did not come back about the same issue within the window you choose. Read every week next to the reopen list, never on its own.Handoff completeness
Handoffs where the agent did not have to ask the customer to repeat anything. Agents tick one box on the ticket; the review reads the misses.Transcript review
A fixed sample of conversations each week, scored against your rubric: correct, cited, on tone and handed over at the right moment.Knowledge gaps closed
Questions logged as unanswerable and how many now have an approved article. The knowledge owner brings the oldest open gaps to the meeting.Wrong-answer reports
Answers an agent or a customer flagged as wrong, each traced to the article or instruction behind it and fixed there, not patched in the prompt.
KPI definitions
What to measure
- First response time, by path
- Time from a customer’s first message to the first reply, reported separately for the assistant and for agents, so an instant automated answer cannot hide a slow human queue.
- Time to resolution after handoff
- From the moment of handoff to the ticket’s solved state. It shows whether the attached summary and history actually shorten the agent’s work.
- Reopen rate on assistant answers
- Assistant-closed conversations that were reopened, or followed by a new contact on the same topic, inside the window you set during scoping.
- Satisfaction, split by path
- Your existing survey score, reported for assistant-only, handed-over and agent-only conversations rather than as one blended figure.
- Escalation reasons
- Why each handoff happened: out of scope, the customer asked for a person, low confidence, a policy exception or a frustrated tone. The mix tells you what to fix or build next.
- Article coverage
- Topics in your ticket history that have a current, approved article the assistant can cite. Coverage, not model tuning, usually decides how much the assistant can take on.
Use cases
Use cases for this team
- Answer the repetitive tickets. Hand the rest over with context.Ticket deflection with handoff
- Order status and returns, with refunds left to your teamOrder lookup, policy returns
- Support in your customers’ languages, checked by people who speak themLanguage-aware support
- Every call answered. Urgent ones put straight through to a person.Call answering and triage
Buying checklist
Before you buy, check these
- Does every answer show the customer which article it came from?
- Can a customer reach a person in one step at any point in the conversation?
- Does a handoff arrive in your helpdesk as a ticket with the full thread and a summary?
- Can your team switch a topic off without a developer or a redeploy?
- Are wrong answers fixed in the source article rather than patched in the prompt?
- Will you see the assistant run against a test set of your own past questions before launch?
- Is there an incident switch for outages and product problems?
- Is deflection always reported next to reopens and satisfaction, never alone?
FAQ
Frequently asked questions
Who on the support team should own the assistant once it is live?
Usually the person who already owns the help centre and macros, backed by the team lead. They approve content changes, decide which topics the assistant answers and run the weekly review. Naming a backup matters as much as naming the owner, because a review that skips two weeks is where quality starts to drift.
How do we stop deflection numbers from hiding unhappy customers?
By never counting a conversation as resolved just because it ended. A closed conversation only counts if the customer did not come back on the same issue within your window, and every report shows deflection beside the reopen rate and satisfaction for that path. The weekly sample deliberately includes resolved conversations, not only failures.
Our agents worry the assistant takes the easy tickets and leaves them the hard ones. Is that what happens?
Yes, the mix shifts towards harder conversations, and it is better to plan for that than to deny it. Handoffs arrive with the history so agents skip the fact-finding, and we suggest measuring handle time by topic rather than overall. Plan to revisit quality rubrics and targets after the first month for the same reason.
Do the weekly reviews need an analyst or a data team?
No. The review is a set of saved views in your helpdesk or a shared sheet: sampled transcripts, flagged handoffs and the gap log. A team lead can run it in a standing meeting. We build the views, write the scoring rubric with you and sit in on the first reviews.
Can the assistant keep our tone of voice and policy wording?
Your macros and articles become its source, and your tone guide is written into its instructions and checked in the test set. For policy topics such as refunds or warranties it quotes the approved wording instead of paraphrasing, because a paraphrase is where an accidental promise slips in.
What does the assistant do during an outage or product incident?
Your team flips an incident switch. The assistant then gives the approved status message, links your status page, stops answering questions related to the incident and routes affected customers to a tagged queue. Switching the assistant off entirely is a single toggle as well, with every new conversation going straight to agents.
How do you report results to our head of support?
A monthly one-page report using the definitions on this page, measured against the baseline period we agree before launch. It lists the knowledge gaps fixed and still open, and includes a few real conversations that went well and a few that did not. It contains no numbers we cannot trace back to your helpdesk.
Next step
Have a workflow in mind? Let’s scope it together.
Book a free 30-minute strategy call. Bring one process you want to automate and we will talk through scope, integrations and a sensible first step.
Free 30-minute call · no commitment