Step 1 (days 1–2): choose one category and record its baseline
Choose the ticket category with the highest volume and the lowest stakes, typically shipping status, password reset instructions, or a how-to question with a documented answer. Record its baseline from the helpdesk: tickets per week, first-line resolution rate, median first response including out of hours, and escalations. Note satisfaction where you measure it.
Resist starting with several categories. One category with a clean record is worth more at day 30 than three with a muddled one, and the second category is faster because the habits transfer.
Step 2 (days 2–4): confirm the knowledge base and connect the helpdesk
Read the knowledge base articles for the category and fix what is wrong or stale, because the Customer Support Digital FTE will cite them and the fact gate treats them as sources. Connect the helpdesk in Setup with a dedicated account scoped to the category's queue where the tool allows, with permission to draft but not to send until a class is promoted.
Set the approved sources for replies: the knowledge base, the policy pages, and nothing else. A reply that cannot cite one of them will escalate rather than guess.
Step 3 (days 4–5): write the always-escalate list and the summary standard
Write the always-escalate list into the remit before the first draft: refunds, credits, and billing disputes; account access and identity; complaints and requests for a manager; legal or safety language; any sign of distress; and anything the sources do not cover. Route each category to the person who owns that judgement. Then write the escalation summary standard: what was asked, what was found, why it did not resolve, a suggested next step.
This list is fixed. It does not shrink as the category earns autonomy, and its presence is what makes gating one category safe to relax later.
- Always human: refunds, account access, complaints, legal, distress
- Confidence triggers: no source found, conflicting sources, unclear intent
- Summary standard: asked, found, why not resolved, next step
- Routing: each category to the person who owns it
Step 4 (week 2): approve cited replies and check every escalation
From the first morning, drafts for the category arrive with the article cited. Review means opening the article, confirming it says what the reply says, checking the tone, and approving; send back with a reason in a sentence when it is wrong. Read every escalation summary this week and send back any that arrived as a raw ticket, because the summary standard is a remit rule the FTE must meet.
Keep the category at Level 0 all week. The record being built is the only evidence week four can use.
Step 5 (week 3): fix what the send-backs point at
Read the send-back reasons grouped. Replies wrong because an article is wrong or ambiguous point at the knowledge base; fix the article and the next reply is right. Replies wrong in tone point at the Brand Brain or the remit's tone rule; fix it once. Escalations that recur for the same missing answer are the FTE's proposed articles waiting in Approvals; approve the good ones and first-line resolution for those questions rises next week.
The send-back rate for the category should fall visibly between week two and week three. If it does not, the category was not as low-risk or as well-documented as it looked, and that is worth knowing.
Step 6 (week 4): review and the auto-send decision
Hold the first monthly review from the FTE's own report: first-line resolution for the category, median first response, escalations and their quality, the approval rate with reasons, and usage. Decide whether the category's record is long and clean enough to promote to auto-send within bounds, such as only when a source is cited and never when the ticket mentions anything on the escalate list.
Promote or deliberately do not, log the decision with its reason, and expect to step it back on the first send-back. Then choose the second category, which will take less than four weeks because the escalate list, the summary standard, and the habits already exist.
What does done look like at day 30?
Done is one category resolved at first line with a cited source on every reply, an escalation path that reaches your own team with summaries, a knowledge base that improved from the tickets that exposed its gaps, a written always-escalate list that has not moved, and a promotion decision made from the record rather than from optimism. The comparison against the baseline is read at the second review, when the first month's setup is out of the numbers.
A support lead who did this spent minutes a day on approvals and an hour a week on reasons and articles, and has a team that watched the first category succeed before the second one arrived.