Large support organisations write rollout plans for AI because they have to: hundreds of people whose jobs change, a union of habits to move. A two-person team doesn’t need the plan. It needs the same five conversations, had quickly and honestly, before the agent goes live — because the alternative is discovering on Thursday that one of you thinks the agent handles refunds and the other has been answering them by hand.
Here they are, in the order they come up.
1. Say plainly what changes on day one
The agent takes the first reply.That’s the whole change, and it’s worth saying out loud, because everything else follows from it. Every conversation on the widget and on WhatsApp gets an answer from your content within seconds. Your team’s inbox now shows the ones marked Waiting on human, the tickets the agent filed, and the conversations someone chose to take over. Nobody is answering “where’s my order” any more unless the agent couldn’t.
Two switches make it real. Automatic replies (Settings → Agent) is the agent itself; off, customers can still write in and nobody gets an automated reply. Live on your site (Settings → Widget) is whether the widget exists on your website at all. Decide who flips them and when, and say so.
2. Reset what a working day looks like
Before the agent, the day was answers: type, send, next. After it, the day is exceptions and upkeep, and the second half is the part teams forget to plan for. Somebody has to read the Worklists, rewrite the document that keeps getting asked about, and prune the handoff list — and if that time isn’t set aside, it never happens, because the inbox always feels more urgent.
Make it explicit: an hour a week, named, on the calendar. The return on that hour is measurable — it’s the resolution rate on whichever intent the hour was spent on — which is more than can be said for most hours in support.
3. Make the numbers shared, not managerial
Anyone with a seat in Anchorcan open the Analytics page, and it should be a habit rather than a report. When the resolution rate is a number the whole team sees, the agent stops being “the owner’s tool” and becomes the thing everyone is trying to make better. The soft-escalation count in particular belongs to the people fixing documents, not to whoever bought the plan.
In practice: once a week, five minutes, the Overview tab open. What resolved, what escalated, what’s flagged under Content gaps. Then the hour from step 2 has a target.
4. Give the team the controls, not just the outcome
A rollout goes badly when the humans feel like the agent’s cleanup crew. It goes well when they can steer it. Anchor puts the steering in the inbox: Take over to step in mid-thread and Return to Anchor AI to hand back; internal notes the customer never sees; Summarize chat for a briefing on demand; tickets routed to a team rather than a person. And Copilot — a private assistant that reads the conversation and your knowledge and suggests, but never messages the customer or changes anything. It’s advice for the person, not a second bot.
The instruction that matters most is small: when you take over and the answer you type is a fact the documentation should have had, write it down afterwards. That’s how the front line improves the system. It’s the only mechanism that scales, and it’s a two-minute habit.
5. Say where this goes for the people
For a large team the rollout question is “what happens to the headcount”. For a small one it’s “what happens to me” — is the person who answered the inbox now redundant, or now something else? Answer it: the person who used to type the answers now owns the knowledge base and the handoff line. It’s a different job, and a better one. The dashboard’s roles — owner, admin, agent — let you give that person the access that says so.
A note on the trial
Anchor’s free trial is one seat, so the team conversation above happens after you subscribe, not during the trial. That’s the right order anyway: the trial is for the owner to point the agent at the help center and see what it answers; the plan is when the rest of the team joins. Two Starter seats is a two-person team with a shared pool of 1,000 credits a month, and each seat adds its credits to that pool.
Start with three questions
Sit down with the other person and ask:
- Which questions should the agent answer, and which should always reach one of us?
- Whose hour is the improvement hour, and when?
- What’s the number we look at together, and how often?
If you can answer all three, the rollout is done. The rest is switching it on.
Common questions
How does a support team's job change after an AI agent is rolled out?
The agent takes the first reply, so the team stops typing the documented answers and works the exceptions: conversations marked waiting on a human, tickets, and anything they choose to take over. The second half of the job — reading the analytics, rewriting the documents that keep getting asked about, pruning the handoff list — needs its own protected time or it doesn't happen.
Who should own the AI agent in a small business?
One named person, usually whoever already answers the inbox. Their job becomes the knowledge base and the handoff line: a weekly look at what escalated, a document rewritten, a topic added or removed. Everyone with a seat can see the numbers; one person is accountable for moving them.
Can my teammates take over a conversation from the AI agent?
Yes. In Anchor's inbox, Take over lets a person step in mid-thread with the transcript, an AI summary, and any collected details in front of them; Return to Anchor AI hands the conversation back. Internal notes stay invisible to the customer, and tickets can be routed to a team rather than an individual.
Can I add my team during the free trial?
No. The trial is one seat, 30 days, and 300 credits, meant for the owner to test the agent on their own content. Teammates join once you're on a plan; every seat adds its credit bundle to one shared monthly pool.