Spaces

First-week checklist for a firm trying an AI agent on client chats

This checklist walks a professional or technical firm through the first week of letting an AI agent answer client and enquirer messages. It covers facts, rules, a trial run and checking the chats.

Runs in your browser. Nothing you enter is sent anywhere or saved.

Before you start

A firm cannot afford a clumsy launch. Clients are on live matters, and one careless reply can sit in a file for years. The safest start is small: a short list of facts, firm rules for handovers, and a week of watching every conversation before you loosen anything.

Treat each line below as a task with an owner. Spread them across the week so nobody is overloaded, and keep a note of what you changed. By day seven you should know what the agent handles well, and where your own team is still needed.

Tick each one as you go. 0 of 10 done.

[ ] Write one page of plain facts: services, hours, location and fees policy.
[ ] List what the agent may never answer, and who takes those chats.
[ ] Name a handover person for each team and a backup.
[ ] Add your ten most common client questions with the answers you give.
[ ] Decide whether replies are held as drafts for a person to check at first.
[ ] Test with five invented messages, including a tricky advice question.
[ ] Tell your regular clients how to reach you and who is replying.
[ ] Check every conversation at the end of days one to three.
[ ] Fix the answers you spot and add what was missing.
[ ] Hold a short review on day seven with the whole team.

A realistic first week

Spend the first two days on facts and rules, with almost no live chats. On day three, let the agent answer only one kind of message, such as appointment questions. Read everything it says at the end of each day. If something sounds off, change the written answer, not just the one reply.

By the end of the week add a second kind of question, such as document receipts. Keep advice, quotes and complaints with your people throughout. Widen the scope slowly, and only after a clean run of days.

Common first-week mistakes

These are easy to avoid with a little planning.

  • Giving the agent a long handbook instead of a short, clear page.
  • Forgetting to say who covers lunch breaks and holidays.
  • Switching on every kind of message at once.
  • Reading only the handed-over chats, and skipping the ones it answered.

FAQ

How long before I can trust the agent with routine questions?

Most firms learn a lot in one week of reading every chat. Trust comes from repeated good replies on narrow topics, not from a date on the calendar.

Do I need to tell my clients about the agent?

It is good practice and often expected. A short line saying that an AI agent helps with routine questions, and that a person is always available, is usually enough.

What if my team is too busy to review the chats?

Give one person twenty minutes a day for the first week. That small time now saves you from a bigger problem later, and the review gets lighter over time.

More free tools for professional and technical service firms

Read the guide: An AI agent for professional, scientific and technical firms.