How do I set up my organization from zero?

The order that works

Teams that start with the channel and work backwards end up with a bot that talks before it knows anything. Start from who you are, then what the bot knows, then where it talks.

Write your Global Mind

Global Mind is your organization’s shared brain. Every bot you ever create reads it. It has four parts:

• Identity. Who you are, what you sell, who your customers are. Two or three paragraphs.
• Brand voice and tone. How you speak. Formal or warm, short or thorough, whether you use first names. Give an example of a good reply.
• Prohibited topics. What the bot must never discuss or promise: legal advice, competitor comparisons, pricing exceptions, whatever applies to you.
• Escalation rules. Situations where the bot should ask a person even if it is confident: a threat to cancel, a mention of a lawyer, an angry customer.

There is also a section for how the bot should communicate with your own team during consultations, including which language to use internally. Your customers can be served in one language and your team briefed in another.

Create a bot

Under Bots, create a bot and fill in its Agent Brain: its instruction (what this bot is for), its limitations, and its work methods. Set the confidence threshold; the default of 80 is a good start. A bot dedicated to one job, such as support for one product line, is easier to tune than a generalist.

Give it knowledge

Create one or two Knowledge Banks and attach them to the bot. See the article on giving the bot knowledge for the details. Do not wait for a complete library; the bot will tell you what it is missing through its consultations, and the learning run will propose additions.

Open a channel

Add a Web Chat channel and attach the bot. Style it in Chat Appearance, paste the snippet on a test page, and talk to your bot. Only once you are happy with how it answers on chat, add email and WhatsApp.

Invite your team

Under Users, invite your people. Admins configure everything. Senior Agents manage knowledge and learning and handle conversations. Agents handle conversations. Each person sets their availability in the console so consultations route to people who are actually there.

Set your Conversation States

Under Conversation States, define the buckets your team will use to sort Live Conversations, and switch on the auto-move triggers. If you connect HubSpot later, each state can map to a pipeline stage.

Send the first message

Open your test page, ask a real question, and watch Live Conversations. Mark the conversation Always learn and add a note about anything the bot should have done differently. Run Update learning now and review what it proposes. That loop, repeated for a week, is how a first bot becomes a good one.

What to look at each morning

Open the Dashboard: how many conversations the bot resolved on its own, how many needed help, and how many are waiting. Open Learning and approve or discard the overnight suggestions. Ten minutes a day is enough.