Boei finishes the job, not just the conversation
Most chatbots stop at "let me connect you to someone". Boei prepares the quote, books the slot, updates the CRM, fires the workflow and sends the follow-up. Every action logged, every step reviewable.
The difference
A chatbot deflects. Boei resolves.
- Answers, then dumps into a contact form
- "Sorry, I can't help with that"
- Human still has to prepare the quote, book the slot, update the CRM
- Team measures "deflection rate", not resolution
- Answers, then calls your APIs to actually do the thing
- Reads your CRM, writes back to it, triggers Zaps, books calendars
- Human reviews outcomes, not tickets
- Team measures resolution rate and revenue captured
Six concrete outcomes. Not one of them is "generate a lead form".
Every action below is a real AI action running against a real API of yours or a pre-built template. Configure per workflow, gate with approval where it matters.
Boei asks for the details it needs (size, quantity, location, tier), pulls prices from your catalog, applies your rules, and sends a written estimate inline. No back-and-forth with sales.
Reads open slots from Google Calendar or your booking API, offers times that suit the customer, writes the event, sends the invite. Never double-books.
Creates or updates the contact in HubSpot, Pipedrive, Salesforce, or your own DB. Deal stage, custom fields, notes, tags. Attaches the transcript. Real-time.
Fire a Zapier zap, a Make scenario, a webhook to your ops stack. Anything downstream of "we just qualified a lead" runs itself.
Confirmation emails, quote PDFs, next-step reminders. Boei drafts and sends through your transactional email provider or a custom endpoint.
Fetch an order status, check stock, look the customer up in your CRM, hit any HTTP endpoint. Answer with the real answer, not a canned one.
Five steps between question and outcome.
The AI reads the visitor's message, decides which action fits, gathers the parameters, calls the tool, reads the result, and replies. All in-conversation, no context loss.
- 1Understand. Boei parses the intent using your training sources + house rules.
- 2Choose the tool. Picks the right AI action from your configured toolset.
- 3Gather parameters. Asks the visitor for anything missing, or pulls from context.
- 4Call the endpoint. HTTP request with your placeholders, credentials, headers.
- 5Reply with the outcome. Reads the response and writes the message the visitor sees.
POST https://api.your-app.com/quotes
Content-Type: application/json
Authorization: Bearer {{credential:API_KEY}}
{
"product": "garden_room",
"size_m2": {{input.SIZE}},
"customer_email": {{context.email}}
}Chain actions into a playbook.
When resolving a case takes more than one call, wrap the sequence in a workflow. Boei loads the playbook, executes each step, waits for approvals where you asked, and reports back when the whole thing is done.
- 1Ask qualifying questions (industry, team size, use case)Auto
- 2Check if the domain already exists in HubSpotAuto
- 3Create or update the contact recordAuto
- 4Send custom offer (discount > 10%)Approval
- 5Book the intro call in the AE's calendarAuto
- 6Notify sales in Slack with the transcriptAuto
Steps run in order. Boei pauses at step 4 if a human approval is required, then resumes without the visitor noticing a gap. Set the approval gate per tool or per workflow.
Start with a template. Or build your own from scratch.
Pre-built AI actions get you live in minutes for common cases. When you need something custom, the underlying generic HTTP tool talks to any endpoint you point it at.
Any REST API works. GET, POST, PUT, PATCH, DELETE. Placeholder injection for visitor context, live input, and encrypted credentials. Salesforce, Pipedrive, Airtable, Notion, your internal ERP, whatever's behind an HTTP endpoint.
- look_up_orderAuto
- check_availabilityAuto
- create_hubspot_contactAuto
- apply_discountApproval
- issue_refundApproval
- escalate_to_humanLive
Automation without the "wait, it did WHAT?"
Every action Boei takes is scoped, logged and reversible where possible. You decide what runs on its own and what waits for a human click.
- Per-action approval flag (safe tools auto-run, sensitive tools wait)
- Per-workflow overrides for stricter playbooks
- Every tool call recorded: inputs, response, human overrides
- Credentials stored encrypted, referenced by name only
- Test any action in the sandbox before it touches customer chats
"The AI actually answers customer questions from our own content instead of dumping them into a generic form. Setup was under an hour and it's been converting visitors to WhatsApp chats every day since."
Stop measuring deflection. Start measuring resolution.
Every conversation ends with a completed action or a warm handoff. Never a dead-end.