Act · Automations
Some work should never be left to a model.
An automation is a rule: when this event happens, if these conditions hold, do exactly this. Same input, same output, no temperature setting. Order confirmations, status replies, reminders and routing belong here — not in a prompt.
Illustrative demo with fictional data.
Why both
Deterministic rules sit beside the agent, not under it.
Scout is good at understanding a sentence. It is the wrong tool for sending the same confirmation message to every order. So the two systems run on the same conversation, with separate records, and neither can edit the other.
| Automation rule | Scout | |
|---|---|---|
| Starts when | A conversation event fires: a message arrives, or a conversation is created, updated, opened or resolved. | A customer's message matches a scenario you declared. |
| Decides how | Your conditions, evaluated literally. No ranking, no confidence score. | Retrieval over approved knowledge, scored, then checked against its sources. |
| Can be wrong by | You writing the wrong condition. Reproducible, and visible in the rule. | Missing knowledge or an unclear question — which is why it hands over instead of guessing. |
| You control it with | An on/off switch and the rule itself. | Scenarios, guardrails, approvals, evaluations and the run trace. |
A rule never calls Scout, and Scout never edits a rule. If both act on one conversation, both appear in its history.
The rule model
When. If. Then.
Three parts, built in settings, switched on or off per rule. Nothing hidden behind a wizard.
- When — the event
- Message created, or a conversation created, updated, opened or resolved. The event decides which conditions and actions are available.
- If — the conditions
- Message contents, inbox, team, assignee, status, priority, language, country, tags and custom fields, joined with AND or OR.
- Then — the actions
- Assign, tag, prioritise, apply an SLA, send a message or attachment, write a private note, call a webhook, snooze, resolve. Added in order, run together.
Events a rule can react to
- Conversation created
- Conversation updated
- Conversation opened
- Conversation resolved
- Message created
Which conditions and actions you can pick depends on the event. Sending on WhatsApp still obeys Meta's messaging rules.
New automation rule
Illustrative demo with fictional data.
One rule, end to end
Watch it fire.
The same rule, step by step, with the conversation changing beside it. Fictional store, fictional order, real rule shape.
The work
The jobs that pay for themselves.
These five are where a wrong answer costs money, and where a deterministic rule is worth more than a clever one.
- Order confirmationConfirm a COD order on WhatsApp before it ships, and cancel the ones nobody confirms.
- Order statusAnswer “where is my order” from the live order record instead of a person copying a tracking link.
- Cart recoveryOne message, inside the window, to the people who left something behind.
- Reminders and follow-upsAppointment reminders, delivery windows, renewal nudges — scheduled, not remembered.
- Lead qualificationCollect what sales needs before a human opens the thread, and route by the answers.
“accurately sends whatsapp, email and sms order confirmations, unsuccessful rate decreased from 30% to 7%”
That is a merchant describing their own order-confirmation automation, not a BlendDesk measurement. We publish no resolution or deflection figure of our own — your numbers will depend on your rules and your traffic.
Execution
A rule can call your systems, not just your inbox.
An action can send a webhook event or call a tool that hits an HTTPS endpoint you control, so the automation reads and writes the system of record instead of guessing at it.
- Shopify
- Order lookup, fulfilment status, cancellation — the order record itself, not a cached summary.
- ERP
- Stock, invoices and delivery windows from the system your operations team actually trusts.
- CRM
- Create or update the contact, attach the conversation, route by owner.
- Custom HTTPS
- Any endpoint you can expose: bearer token, API key or basic auth, HTTPS only, size and time limits enforced, private network addresses refused.
Tool call · run 2291
200 OKIllustrative demo with fictional data.
Limits
What an automation may not do on its own.
Automations are deterministic, which makes them fast and makes a mistake repeat perfectly. The same controls that bind Scout bind them.
- Touch money without a person
- Refunds, cancellations and anything that moves a balance go to an approval queue with their exact arguments, and expire if nobody decides.
- Message outside the window
- Outside the 24-hour customer service window, only templates Meta has approved can send. A rule cannot route around that.
- Send marketing as utility
- Template category is Meta's call, not a field you set to save money. Rules that send are checked against the category the template was approved under.
- Run without a record
- Every firing is on the conversation: which rule, which conditions matched, what it did and when.
Outside the 24-hour customer service window, only approved templates send.
An Egypt marketing template costs $0.0644 and a utility template $0.0036 — about 18× the price for the same characters.
Category routing is the biggest per-message lever a merchant controls, and it is decided by what the template says, not by the rule that sends it.
Next
Keep reading
- IntegrationsShopify, ERP, CRM and your own HTTPS endpoints.
- Guardrails and approvalsWhat needs a person before it executes.
- ScoutThe agent that handles the sentences a rule cannot parse.
- Order confirmationThe highest-value automation for a COD business.
- AnalyticsWhat fired, how often, and the conversations behind it.
Questions
Automations, answered plainly.
What can a rule react to?
Five events: conversation created, updated, opened or resolved, and message created. Each event exposes its own conditions and actions.
Can a rule message a customer?
Yes, with the send-message action. On WhatsApp the 24-hour customer service window applies, and outside it only templates Meta has approved will send.
Do rules and Scout fight over a conversation?
No. A rule reacts to events and acts literally; Scout reacts to what a customer said. Both leave their own entries in the conversation history, and neither can modify the other's configuration.
Can a rule call my backend?
Yes — a webhook event, or a tool that calls an HTTPS endpoint you control with bearer, API-key or basic auth. The request, its arguments and the response are recorded on the run.
What happens when my endpoint is down?
The call is recorded as failed with its error, the automation stops at that step rather than inventing a result, and the conversation goes to a person.
Is the walkthrough on this page real data?
No. Nile Home, its customers and its orders are fictional. The events, conditions and actions are the ones in the product.
Bring your three worst repetitive jobs.
We will build the rules for them in the demo — the events, the conditions, the actions and the approval step — and show you the record each one leaves.