Automations
An automation is a flow: something happens, and the steps you set follow. You build it on a canvas, publish it, and see per run what it did.
Having a flow generated
You don't have to build a flow node by node yourself. In the builder you describe what should happen in plain language, and you get a plan first: a summary, the trigger, and the steps in order. Confirm the plan and the flow appears as a draft on the same canvas described below.
Nothing is saved or switched on until you press that button yourself. You check the draft, ask for a change if you want one, and then turn it on with the same button you'd use for a flow you built by hand.
What starts a flow
Every flow has exactly 1 starting point. There are 4 to choose from.
- A keyword in an incoming message. You choose whether the message has to contain the word, start or end with it, match it exactly, or match a pattern you write.
- A booking from FareHarbor: created, changed, cancelled, or rebooked.
- A webhook. The flow gets an address of its own that another system can post to.
- A new contact. This fires once, the moment the contact is first created.
The steps you hang behind it
You drag a step onto the canvas and connect it to the previous one. These are the groups.
| Group | Steps |
|---|---|
| Message | Send text, send media, send a template, ask a question with up to 3 buttons, ask a question with a list |
| Wait | Wait a duration, wait until a moment or until your next opening hours, wait for a reply, wait for a webhook |
| Branch | A condition on a field, on your opening hours, a split across variants, or an AI step that picks a label or pulls data from the message |
| Contact | Save the sender as a contact with labels, or opt a contact out |
| Hand over | Assign to a colleague, either a fixed person or in rotation |
| External | Call another API and store the answer as a variable |
Waiting and timing
A wait can be relative, say 2 hours after the previous step, or aimed at a moment. With a booking you can anchor the flow to the start or the end of the activity, for instance 1 day after it finishes.
There's also a step that waits until you're open again, so a question that lands on Saturday evening gets answered Monday morning. Opening hours and time zone are set per organization.
Branching
A condition splits the flow in two. You can compare against the customer's last answer, the content of the message, the contact's name, or a field from the event that started the flow.
You can also branch on your opening hours, or split contacts randomly but consistently across variants, which is how you put two versions of a message against each other.
AI step: labelling, pulling out data, or answering
This step reads whichever message is relevant at that point and does one of three things. Classify picks exactly one label from a list you define, say booking, complaint, or general question, and the flow branches on that label. Extract pulls fixed fields out of the message, such as an order number or a date, and stores them as a variable for later in the flow. Respond writes the text of the reply itself, from the knowledge you entered in Settings. Classify and extract never send anything to the customer on their own; a reply lands in the inbox as a draft by default and only goes out by itself once you switch that on per step.
Every AI step has a required fallback branch for when it doesn't work out: a slow reply, an unclear answer, or a limit reached. The flow just continues on that branch instead of getting stuck. For replies that also covers the answer not being in your knowledge: the step invents nothing and takes that same branch.
Publishing and testing
Publishing runs a check over the flow. Disconnected steps, missing fields, and any path that would send plain text outside the 24-hour window are blocked, with the reason attached.
You can switch a flow on for 1 phone number first. It then runs for that number only, and the flow is marked as being in test mode.
A flow can be tied to specific WhatsApp numbers. Tie it to none and it applies to all of them.
Seeing what happened
Per flow you get its runs: which are in progress, which finished, and which stalled. Open one and you get the steps in order, with the error attached where something went wrong.
Build on a computer: the canvas relies on dragging and connecting, which doesn't fit a phone screen. On a phone the app shows a message instead of the builder. The list of flows and their runs are readable anywhere.