Building chatbot flows

This page covers how the pieces of a chatbot flow fit together: intents that recognise what people want, parameters that pull details out of their messages, replies with buttons, a fallback for everything else, and the nodes that fetch data or hand over to a person. For canvas basics such as adding, connecting and grouping nodes, see Building flows.

Plan the conversation firstCopy Anchor Link

Before you add nodes, list:

1. What people will ask for. Each distinct request becomes an intent, for example "order a pizza", "track my order", "opening hours" and "talk to a person".
2. What you need from them. Details such as a pizza size or an order number become parameters on the intent, or fields in a Form when there are several.
3. Where the answer comes from. Fixed text goes in a Reply; live data comes from an API call or a Data Store; open questions can go to an AI Agent with your knowledge base.
4. What happens when the bot cannot help. A fallback reply, and a Live Chat handover if your team takes chats.

Trigger
 ├─ OrderPizza    → Reply (size buttons)
 ├─ SelectSize    → Data Store (save order) → Reply (confirmation)
 ├─ TrackOrder    → API (order status) → Reply
 ├─ TalkToPerson  → Live Chat
 └─ Fallback      → Reply

Connection rulesCopy Anchor Link

A chatbot flow starts at its single Trigger node. The canvas refuses connections that break these rules:

IntentsCopy Anchor Link

An Intent node describes one thing people ask for. When a message arrives, Buni.ai compares it with the sample utterances of your intents and runs the one that matches best.

The Intent panel has three tabs:

Below the tabs, Output › Output variable names where the intent's result is stored. See Where the values go.

Writing good sample utterancesCopy Anchor Link

• Add at least 10, worded differently: questions, statements, short and long, with and without typos your users make.
• Give each intent its own phrases. The same or overlapping utterances on two intents make matching unpredictable, and validation flags them.
• Keep each utterance on one line and under 200 characters.
• Mark a detail with a placeholder in braces, such as I'd like a {pizzaSize} pizza. The name must match a parameter on the same intent. Use {slotName}, not ${slotName}, in utterances.
• Only one intent in the whole flow can have an utterance that is nothing but a placeholder, such as {pizzaSize}.
• Do not use the reserved names CancelIntent, FallbackIntent, HelpIntent, StartOverIntent or StopIntent.

ParametersCopy Anchor Link

A parameter pulls one detail out of the message, such as a date, a number or a city. Select Add Parameter on the Parameters tab and fill in:

A Custom Parameter also needs a type name and sample values with synonyms, for example large with synonyms big and family size. Its Resolution Strategy is either Expand Values (accept variations and synonyms) or Restrict to Slot Values (accept only the values and synonyms you defined).

Use one parameter type at most once per intent.

Where the values goCopy Anchor Link

When an intent matches, its parameter values are saved to state under the intent's Output variable. If you leave it empty, Buni.ai uses the intent name without spaces.

For an intent OrderPizza with output variable order and a parameter pizzaSize, later nodes read the size as {{state.order.pizzaSize}}. See Variables for how to use values in text and conditions.

RepliesCopy Anchor Link

A Reply node sends one or more messages. On the Response Content tab, add elements in the order they should be sent:

The Preview tab shows roughly how the reply will look.

Buttons and quick repliesCopy Anchor Link

Each button has Button Text (what the person sees), Button Value (what the bot receives when they tap it) and an optional Target Intent. Quick replies have the same three fields as Reply Text, Reply Value and Target Intent.

Set Target Intent to send a tap straight to that intent instead of matching the tapped text against all intents. This is how you continue a conversation after a Reply, since an Intent node cannot be connected after a Reply.

Channel limits apply when the message is sent:

If you need more than three choices on these channels, split them across several Reply elements.

Replies built from dataCopy Anchor Link

On the Data-Driven tab, turn on Enable Data-Driven Responses to send one message per item in a list, for example a list of products returned by an API call. Set Data Source Variable to the variable that holds the list, map the fields (Title Field, Subtitle Field, Image URL Field, Button Text Field and so on), and choose a Message Grouping: Individual Messages, Carousel or List.

Handle messages the bot does not understandCopy Anchor Link

Add an Intent node whose only sample utterance is Fallback, connect it to the Trigger, and connect it to the reply you want unrecognised messages to get. Buni.ai routes every message that matches no intent there.

Without one, unrecognised messages get a built-in English reply, "Sorry, I don't understand that. Can you try again?", and validation shows a No Fallback Intent warning.

Good fallbacks tell the person what the bot can do, offer buttons for the most common requests, or hand over to a person with a Live Chat node.

Collect several detailsCopy Anchor Link

To ask for several values in a row, such as name, email and preferred date, use a Form node rather than a chain of intents. A form keeps the conversation on its own fields until the person finishes, validates each answer, and lets them review before submitting. Its answers land in {{state.variableName}}, where variableName is the form's output variable.

Work with data and AICopy Anchor Link

After an intent, use any of these to do the work:

For failed API calls and other errors, see Error handling.

Hand over to a personCopy Anchor Link

Connect an intent (for example TalkToPerson) or a fallback path to a Live Chat node to pass the conversation to your team in the Inbox. A Live Chat node ends the bot's part of the flow. See Live chat handoff.

Common patternsCopy Anchor Link

Question and answer

Trigger → Intent (OpeningHours) → Reply

Look something up

Trigger → Intent (TrackOrder, parameter orderNumber) → API → Router
  ├─ found     → Reply (status)
  └─ not found → Reply (check the number)

Confirm with buttons

Trigger → Intent (OrderPizza) → Reply (buttons Small / Medium / Large, Target Intent: SelectSize)
Trigger → Intent (SelectSize) → Data Store (save) → Reply (confirmation)

Answer from documents, then escalate

Trigger → Intent (Fallback) → AI Agent (knowledge base)
Trigger → Intent (TalkToPerson) → Live Chat

Checks before you publishCopy Anchor Link

The canvas validates the flow as you build. Fix anything it reports before publishing. The chatbot-specific checks include:

• The Trigger has an Introduction Message and at least one platform.
• Every Intent has a unique Intent Name, at least one sample utterance, and no utterances shared with another intent.
• Every required parameter has Elicitation Prompts, and parameter names are camelCase.
• Every Reply has at least one response element, and each element has content.
• There is a fallback intent.
• A Live Chat node has an agent group and a transfer message.

RelatedCopy Anchor Link