Condition: AI classifier (route by intent)
The AI classifier node reads the customer's last message and routes it into one of several categories you define — perfect for triaging inbound to Sales, Support, or Renewals.
Watch this step in the video · Respond with AI and AI classifier
What it is and when to use it
The AI classifier (route by intent) is a condition node that asks an AI model to classify the customer's last message into one of several categories you define, then sends the run down the matching branch. It isn't a chat — it's a single "classify and route" call.
It's perfect for triaging WhatsApp (or any channel) inbound into Sales / Support / Renewals, or for labeling a lead's intent before the rest of the flow runs. The model handles language natively: you define the labels (usually in your own language) and the AI classifies the customer's text whatever language they wrote in.
Each classification consumes AI tokens from your plan's monthly AI allowance and, once that runs out, from the top-up AI wallet (recorded in the AI usage ledger), just like other AI features. Keep that in mind for high-volume flows.
Requirements
- The workflow engine is on all plans (from Lite), with a monthly run quota; AI usage draws down your plan's monthly AI allowance and then the AI wallet.
- A conversation with at least one inbound customer message to classify.
Steps
- Go to Settings → Workflows, open the workflow, and in the Nodes tab expand Conditions.
- Drag AI classifier (route by intent) onto the canvas and connect it after the node that receives the message.
- In the inspector, add your categories. For each one set:
- Label (e.g. Sales) — this becomes the output name.
- Hint (optional but recommended) — describe what belongs in that category (e.g. "pricing questions, demos, plans").
- Adjust the options:
- Minimum confidence (0–1, default 0.6): if the AI doesn't clear that threshold, the run takes the No match branch.
- Save to variable (optional): store the chosen category in a variable (e.g.
ai_category) to reuse later. - Include history: when on, the AI considers the last few inbound messages to disambiguate sequences like "hi" → "I have a problem with a charge".
- AI agent / model: optionally reuse an existing AI agent's model.
- Wire one branch per category plus the fallback No match branch.
- Click Save.
Verify
Run a Test with a sample message and check the execution timeline: you'll see the chosen category and its confidence. If you saved the category to a variable, it'll show in the variables panel.
If it doesn't work
- Everything lands on No match: lower the minimum confidence or improve each category's hint — the clearer they are, the better it classifies.
- It misclassifies: avoid overlapping categories (e.g. "Support" and "Issues"). Define mutually exclusive labels.
- No AI usage / no response: confirm you still have balance in your plan's monthly AI allowance or in the AI wallet; if both are used up, top up the wallet so the node can classify.
Tips
- Start with 3–4 clear categories and a good hint on each; add more only when the flow needs it.
- Save the category to a variable and reuse it for Add tag or Hand off to an agent instead of classifying twice.
Related articles
Related articles
-
Build your first workflow
Omnifox workflows automate your operation: a trigger starts the automation and action nodes do the work. Build a welcome message in minutes.
-
Workflow triggers catalog
Learn which events can start a workflow automatically in Omnifox.
-
Workflows complete guide
Workflows in depth, in 54 chapters: triggers, nodes, variables, testing, history, templates, import and export, with every setting explained and a real workflow end to end.
-
Action nodes available in workflows
Discover the actions a workflow can run automatically.
-
Conditions and branching (if/then) in a workflow
Make your flows take decisions and follow different paths based on each contact data.