The problem explored
Routing an incoming support conversation between escalation and FAQ-assisted responses.
A conversational support flow with classification, FAQ context and escalation branches. The supplied diagrams document the configured connections, rather than verified production outcomes.
Scroll sideways to inspect the diagram. Zoom enlarges labels; Reset zoom restores this preview.
Architecture summary · read alongside the source
- Receive a message
- Classify the request
- Escalate a branch
- Use FAQ context
Inside the processing logic
Receive a message
- A WhatsApp Trigger starts the diagram.
- Customer Log and an If branch lead into sheet updates and field preparation.
- Set Fields prepares data before classification.
Classify the request
- An AI Agent Classifier connects to a Gemini model.
- An output parser feeds an If node.
- The exact classification labels and routing condition are not shown.
Escalate a branch
- One branch sends a WhatsApp message.
- Slack Send Message follows on that branch.
- A HubSpot create-or-update-contact node is also visible.
Use FAQ context
- The other branch reads rows before an AI Agent Responder.
- The responder has Simple Memory and a SkincareFAQs Sheets tool.
- Parsed output routes to WhatsApp, HubSpot and a sheet update.
Step through the explanation
Dennice, an original illustrated robot guide, walks through the diagram. Start with Next, or choose autoplay. Captions and a full transcript are included; sound is optional.
Source evidence
These are the supplied local files used to write this explanation. Node labels and connections are visible; exported workflow JSON and execution logs were not supplied.
Visible outputs
WhatsApp responses, contact updates and escalation or logging steps.
These outputs are represented by nodes in the source. Delivery success and business results require execution evidence.
What this case study does—and doesn't—verify
SkincareFAQs is a visible tool label, not evidence of a client relationship. Response quality, escalation thresholds and production use are unverified. The case study describes the visible routing and integrations only.
Next validation steps
Before operational use, I would inspect the workflow export, confirm credentials and permissions, test representative inputs and error paths, and review the outgoing messages with the intended user. These are proposed checks, not completed workflow execution tests.