- Business context
- A Texas property business serving English and Spanish enquiries.
- My contribution
- Conversation architecture, n8n orchestration, property routing and implementation.
- Implemented scope
- Persistent conversation context, structured qualification and a separate reply delivery layer.
Every enquiry starts somewhere different.
A property enquiry rarely arrives as a complete brief. A person might ask about a location, switch between English and Spanish, or return to an earlier conversation with a new question. The useful response depends on what has already been discussed.
I built a conversation system around that context. It connects incoming messages, property specific instructions, conversation history and operational records so qualification can continue as the enquiry becomes clearer.
The central design question was how to keep a natural conversation connected to the next action the business needs. That required both AI interpretation and explicit workflow state.
Follow the enquiry.
Start with what the person says.
A question may identify a property immediately or leave the intended property unclear. The incoming message and contact context enter the workflow together.
- Example message
- ¿Me puedes dar más información?
- Language
- Spanish
- Property intent
- Needs clarification
Keep the history as the route changes.
The workflow loads the conversation context and routes a known property to its qualification instructions. The same conversation can continue after the route is identified.
- Conversation
- Existing history retained
- Property
- Identified from the enquiry
- Instructions
- Selected for that property
Ask for the information that matters.
The reply and the operational fields have different jobs. One continues the conversation; the others record the next step, contact details and whether a person needs to take over.
- Reply
- Relevant answer and next question
- Working record
- Contact details and next step
- Escalation
- Explicit handoff fields
Give the next person useful context.
The workflow records the updated state and passes the reply to the messaging layer. A human handoff can carry the property context and the information already collected.
- Reply delivery
- Separate messaging layer
- Context for the team
- Property and collected information
- Next step
- Recorded for the workflow or person
Illustrative walkthrough. The descriptions and example records are recreated for this case study.
Give each part a clear responsibility.
- Conversation memory
- The workflow loads the relevant lead context and carries the conversation forward as property intent is identified. A route change does not require starting a new conversation.
- Property routing
- A general route handles unclear intent. Once the property is known, the conversation uses the corresponding qualification instructions.
- Structured next steps
- The AI response includes fields for qualification, contact details, follow up and handoff. Those fields connect the conversation to the surrounding workflow.
- Coordinated replies
- A per contact lock coordinates parallel work. Generated replies are passed to a separate messaging layer, while the workflow maintains operational state.
The work between the tools.
My contribution covered the connection between messaging flows, the n8n conversation engine, stored context and property qualification prompts. I designed the responsibilities of those parts and implemented the transitions between them.
The customer facing conversation was one part of the build. The surrounding work included deciding which information to retain, when to change the route, what the reply contract should contain and how another process or person could use it.
This is the kind of project where the interface can look simple while the decisions behind it require care. The system needs a consistent way to connect what a person says with what the business should do next.
A connected qualification process.
The implemented system brought bilingual messaging, property context, conversation memory and structured qualification into one defined process. It gave the workflow a shared understanding of the enquiry as the conversation progressed.
The case is supported by review of the implemented workflows, recorded execution activity and the prompt system. It describes the work I built and the behavior documented during that review. The walkthrough above is an explanatory example, with identifying information removed.
For a similar engagement, I would start with the actual questions customers ask, the information your team needs and the situations that require a person. Those details determine the first useful version and the checks it needs to pass.
Written by Younes Nadif