An agent is not a chatbot
A chatbot answers from a fixed list. An agent understands the request, searches your data, decides a next step, then executes or hands over.
In property the next step is usually: qualify, book a viewing, or hand over to an agent. If the system does none of these, it is a reply bot, not an agent.
Four agent types
| Type | What it does | Risk |
|---|---|---|
| Qualification agent | Asks budget, area, timing | Too many questions repel the buyer |
| Matching agent | Suggests suitable units from inventory | A wrong match loses trust |
| Booking agent | Finds a viewing slot in the agent's calendar | Booking an unavailable time |
| Follow-up agent | Chases those who have not decided | Annoyance and opt-outs |
Most brokerages need the first and third first — they are closest to the result.
What the agent must know
- Available units, prices and payment plans.
- Location and surroundings (schools, roads, services).
- Delivery terms and dates.
- Approved FAQ answers.
- Limits: what it must never answer.
Boundaries: when it hands over
The most important design decision is not what the agent knows but when it stops. Rule: any question about final price, negotiation, contracting or a complaint goes to a human immediately.
Anything outside its knowledge base is handed over rather than guessed. Guessing in property is expensive: a wrong price can lose the buyer permanently.
How to measure it
| Metric | What it measures | Target |
|---|---|---|
| Share of qualified chats | Qualification quality | Improving monthly |
| Handover rate | Boundary accuracy | Low and justified |
| Time to first reply | Speed | Seconds |
| Customer sentiment | Conversation quality | No complaints |
| Completed bookings | Scheduling | Above 70% |
Common mistakes
- Training it on everything with no boundaries.
- Hiding that it is automated — discovery breaks trust.
- Not connecting it to inventory, so it quotes stale prices.
- Leaving it unmonitored in month one.
- Measuring messages instead of qualified leads.
Questions for a vendor
- Where does it read prices and availability, and how are they updated?
- What does it do with a question outside its knowledge?
- How does it hand over, and is the handover logged?
- Can we review conversations and edit answers?
- What is the running cost if volume doubles?
- Is our data isolated from other clients?
In short
A good agent is not the cleverest talker but the most precise at stopping: it knows what it knows and hands over what it does not.
Start with a qualification and booking agent, connect it to inventory, and review the first hundred conversations yourself.