AI chat tenants can use, without the hallucinations
To date, Omfax has never provided a chatbot. So, we have been‘ horizoning’- thinking about what the best version of an AI Large Language Model chat might provide.
TENANTS want to use their own words, their own language, and their own channel— but need reassurance that a person is just one step away. Landlords need to know that AI won’ t dream up answers but follow policies and procedures.
The safe bot, or the usable one
Most landlords have looked at two options and liked neither. The self-service diagnostic decision tree is auditable and updateable, but some tenants pick the wrong answer because they don’ t know what a“ gully” is.
An AI chatbot is wonderfully fluent. But without rails, it will tell a tenant a repair is the landlord’ s responsibility when your policy says otherwise. And when the complaint arrives, there is nothing to review but a log of something invented on the spot.
What if you didn’ t have to choose? Your diagnostic scripts own the outcome— the questions, the branching, the priority, the service raised. The AI owns the understanding: it works out which of your answers the tenant meant. It is never asked to produce an answer, only to choose among those on offer— and where nothing corresponds it says so and asks again. There is no room for hallucination and no need for training.
What that feels like
Part way through a repair the tenant is asked:
Main drain( Manhole) · gully · wastepipe soilpipe / toilet, and the tenant types“ what’ s a gully?”. They have used the word- but they are asking, not choosing.
AI should answer—“ A gully is typically a drainage channel which collects surface water and directs it away from an area.”
And maybe show an image. Then re-present the four options. Nothing is recorded; the question is still open. No penalty for asking, no three-strikes limit.
Then:“ oh yeah, it’ s a gully.” Now the tenant is choosing, and the conversation continues. Tenants are allowed not to know a word, without needing a member of staff and without being routed somewhere wrong.
In their own language
All tenants answer identical questions— same priority, same works order, but in their own language
Using their favourite comms channel
The same conversation should run on your website chat, WhatsApp, Facebook Messenger, Instagram Direct and voice chat via telephone. Each gets what it expects— reply buttons in messaging, spoken prompts on the phone— and photos and voice notes are welcome.
Typing into the chat is always available: buttons are a convenience, never a replacement.( One caveat: social messages will cross Meta’ s infrastructure before reaching any diagnostic, unlike website and voice chat).
Landlord Policy Understanding
A tenant should be able to ask what the landlord’ s policy is on anything. The chat should provide brief summarised answers with cited URL links off to the published policy if the tenant wants to dig deeper.
Easy escalation to Staff
An advisor should always be one step away Asking for a person works at any point and is never obstructed— no maze, no need to fail three times first.
The tenant doesn’ t have to ask: Chat should hand over to a person by itself
• in an emergency
• if the tenant sounds distressed or frustrated
• when a conversation goes round in circles
• when the enquiry falls outside your scripts.
The hand-over is a supervised transfer with re-ring and position announcements, not a blind push. If nobody picks up, the call becomes a tracked callback promise rather than a dropped caller. And whoever answers receives what has already been gathered, so the tenant doesn’ t start again from scratch.
Blue Sky Thinking
Confession time: We have gone beyond‘ horizoning’. All these features are ready to be delivered as part of Keyfax V5 in early 2027.
Contact Omfax for more information or a demo: sales @ omfax. co. uk
Intelligent software made simple
96 Issue 28