Inputs that matter
Hard filters for non-negotiables—possession window, pet policy, parking count—should eliminate noise first. Soft scoring handles nice-to-haves like higher floor or lake view.
Capture preference updates from every WhatsApp branch and call summary so the model weights recent intent higher than stale form fills from six months ago.
Explainability builds rep adoption
Agents should see the top three reasons a listing ranked: price fit, commute score, inventory velocity. Trust drives adoption; black-box scores get ignored.
When inventory is thin, suggest adjacent communities transparently with tradeoffs instead of hiding distance with marketing language.
- A/B test ranking tweaks on holdout territories before global rollout.
- Log overrides when reps pin alternate units—those labels train better rules.
Connecting to site visits
When a ranked shortlist drives a booking, close the loop: store which ranked position they chose to refine future scoring for similar buyer personas.
For regulated topics (WhatsApp, email, ads), validate opt-in, record keeping, and regional rules with qualified counsel. This content is operational guidance, not legal advice.