89% Match Accuracy.
From Buyer Conversation to Agent-Ready Lead.
Agix built Properti AI's Property Discovery Automation, a six-stage conversational AI pipeline that extracts buyer lifestyle needs, enriches listings with live neighborhood intelligence, and delivers semantically ranked property matches with 89% accuracy. Buyers find their home in 4.2 viewings. Agents spend zero time on qualification.
The all-in-one AI marketing platform built for real estate networks that can't afford to look different in every suburb.
Properti AI powers multi-office real estate agencies across Australia and the UK, combining AI content creation, smart scheduling, brand governance, and performance analytics into a single platform that gives head office control without strangling local agents. Built for franchise networks where every office needs to move fast and every post still has to look like it came from the same brand, Properti AI eliminates the tension between consistency and autonomy that has plagued real estate marketing for decades. When Properti AI approached Agix, the brief was precise: build the buyer-facing intelligence layer, a conversational discovery system that could match buyers to properties before agents spent a minute on qualification, turning every inbound inquiry into a ranked shortlist rather than a cold conversation starting from zero.
Nine connected products. One platform. Runs on autopilot.
Properti AI isn't a single tool, it's the whole marketing stack for real estate networks. Grader, Social, Ads, Reviews, Studio, AI, Autopilot, Insights, and Franchise work as one connected system that creates, publishes, advertises, and measures marketing across every office simultaneously.
Agents spending 60% of their week qualifying buyers who already knew what they wanted, they just couldn't say it.
Real estate discovery was a human bottleneck problem disguised as a data problem. The listings existed. The buyer preferences existed. The gap was a translation layer that could turn a conversation about lifestyle into a ranked shortlist of properties, without a single agent hour spent in between.
Buyers described their dream home in lifestyle language. Listings were indexed by spec.
Buyers describe their ideal home in lifestyle language, good schools, short commute, backyard for the kids, but listing portals only filtered by bedrooms, price, and suburb. Agents spent the first 45 minutes of every buyer relationship manually translating that gap. 58% of qualification time went to questions with objective, database-answerable answers that required no human judgment at all.
Buyers viewed 18+ properties before purchasing, most of them wrong from the first showing.
18 properties viewed before purchase wasn't buyer indecision, it was evidence that recommendation systems were inaccurate. Agents matched buyers based on memory and recency bias, not real-time data. Buyers who viewed 18+ properties reported the lowest settlement satisfaction scores: the exhausting journey tainted the outcome, even when they found the right home.
Lead conversion was collapsing at the first follow-up, buyers ghosted before agents could qualify them.
Inquiry-to-buyer conversion sat below 30%, almost entirely explained by response latency. A buyer who enquires at 9 PM Sunday and hears back Tuesday morning has already scrolled 60 listings and booked a viewing with a competitor. The first-mover advantage is enormous, and it was being surrendered entirely to whoever had an agent available at the moment of inquiry.
Six stages. One conversation. From "what matters to you" to agent-ready shortlist.
Conversational Onboarding → Intent Profile Construction → Listing & Neighborhood Enrichment → Semantic Match Engine → Recommendation Delivery → Agent Handoff & Feedback Loop, every stage compounds on the last, building a buyer profile that gets smarter with each response.
Six pipeline stages that turn a buyer's first message into a ranked shortlist an agent can act on immediately.
Each stage operates as an independent module and as part of a unified context pipeline, passing a buyer profile object that accumulates preference signals, enriched listing data, match scores, and agent-readiness flags from first message to final handoff.
Conversational Onboarding
A buyer-facing chat interface that opens with a single open question, "What's most important in your next home?", and extracts structured requirements from natural-language responses. When a buyer describes schools, commute, bedrooms, budget, and outdoor space in a single message, the system parses all five dimensions simultaneously. Buyers who complete onboarding convert to viewing-ready leads at 3.4× the rate of those using traditional search filtering.
Intent Profile Construction
Transforms conversational responses into a structured, weighted buyer profile, combining explicit requirement extraction, implicit preference inference, lifestyle scoring, and priority weighting. A buyer who mentions "the kids" and "schools" without mentioning a backyard still gets a yard-size signal, because lifestyle context implies unspoken preferences. Profiles are versioned so agents can see exactly what drove each match score.
Listing & Neighborhood Enrichment
Augments every listing in real time with six neighborhood dimensions: school ratings, walkability score, commute time, crime context, live listing feed, and 5-year price appreciation. A property listed at 9:00 AM is matchable by 9:04 AM, giving agents a first-mover advantage on their own recommended stock. No nightly batches, no stale matches.
Semantic Match Engine
Scores enriched listings against buyer profiles using semantic similarity, a property 5% over budget but with superior school and commute scores can outrank a cheaper listing that fails on lifestyle. Every recommendation shows component scores (School Fit, Commute Fit, Neighborhood Fit, Budget Fit) so buyers and agents understand the reasoning. Diversity balancing prevents the shortlist from clustering in a single suburb.
Recommendation Delivery
Presents the ranked shortlist with match percentages, component score breakdowns, and daily new match alerts, turning discovery from a one-time search into an ongoing relationship. Buyers can signal disagreement with any score, updating their profile without repeating onboarding. Buyers who received daily alerts converted to viewing-ready status 19 days faster on average.
Agent Handoff & Feedback Loop
Delivers every buyer to an agent as a fully profiled lead, readiness score, preference breakdown, shortlisted properties, and CRM sync included. Agents spend 68% less time on qualification because every question has already been answered. Viewing feedback loops back into the model, reranking the shortlist and compounding match accuracy with every interaction.
Every agent gets a live visibility score — and knows exactly how to improve it.
Properti AI gives each agent a single number, their Overall Visibility Score, broken down across AI search presence, social reach, and review performance. It's not a vanity metric. It's a ranked leaderboard position that drives referrals, listing wins, and inbound enquiries. Agents who improve their score by 10 points see an average 23% increase in inbound listing appointments.
AI search visibility tracked in real time
As buyers increasingly use AI-powered search to find agents, Properti scores how well each agent surfaces in those results and surfaces specific actions to improve.
Social reach scored across every channel
Properti measures posting consistency, engagement rate, and audience growth across Facebook, Instagram, and LinkedIn, giving agents a single social reach score instead of four separate dashboards.
Review velocity and sentiment analysed automatically
Review recency, volume, and response rate feed directly into the agent's score, with AI-generated response suggestions so agents never leave a review unanswered.
Every agent gets a visibility score. Every office stays on brand. Every network runs on data.
The buyer discovery pipeline feeds directly into Properti AI's agent intelligence layer, giving individual agents a live visibility score across AI search, social reach, and reviews, while head office maintains brand consistency across every office in the network without reviewing every post. Local teams move fast. The platform keeps everything on brand.
What happened when buyers stopped scrolling through 18 listings and agents stopped asking questions they already knew the answers to.
Measured across Properti AI network deployments spanning multi-office agencies in Australia and the UK in the 12 months following full platform launch.
Increase in inbound inquiry-to-active-buyer-relationship conversion rate following deployment of the conversational onboarding and instant recommendation pipeline. The driver was response latency: buyers who received a personalised shortlist within 60 seconds of submitting an inquiry converted at 3.1× the rate of those who waited for agent follow-up. The 156% lift came almost entirely from the night and weekend inquiry population, leads that previously went cold before Monday morning, now engaging with a platform that never stopped being available.
Proportion of properties in AI-generated shortlists that buyers rated as "genuinely relevant" after viewing, versus 34% for manually curated agent shortlists in the same markets. The accuracy improvement reflects the enrichment layer's ability to score school catchment, commute times, and walkability against each buyer's stated thresholds, dimensions agents could match only approximately, based on memory and local knowledge, not real-time data lookups across every listing simultaneously.
Reduction in time agents spend on buyer qualification calls per lead, from an average of 47 minutes to 15 minutes. The remaining 15 minutes are spent on judgment work that machines cannot do, nuancing the shortlist based on the buyer's emotional response to specific properties, identifying hidden motivations that affect the purchase decision, and building the relationship that drives referrals. Agents reported higher job satisfaction and stronger buyer relationships as a result of spending qualification time on conversation rather than data gathering.
Down from an industry average of 18 properties before the conversational match layer was deployed. The reduction reflects match precision: buyers who receive a shortlist generated from their actual preference profile rather than a keyword-filtered portal result set attend fewer viewings because more of the properties they view are genuinely candidates. The 4.2 figure includes properties viewed for comparison, buyers who purchased the first property they viewed accounted for 31% of all transactions in the dataset, up from 6% before AI-assisted discovery.
Our agents used to spend the first hour of every buyer relationship asking questions the buyer had already answered on the inquiry form. Now they spend it on what they're actually good at, reading the buyer, reading the market, and closing. The AI did the groundwork. The agents do the deal. Lead conversion is up 156% and our top agents are handling 40% more buyers without working longer hours.
Four decisions that made 89% match accuracy possible, and repeatable.
Semantic matching beats filter logic, lifestyle signals outperform hard specs
The foundational architectural decision was to abandon filter-based matching in favour of semantic scoring. Filter logic asks: does this listing meet these criteria? Semantic matching asks: how well does this listing fit this buyer's entire preference profile? The difference is enormous in practice. A property 5% over a buyer's stated budget but in a suburb with schools rated two points higher than their threshold, a commute 8 minutes shorter, and a walkability score 15 points better will outperform in match accuracy, and in eventual purchase satisfaction, over a property that passes all three hard filters but underdelivers on unmeasured lifestyle dimensions. 89% match accuracy was not achievable with filter logic. It required building a scoring system sophisticated enough to model the tradeoffs buyers actually make when they're in the field.
Real-time enrichment prevents stale matches, data freshness is a product requirement, not a nice-to-have
A semantic match engine is only as good as the data it scores against. Enriching listing databases in nightly batches meant that a property listed at 9 PM on Monday wouldn't enter the match pool until Tuesday morning, a 12-hour window during which high-fit buyers might have already seen it on portal, inquired through a competitor agent, and committed to a viewing. Real-time enrichment, live listing feed, continuous school data sync, commute-time API calls per listing per buyer on demand, closed that window to under 5 minutes. The accuracy advantage over nightly-batch competitors wasn't just in match quality. It was in freshness: AI-matched buyers saw new fit properties before manually-browsing buyers in the same market, giving Properti AI agents a consistent first-mover advantage on their own recommended stock.
Implicit preference inference surfaces the requirements buyers don't know to state
The highest-value innovation in the intent profile layer was inferring preferences that buyers didn't explicitly state. A buyer who mentions "the kids" and "school" but doesn't mention outdoor space almost certainly wants a yard, the lifestyle context makes it predictable. A buyer who mentions being "close to the city" as their top priority but also mentions "weekend markets" probably means walking access to amenities, not just commute time, the system infers a walkability weight from the second signal that the first didn't surface. These implicit preference signals account for 31% of the eventual match score in purchases where buyers reported "I didn't know I needed this until I saw it." The onboarding stage was designed to elicit enough lifestyle context to make these inferences, not just to capture the explicit requirements a search form would have asked for.
Feedback loops compound accuracy, every viewing makes the next recommendation better
The 89% accuracy figure is not the accuracy at launch, it's the accuracy at month six, after the feedback loop had processed thousands of viewing outcomes into preference model refinements. At launch, match accuracy was 74%. Accuracy at month one was 79%. The compound improvement came from the feedback mechanism: when buyers attended viewings and provided signals (liked the street, not the layout; would stretch the budget for a better school score; didn't realise how much street noise bothered them), those signals updated their preference profiles and recalibrated the weighting of each dimension in future matches. At scale, the model learns that "proximity to parks" is strongly correlated with buyer satisfaction in families with young children, and that "five-year price appreciation" is more predictive of purchase intent than stated when buyers mention "investment potential" in onboarding. These correlations are invisible to any individual agent and only emerge from population-level viewing and purchase data.
What powers this system.
Conversational AI
Natural-language buyer onboarding, intent extraction, and preference profiling, turning unstructured conversation into structured buyer profiles that drive semantic matching and agent handoffs.
Semantic Search & Matching
Vector-embedded semantic scoring that matches buyer lifestyle profiles against enriched listing databases, producing ranked shortlists with per-dimension explanations that agents and buyers can interrogate.
Real-Time Data Enrichment
Live integration with school ratings, commute time APIs, walkability indexes, crime data, and price appreciation models, augmenting listing databases in real time so matches are always scored against current neighborhood intelligence.
Buyer Intelligence & CRM Integration
Structured buyer profile handoffs with readiness scores, preference breakdowns, and shortlisted properties, synced to CRM systems so agents start every buyer relationship with complete context rather than cold qualification calls.
Feedback Loop & Model Improvement
Continuous preference model refinement through viewing outcome data, match accuracy that compounds from 74% at launch to 89% at month six, as the system learns which dimensions most reliably predict purchase satisfaction from population-level behaviour.
PropTech & Real Estate AI
Sector-specific AI for real estate networks, from buyer discovery automation to agent performance analytics and franchise brand governance, built for multi-office agencies that need scale without sacrificing local relevance.
Common questions about building buyer discovery AI that improves with every transaction.
Partial onboarding produces partial profiles. The system generates recommendations with the available signals rather than blocking delivery until a complete profile exists, but lower profile completeness produces lower match confidence scores, which are visible to buyers and agents. A buyer who provides price, suburb, and bedroom count but no lifestyle context receives a recommendation set that performs closer to traditional filter logic (around 60% perceived relevance) than the full semantic match (89%). The recommendation interface shows which preference dimensions are unset and invites buyers to complete them, but never gates the experience behind a mandatory form. Most buyers who receive recommendations with low match confidence on specific dimensions complete the missing signals within 24 hours of receiving the first shortlist, because seeing what a more complete profile would unlock is a stronger motivator than being told to complete a form before you see anything.
Preference evolution is built into the model. Buyer profiles are versioned, not fixed, when a buyer provides viewing feedback ("loved the suburb, the house was too small") or explicitly updates a preference, the profile recalibrates and the entire shortlist reranks in real time. Agents can also update buyer profiles directly in the handoff dashboard when a qualification call reveals that the stated budget ceiling was softer than indicated, or that the commute threshold wasn't as fixed as the onboarding suggested. The system treats preference signals as hypotheses with varying confidence levels rather than hard constraints, early-session signals carry less weight than post-viewing feedback signals, which in turn carry less weight than the agent's direct qualification assessment. Profiles that have been refined through three or more viewing feedback cycles show 91% match accuracy on subsequent recommendations, versus 79% for new profiles with no feedback history.
The discovery pipeline integrates with major real estate CRM systems (Rex, Salesforce Real Estate Cloud, Agentbox) via API, pushing completed buyer profiles as new leads with enriched custom fields for preference dimensions and readiness scores. On the listing side, the enrichment pipeline connects directly to the listing portals and agency listing management systems through their APIs, pulling live listing data, triggering enrichment calls to school, commute, walkability, and safety data providers, and writing enriched listing records back to the indexed database within minutes of publication. For agencies that use proprietary or legacy CRM systems, the handoff API produces a standard JSON buyer profile that any system with REST API access can consume. The onboarding and recommendation layer deploys as an embeddable widget that can be placed on any agency website or listing portal with a single script tag, no CMS integration or technology upgrade required to get buyers into the discovery funnel.
Match accuracy is measured as the proportion of properties in AI-generated shortlists that buyers rated as "genuinely relevant" when asked after attending a viewing, defined as "I would seriously consider purchasing this property if the price was right." This is a stricter standard than "I was interested enough to attend" (which would produce higher numbers) and a more practical standard than "I actually purchased it" (which would produce lower numbers). The 89% figure is an average across all buyers with at least three viewing feedback data points, in all markets where the platform has been deployed for more than 90 days. At launch, before the feedback loop has generated sufficient signal to calibrate the model, accuracy for new deployments is typically 72–76%. The accuracy improvement trajectory is predictable: agencies that deploy consistently see 2–3 percentage points of accuracy improvement per month for the first six months, after which the curve flattens as diminishing returns on additional feedback signal set in at around 89–91%.
The 68% reduction in qualification time is entirely reallocated to higher-value agent work, it doesn't translate to fewer agents. In every deployment, agents who adopted the AI-assisted workflow handled 35–40% more buyer relationships per month without increasing their hours, because the 47-minute average qualification call compressed to 15 minutes and the 3.4-hour shortlist preparation process (reviewing listings, checking school catchments, mapping commute times) was replaced by the AI's real-time output. Agent earnings increased in 84% of cases where performance bonuses were commission-linked, because more buyer relationships at higher conversion rates with lower qualification overhead produces more transactions per agent per month. The agents who reported the highest satisfaction increases described the same change: they were finally spending their time on the parts of the job no database could do, reading the room, building trust, navigating the emotional complexity of the largest purchase their buyer would ever make.
Ready to take buyer qualification off your agents' plates entirely?
Most projects go from kickoff to deployed AI system in 8–16 weeks. Let's talk about what a 156% lead conversion lift, 89% match accuracy, and 4.2 average properties viewed before purchase could mean for your network.