Enquiry to viewing
A sales platform for a property brokerage, built around an AI agent that works the top of the funnel. Portal enquiries are captured and scored, the conversation runs in WhatsApp and email, viewings are booked against real consultant calendars - and every step is visible and switchable.
Sector
- Residential property
- Off-plan and ready units
Platform
- Web workspace
- 11 destinations, 3 groups
What we did
- Workflow modelling
- UX + UI design
- Front-end engineering
Channels
- Voice call
- Web + social capture
What problem does it solve?
Property enquiries arrive from portals, a website and social channels at every hour, and they go cold fast. The bottleneck is almost never the consultant's ability to sell - it is the hours between an enquiry landing and a human seeing it.
An AI agent that works the top of the funnel, in the open
AQARI puts a named assistant on the first contact: it deduplicates and scores the lead against live inventory, opens a WhatsApp thread about the exact listing the buyer asked about, answers questions from a project knowledge base in English or Arabic, and offers viewing slots read from the consultant's calendar. The consultant appears once - at the viewing, briefed.
Why do most property leads never get a conversation?
Three things break the top of a property funnel, and none of them are solved by a faster CRM form: the first reply arrives too late, the qualifying questions are repetitive but not scriptable, and the calendar is the real bottleneck.
What can you actually click?
Eleven destinations across three groups: the sales workspace, the AI agent's own surfaces, and inventory. The walkthrough runs an enquiry from capture through conversation to a booked viewing, then opens the machinery behind it.
How did we build it?
An assistant that touches customers is only adoptable if the team can see what it will do before it does it. That constraint shaped the build order more than any feature list did.
What does an AI-assisted sales workflow actually need?
Six rules came out of building this one, and they hold for any business where an assistant handles first contact - property, clinics, education, high-ticket services.
- Ground every answer in your own inventory. An assistant that improvises a price or an availability date does more damage in one message than it saves in a month.
- Draw the workflow where the operator can see it. Trust comes from a visible map with switches, not from a promise that the automation is sensible.
- Make every stage individually reversible. Teams adopt this by turning on one step at a time; an all-or-nothing sequence gets turned off entirely at the first bad message.
- Detect language, do not ask for it. In a multilingual market the reply rate follows the language of the buyer's first message.
- Book against the real calendar. Offering slots that a consultant has to confirm afterwards reintroduces exactly the delay the automation removed.
- Hand over with a brief, not a transcript. The human step is worth protecting - so what reaches them should be budget, intent and timeline, already extracted.
What does the interface look like, and why?
One rule drives the whole palette: a person must always be able to tell at a glance what the assistant did and what a human did. Brand blue carries the product, and a warm amber is reserved exclusively for AI activity - the agent's badge, its highlighted metric, its rows in the activity feed. The messaging surface keeps WhatsApp's own colours instead of restyling them, because a thread that looks like the channel it came from is easier to trust.
Aa
Primary typeface
Instrument Sans
A B C D E F G H I J K L M N O P Q R S T U V W X Y Z
a b c d e f g h i j k l m n o p q r s t u v w x y z
0 1 2 3 4 5 6 7 8 9 & ? ! £ $ €
30pxAED 50.41MHeadline figure24pxGood afternoon, SarahH218pxSection headingH316pxPage title and lead copyH1 / Lead14pxBody text, rows and message bubblesBody12pxLEAD SOURCE · AI SCORELabelThe type scale is a fixed nine-step ramp from 11px to 30px held in design tokens, which is why dense tables and conversational bubbles can sit on the same screen without either one shouting. Scores, response times and currency are set in a monospaced face so a column of figures can be compared vertically. Instrument Sans and IBM Plex Mono are open-licence typefaces; the specimen above falls back to the nearest available face if they are not installed on your device.
Building an AI-assisted sales platform
The questions brokerages and sales leaders ask us first, answered plainly.
Ask us yoursHow long does it take to build an AI sales assistant like this?
A first production release at this scope - multichannel capture, a grounded conversational agent, calendar booking and a visible workflow - is a matter of months, not weeks. The conversation design and the knowledge grounding take longer than the interface. What moves the date most is how well structured your inventory data is, since that is what the assistant has to answer from.
Will it say something wrong to a customer?
That risk is exactly why the platform is built the way it is. Answers are grounded in the project's own inventory and knowledge base rather than generated freely, the workflow is visible with a switch on every stage, and a consultant can take over any thread at any point. You start with the safe steps enabled and widen the scope as you see the transcripts.
Does it integrate with WhatsApp and property portals?
Those are the assumed channels. Capture is modelled from property portals, a website form and social messaging, and the outbound conversation runs over WhatsApp with an email fallback and an optional voice call. In a real rollout the portal and messaging integrations are the substantial engineering; the workspace on top of them is well-trodden.
How much does a platform like this cost?
It depends on scope, and a feature list alone is not enough to quote from honestly. The main drivers are the number of channels and portals to integrate, whether voice is in scope, and how much of your inventory and document library has to be prepared before the assistant can answer from it. Shanti Infosoft scopes it on a short call and returns a fixed quote within 48 hours.
Does this replace the sales team?
No, and it is not designed to. Everything before the viewing is repetitive work that scales badly with headcount; the viewing, the negotiation and the relationship are not. The point of the workflow map is to make that division explicit - the human step is marked, and it is protected.
Related work
Leads going cold before anyone replies?
Tell us how your business actually works and we will tell you honestly what it takes to build. See more of our work or what we do.
Book a 15-minute call