Imobiox logo

Designing the Rules of a Shared Real Estate Network

Company
Imobiox
Role
Product Designer
Year
2026

Imobiox dashboard with pending items, portfolio numbers and partnership requests

The dashboard, opening with what is waiting for the broker.

Overview

Imobiox connects builders, agencies and independent brokers in a shared property network. Each one manages their own inventory, opens it to partners, and sells properties that belong to someone else.

I designed the MVP. The screens were the smaller part of the work. Most of it was deciding the rules: what two people who have never met need to know before they agree to sell the same property, what each of them sees when the other one changes something, and where AI actually removes work instead of adding a feature.

The Challenge

The real estate market still operates in a fragmented way, with disconnected systems, rework and little integration between the people involved in a sale. Three problems shaped the work:

  • Every partnership has two sides. The same property lives in the owner’s inventory and on the partner’s website. Any change made by one of them has a consequence for the other.
  • Trust before the first conversation. A broker asks a stranger for permission to sell. Without the commission, the terms and the rules in front of them, the request is a shot in the dark.
  • AI with a reason to exist. Brokers register properties from memory, after a visit, into a long form. AI had to earn its place by solving that kind of problem.

My Role

I worked as the product designer on Imobiox. I was responsible for:

  • The end-to-end flows: account creation and CRECI verification, dashboard, inventory, property network, sale permissions, clients, website settings and listing portals
  • The partnership rules written into the interface: what each side sees, what every status change does, and who gets notified
  • The AI features: property pre-registration by audio and the AI Studio
  • The component library, with variants for the property card, the network card, the client record and the filters, plus the empty, error and no-permission states
  • Working concepts in code for the flows that had to be used to be understood

How I Worked

I designed the base product in Figma, then ran a second round with Claude to go deeper into the parts that were still thin: notifications, the life of a partnership, status changes and the AI flows. Every proposal in that round is documented against the screen it comes from. It either evolves it, replaces it or reuses it as it is, so nothing was redesigned without a reason.

The property photos and the illustrations were generated with image models from prompts I wrote. That way the screens carry realistic content, not gray boxes.

For the flows where timing and feedback matter more than layout, I built working concepts in code and shared them as deliverables.

Pre-registration concept ↗ Sale permissions concept ↗

Key Decisions

1. Show the terms before the request

In the first version, a broker could ask for permission to sell a property without knowing the conditions. I redesigned the request as a drawer that puts everything on the table first: who is responsible for the property, the commission as a percentage and as an approximate value, the term that applies, and who keeps the leads that arrive through the partner’s website. Below that, a short list explains what happens if the request is approved.

The button says what the broker is doing: accept the conditions and request.

Drawer for requesting sale permission, showing commission, term and what happens if approved

The request drawer, with the conditions of the partnership before the action.

2. No answer is a dead end

A request can be pending, declined or approved, and each state got its own design instead of a status label:

  • Pending shows the timeline, who the broker is waiting for and where the answer will arrive, with the option to cancel.
  • Declined shows the reason written by the owner and, right below, similar properties available in the network, each one with its commission.
  • Approved confirms that the property is already in the broker’s inventory and website, repeats the agreed conditions and links to both places.

Declined request showing the owner’s reason and similar properties in the network

Approved request with the timeline, the conditions and links to the inventory and the website

A declined request points to similar properties. An approved one points to the next step.

3. A partnership is a relationship, not a permission

Approval is where the partnership starts, so I designed what comes after it. The inventory separates the broker’s own properties from the ones they sell in partnership, and each card says how many partners are selling it and how many requests are waiting.

Ending a partnership explains its consequences before the owner confirms: the property leaves the partner’s website, the partner is notified, and the leads the partner already received stay with them. The owner can add a reason, and the form says that the partner will see it.

When a property is sold by someone else, the partner sees who sold it and when, keeps their contacts for that property, and gets similar properties from the network to offer those clients.

Inventory with tabs for own properties and partnership properties

Modal for ending a partnership, explaining what happens to the partner

Property detail after being sold by a partner, with the broker’s contacts preserved and similar properties suggested

The inventory, ending a partnership, and what a partner sees when the property is sold by someone else.

4. Status changes say what they do

Changing a property from Available to Reserved looks like a small action, but it affects the website, the network, the listing portals and every partner. I moved the status change to the card and made the menu describe the effect of each option. Available is applied directly. Reserved, Sold and Suspended ask for confirmation, because other people are affected.

Publication follows the same idea. Each channel has its own line with its own state: the broker’s website, the network and each portal. When a portal rejects a listing, the error says why and offers the fix in the same place.

Publication tab with the state of each channel and a portal error with its fix

The publication tab, channel by channel.

5. One feedback system, defined by rules

With partnerships, a lot happens while the broker is not looking: a request arrives, a price changes, a property is reserved by someone else. I replaced scattered alerts with one system made of four parts: pending items on the dashboard, a notification center, toasts and transactional emails.

Then I wrote the rules. Each event has one main channel, and critical events are also sent by email. Toasts only confirm an action the user just took. Pending items disappear on their own once the broker acts, and the center keeps the history.

Notification center grouped by day with partnership, contact and portal events

The notification center, grouped by day.

6. Registering a property by voice, in a way brokers can trust

The AI pre-registration replaces the first pass of the form with an audio note. I went through two versions of this flow. The second one is built on four ideas:

  • One modal, three states. Before recording, recording and reviewing the audio happen in the same place. Recording only starts when the broker taps the button.
  • A script for the person, not for the model. The modal lists what is worth mentioning (type and neighborhood, price, rooms and area, highlights, address, condo fee and tax) and confirms that the microphone is working.
  • Processing that shows its work. While the draft is being built, the modal shows what the AI has already understood and the transcript. The broker can close it and is notified when the draft is ready.
  • A way out when it fails. If the audio is too short, the product says exactly what it identified and offers three paths: record again, continue with what was identified, or register manually.

The result is a normal registration form. Every field filled in by the AI is marked, the audio and the transcript stay next to it, and a checklist shows what is still missing before publishing.

Modal before recording, listing the points to mention and the microphone status

Processing step showing what the AI already understood and the transcript

Insufficient information state with three ways to continue

Property form pre-filled by AI, with missing items highlighted

Before recording, processing, the failure state and the draft the broker reviews.

7. An AI Studio built on what already exists

The AI Studio groups the tools that turn a registered property into marketing material: social media posts, photo enhancement, property video, virtual staging, descriptions, and stories and captions. Every tool starts by choosing a property from the inventory, so nothing is described twice. Generation runs in the background and the results are kept in one place.

Photo enhancement and virtual staging are the same features available during registration, reused here instead of rebuilt. In photo enhancement, the broker chooses what to improve and compares before and after. Nothing replaces the listing photos until they decide to use the result.

AI Studio with tools for posts, photos, video, staging and descriptions

Photo enhancement tool with a before and after comparison

The AI Studio and the photo enhancement tool.

Design Foundations

The interface is dark by default, with a single green accent reserved for primary actions and positive status. Empty states, no results, loading errors and missing-permission screens were designed as part of the system, not screen by screen, which keeps the product predictable when something is not there.

What’s Next

After the MVP, each new initiative is prioritized by its potential to generate value for the ecosystem, increase engagement across the network and prepare the platform for a national expansion.

Next case · Design system

Scaling a design system across ~20 product teams