60% weekly adoption
60% weekly adoption
OF BETA AGENTS USING AI-RECOMMENDED ACTIONS
OF BETA AGENTS USING AI-RECOMMENDED ACTIONS
$40K ann. value per agent
$40K ann. value per agent
reactivating an average of 5 trusted clients
reactivating an average of 5 trusted clients
INTRODUCTION
INTRODUCTION
INTRODUCTION
Agents wanted to show up for their clients through life’s biggest moments, not just the day they buy or sell a home. But the platform gives them no way to act on what they tracked, leaving them to carry every client’s story in their own head, hoping to catch a life event before someone feels forgotten.
Agents wanted to show up for their clients through life’s biggest moments, not just the day they buy or sell a home. But the platform gives them no way to act on what they tracked, leaving them to carry every client’s story in their own head, hoping to catch a life event before someone feels forgotten.
PROBLEMS AGENTS FACED
PROBLEMS AGENTS FACED
PROBLEMS AGENTS FACED
40% of consumers
40% of consumers
DON'T USE THE SAME AGENT, LOSING $8K IN COMMISSIONS.
DON'T USE THE SAME AGENT, LOSING $8K IN COMMISSIONS.
9k spent yearly
9k spent yearly
on duplicated software already provided
on duplicated software already provided.
Discovering the problem
Discovering the problem
Discovering the problem
This came from agent platform feedback: agents wanted clearer help turning what they track in notes, calls, texts, and emails into action, specifically lead scores that resurface clients at the exact moment they're ready to change homes.
This came from agent platform feedback: agents wanted clearer help turning what they track in notes, calls, texts, and emails into action, specifically lead scores that resurface clients at the exact moment they're ready to change homes.
"I've been using Follow Up Boss for a long time — not having recommendations on how to stay connected with my clients’ lives on Compass CRM is a bit disappointing.
Recommendations on the 5 D’s:
Diamonds
Diapers
Diplomas
Divorce
Death
The activity log can help with this, but it's hard to easily know what note I put that information in."
"I've been using Follow Up Boss for a long time — not having recommendations on how to stay connected with my clients’ lives on Compass CRM is a bit disappointing.
Recommendations on the 5 D’s:
Diamonds
Diapers
Diplomas
Divorce
Death
The activity log can help with this, but it's hard to easily know what note I put that information in."
"I've been using Follow Up Boss for a long time — not having recommendations on how to stay connected with my clients’ lives on Compass CRM is a bit disappointing.
Recommendations on the 5 D’s:
Diamonds
Diapers
Diplomas
Divorce
Death
The activity log can help with this, but it's hard to easily know what note I put that information in."
32
32
32
HOW MIGHT WE...
HOW MIGHT WE...
HOW MIGHT WE...
free agents from remembering every client detail, and use CRM data to resurface trusted clients with actionable recommendations, right when they're ready to buy or sell?
free agents from remembering every client detail, and use CRM data to resurface trusted clients with actionable recommendations, right when they're ready to buy or sell?
free agents from remembering every client detail, and use CRM data to resurface trusted clients with actionable recommendations, right when they're ready to buy or sell?
TALKING WITH TEAMS
TALKING WITH TEAMS
TALKING WITH TEAMS
When sitting with 5 different brokerage teams, we noticed that different teams and agents all vet leads in their own unique way. This made the solution not a clear button-to-button experience, but a dynamic one that needed artificial intelligence learning how each agent or team uniquely works.
When sitting with 5 different brokerage teams, we noticed that different teams and agents all vet leads in their own unique way. This made the solution not a clear button-to-button experience, but a dynamic one that needed artificial intelligence learning how each agent or team uniquely works.
When sitting with 5 different brokerage teams, we noticed that different teams and agents all vet leads in their own unique way. This made the solution not a clear button-to-button experience, but a dynamic one that needed artificial intelligence learning how each agent or team uniquely works.
WHERE AI CAN SUPPORT
WHERE AI CAN SUPPORT
WHERE AI CAN SUPPORT
Product, engineering, and design collaborated on understanding how the API would sort, clean, assign, and track lead client information as it flows through our platform and across the variety of agent and broker applications. The goal was to clearly agree on where AI could start solving problems for leads in the CRM, which helped us answer the bigger question of AI's potential as a solution.
Product, engineering, and design collaborated on understanding how the API would sort, clean, assign, and track lead client information as it flows through our platform and across the variety of agent and broker applications. The goal was to clearly agree on where AI could start solving problems for leads in the CRM, which helped us answer the bigger question of AI's potential as a solution.
Product, engineering, and design collaborated on understanding how the API would sort, clean, assign, and track lead client information as it flows through our platform and across the variety of agent and broker applications. The goal was to clearly agree on where AI could start solving problems for leads in the CRM, which helped us answer the bigger question of AI's potential as a solution.
HOW AI MAKES DECISIONS
HOW AI MAKES DECISIONS
HOW AI MAKES DECISIONS
Agents want AI to get the work 80% of the way done for them. Here is a decision tree that allows AI to quickly understand the level of action needed for quality impact based on the data around the customer, from recommending and drafting an email or text response, to building marketing assets for them to review!
Agents want AI to get the work 80% of the way done for them. Here is a decision tree that allows AI to quickly understand the level of action needed for quality impact based on the data around the customer, from recommending and drafting an email or text response, to building marketing assets for them to review!
Agents want AI to get the work 80% of the way done for them. Here is a decision tree that allows AI to quickly understand the level of action needed for quality impact based on the data around the customer, from recommending and drafting an email or text response, to building marketing assets for them to review!
Screen Interactions
Screen Interactions
Screen Interactions
Original Pattern
Original Pattern
Original Pattern
An approved design system pattern, already live in Compass for new leads, that became the foundation for the AI recommendation experience.
An approved design system pattern, already live in Compass for new leads, that became the foundation for the AI recommendation experience.
An approved design system pattern, already live in Compass for new leads, that became the foundation for the AI recommendation experience.
Round 1
Round 1
Round 1
The goal in round one was to stay as close to the original component as possible while adapting it to support an AI recommendation.
The goal in round one was to stay as close to the original component as possible while adapting it to support an AI recommendation.
The goal in round one was to stay as close to the original component as possible while adapting it to support an AI recommendation.
Round 1 Feedback
Round 1 Feedback
Agents said the card took up too much space, especially if the recommendation wasn't useful, and asked for better scannability.
Agents said the card took up too much space, especially if the recommendation wasn't useful, and asked for better scannability.
Round 2
Round 2
Reworded AI recommendations to be more readable, gave the card its own dedicated section, and stacked multiple cards to reduce the amount of space it takes up above the fold.
Reworded AI recommendations to be more readable, gave the card its own dedicated section, and stacked multiple cards to reduce the amount of space it takes up above the fold.
Round 1 Feedback
Agents said the card took up too much space, especially if the recommendation wasn't useful, and asked for better scannability.
Round 2
Reworded AI recommendations to be more readable, gave the card its own dedicated section, and stacked multiple cards to reduce the amount of space it takes up above the fold.
Round 2 Feedback
Round 2 Feedback
Agents wanted clearer wording, a visually labeled LeadScore instead of a bare number, and all cards visible at once.
Agents wanted clearer wording, a visually labeled LeadScore instead of a bare number, and all cards visible at once.
Final Design
Final Design
Added an item count, a clearer LeadScore, and a full-page view for three or more recommendations.
Added an item count, a clearer LeadScore, and a full-page view for three or more recommendations.
Round 2 Feedback
Agents wanted clearer wording, a visually labeled LeadScore instead of a bare number, and all cards visible at once.
Final Design
Added an item count, a clearer LeadScore, and a full-page view for three or more recommendations.
VOICE OF THE AGENT
VOICE OF THE AGENT
VOICE OF THE AGENT
"We make money interacting with people and having conversations about real estate, so being bogged down with administrative work is expensive."
"We make money interacting with people and having conversations about real estate, so being bogged down with administrative work is expensive."
FINAL Feature
FINAL Feature
FINAL Feature
After round three, I split recommendations by readiness. Buy or sell signals get a LeadScore and route to Leads, while life or business events without that signal surface as touchpoints on Contacts instead.
After round three, I split recommendations by readiness. Buy or sell signals get a LeadScore and route to Leads, while life or business events without that signal surface as touchpoints on Contacts instead.
Full FLow
Full FLow
Full FLow
Intuitive action cards
Intuitive action cards
Each recommendation card shows the LeadScore and AI's reasoning up front, so agents can act without leaving the list.
Each recommendation card shows the LeadScore and AI's reasoning up front, so agents can act without leaving the list.
Each recommendation card shows the LeadScore and AI's reasoning up front, so agents can act without leaving the list.
Drafted responses for quick action
Drafted responses for quick action
Tapping a recommendation surface an AI-drafted message ready to send. Agents review and edit instead of writing from scratch.
Tapping a recommendation surface an AI-drafted message ready to send. Agents review and edit instead of writing from scratch.
Tapping a recommendation surface an AI-drafted message ready to send. Agents review and edit instead of writing from scratch.
Feedback you can give later
Feedback you can give later
Thumbs up or down calibrates the AI without slowing agents down, helping it learn their business and refine future recommendations.
Thumbs up or down calibrates the AI without slowing agents down, helping it learn their business and refine future recommendations.
Thumbs up or down calibrates the AI without slowing agents down, helping it learn their business and refine future recommendations.
REFLECTION
REFLECTION
REFLECTION
Done better
Done better
Done better
We noticed an unexpected behavior change: once agents knew their notes were feeding an AI assistant, not just serving as personal reference, they became noticeably more detailed and factual instead of short and vague. In hindsight, we should have told agents that upfront, rather than letting them discover it and adjust their behavior on their own.
We noticed an unexpected behavior change: once agents knew their notes were feeding an AI assistant, not just serving as personal reference, they became noticeably more detailed and factual instead of short and vague. In hindsight, we should have told agents that upfront, rather than letting them discover it and adjust their behavior on their own.
Coming next
Coming next
Coming next
Giving AI an active role in the conversation: texting reactivated clients directly to gather pre-qualification details before the agent even reaches out. Instead of a raw lead, agents would get a ready-to-act snapshot of the hottest, most sales-ready relationships.
Giving AI an active role in the conversation: texting reactivated clients directly to gather pre-qualification details before the agent even reaches out. Instead of a raw lead, agents would get a ready-to-act snapshot of the hottest, most sales-ready relationships.