An AI-assisted CMA tool for pricing homes with confidence
In 2020, I led design at Compass to bring a new Comparative Market Analysis (CMA) tool to their real estate platform, helping realtors guide their clients toward prices that balance client expectations with market realities. I collaborated closely with a cross-functional team of front-end engineers, AI/ML engineers, product managers, and user researchers from initial kickoff through final launch.
Company
Compass
Role
Lead designer
Year
2020
An AI-assisted CMA tool for pricing homes with confidence
In 2020, I led design at Compass to bring a new Comparative Market Analysis (CMA) tool to their real estate platform, helping realtors guide their clients toward prices that balance client expectations with market realities. I collaborated closely with a cross-functional team of front-end engineers, AI/ML engineers, product managers, and user researchers from initial kickoff through final launch.
Company
Compass
Role
Lead designer
Year
2020
An AI-assisted CMA tool for pricing homes with confidence
In 2020, I led design at Compass to bring a new Comparative Market Analysis (CMA) tool to their real estate platform, helping realtors guide their clients toward prices that balance client expectations with market realities. I collaborated closely with a cross-functional team of front-end engineers, AI/ML engineers, product managers, and user researchers from initial kickoff through final launch.
Company
Compass
Role
Lead designer
Year
2020
The problem
Homes are unique. Each has a range of plausible market values that are shaped by hard-to-measure qualities, like buyer demand, craftsmanship, recent sales, timing, and location.
Home buyers and sellers mostly lack the experience and market knowledge needed to accurately assess a property’s likely sale value. Realtors are the ones clients rely on to guide them through this complex and consequential process.
The problem
Homes are unique. Each has a range of plausible market values that are shaped by hard-to-measure qualities, like buyer demand, craftsmanship, recent sales, timing, and location.
Home buyers and sellers mostly lack the experience and market knowledge needed to accurately assess a property’s likely sale value. Realtors are the ones clients rely on to guide them through this complex and consequential process.
The problem
Homes are unique. Each has a range of plausible market values that are shaped by hard-to-measure qualities, like buyer demand, craftsmanship, recent sales, timing, and location.
Home buyers and sellers mostly lack the experience and market knowledge needed to accurately assess a property’s likely sale value. Realtors are the ones clients rely on to guide them through this complex and consequential process.
Competitive market analysis (CMAs)
Realtors use CMAs to structure pricing conversations with their clients, showing how market conditions and subtle property differences influence value. A strong comparable might match the subject in size and layout, but photos reveal a dated kitchen. Another similar home may look like a perfect match—until you see it sits inches from a noisy highway. These details help align the realtor’s perspective with the client’s and reinforce the realtor’s expertise and value.
Competitive market analysis (CMAs)
Realtors use CMAs to structure pricing conversations with their clients, showing how market conditions and subtle property differences influence value. A strong comparable might match the subject in size and layout, but photos reveal a dated kitchen. Another similar home may look like a perfect match—until you see it sits inches from a noisy highway. These details help align the realtor’s perspective with the client’s and reinforce the realtor’s expertise and value.
Competitive market analysis (CMAs)
Realtors use CMAs to structure pricing conversations with their clients, showing how market conditions and subtle property differences influence value. A strong comparable might match the subject in size and layout, but photos reveal a dated kitchen. Another similar home may look like a perfect match—until you see it sits inches from a noisy highway. These details help align the realtor’s perspective with the client’s and reinforce the realtor’s expertise and value.
CMA builder and editor
Building CMAs requires searching for, analyzing, and selecting comparable properties. Realtors then assess location and neighborhood quality, and add notes and adjustments to highlight nuances and clarify their pricing strategy.
CMA builder and editor
Building CMAs requires searching for, analyzing, and selecting comparable properties. Realtors then assess location and neighborhood quality, and add notes and adjustments to highlight nuances and clarify their pricing strategy.
CMA builder and editor
Building CMAs requires searching for, analyzing, and selecting comparable properties. Realtors then assess location and neighborhood quality, and add notes and adjustments to highlight nuances and clarify their pricing strategy.
Comparable properties
Comparables are used as pricing analogies for the subject property. The challenge for realtors is that no two properties are exactly alike. For example, a comp may be a close match but still differ in important ways. It may have a better view, be located in a less desirable part of town, or have sold during stronger market conditions. These differences can significantly affect the final sale price.
Comparable properties
Comparables are used as pricing analogies for the subject property. The challenge for realtors is that no two properties are exactly alike. For example, a comp may be a close match but still differ in important ways. It may have a better view, be located in a less desirable part of town, or have sold during stronger market conditions. These differences can significantly affect the final sale price.
Comparable properties
Comparables are used as pricing analogies for the subject property. The challenge for realtors is that no two properties are exactly alike. For example, a comp may be a close match but still differ in important ways. It may have a better view, be located in a less desirable part of town, or have sold during stronger market conditions. These differences can significantly affect the final sale price.
Suggested price adjustments
Highlight areas where the ML thinks the comparables differ from the subject property.
Suggested price adjustments
Highlight areas where the ML thinks the comparables differ from the subject property.
Suggested price adjustments
Highlight areas where the ML thinks the comparables differ from the subject property.
Adjustment tooltip
Tool tips explain the suggestion and how we calculated it using our model.
Adjustment tooltip
Tool tips explain the suggestion and how we calculated it using our model.
Adjustment tooltip
Tool tips explain the suggestion and how we calculated it using our model.
AI / ML suggested adjustments
I worked closely with ML engineers to identify opportunities where models could surface useful signals without overriding agent judgment. We leveraged Compass's wealth of listing data — including property details, historical sales, market trends, and amenities — we designed an AI-assisted approach to accelerate CMA creation while generating more meaningful pricing insights.
The goal was not to replace agents with automation, but to combine human expertise with machine intelligence to support better, more accurate pricing decisions. The suggested price adjustments were one such example.
AI / ML suggested adjustments
I worked closely with ML engineers to identify opportunities where models could surface useful signals without overriding agent judgment. We leveraged Compass's wealth of listing data — including property details, historical sales, market trends, and amenities — we designed an AI-assisted approach to accelerate CMA creation while generating more meaningful pricing insights.
The goal was not to replace agents with automation, but to combine human expertise with machine intelligence to support better, more accurate pricing decisions. The suggested price adjustments were one such example.
AI / ML suggested adjustments
I worked closely with ML engineers to identify opportunities where models could surface useful signals without overriding agent judgment. We leveraged Compass's wealth of listing data — including property details, historical sales, market trends, and amenities — we designed an AI-assisted approach to accelerate CMA creation while generating more meaningful pricing insights.
The goal was not to replace agents with automation, but to combine human expertise with machine intelligence to support better, more accurate pricing decisions. The suggested price adjustments were one such example.
Feedback loop to help improve the model
Provided a way for agents to leave feedback about the suggestions. The feedback gathered from this menu helped us gauge the quality of the suggestions and helped us iteratively improve the model in the future.
Feedback loop to help improve the model
Provided a way for agents to leave feedback about the suggestions. The feedback gathered from this menu helped us gauge the quality of the suggestions and helped us iteratively improve the model in the future.
Feedback loop to help improve the model
Provided a way for agents to leave feedback about the suggestions. The feedback gathered from this menu helped us gauge the quality of the suggestions and helped us iteratively improve the model in the future.
Learnings and explorations
We explored ways to keep the adjustment builder as simple as possible. An earlier version displayed one adjustment at a time. It looked good in certain situations but became complicated when the ML suggested 2+ adjustments. Research revealed that the interface put too much visual weight on the suggested adjustments, overshadowing the primary input field. The visual and conceptual imbalance triggered a visceral negative reaction in the realtors we spoke with. We wanted to instill the appropriate amount of trust in our users.
Realtors firmly told us during usability studies that they were the experts, and the design oversold the technology’s abilities to do the work for them. They saw the primary value of technology as something that augments their knowledge and assists them, not replaces them. We took this to heart and used it to guide future iterations and the final build.
Learnings and explorations
We explored ways to keep the adjustment builder as simple as possible. An earlier version displayed one adjustment at a time. It looked good in certain situations but became complicated when the ML suggested 2+ adjustments. Research revealed that the interface put too much visual weight on the suggested adjustments, overshadowing the primary input field. The visual and conceptual imbalance triggered a visceral negative reaction in the realtors we spoke with. We wanted to instill the appropriate amount of trust in our users.
Realtors firmly told us during usability studies that they were the experts, and the design oversold the technology’s abilities to do the work for them. They saw the primary value of technology as something that augments their knowledge and assists them, not replaces them. We took this to heart and used it to guide future iterations and the final build.
Learnings and explorations
We explored ways to keep the adjustment builder as simple as possible. An earlier version displayed one adjustment at a time. It looked good in certain situations but became complicated when the ML suggested 2+ adjustments. Research revealed that the interface put too much visual weight on the suggested adjustments, overshadowing the primary input field. The visual and conceptual imbalance triggered a visceral negative reaction in the realtors we spoke with. We wanted to instill the appropriate amount of trust in our users.
Realtors firmly told us during usability studies that they were the experts, and the design oversold the technology’s abilities to do the work for them. They saw the primary value of technology as something that augments their knowledge and assists them, not replaces them. We took this to heart and used it to guide future iterations and the final build.
The results
We launched the CMA and the adjustment builder experience to the Compass Real Estate platform in 2020. Following the launch, CMA creation tripled and realtors responded positively. The tool played a key role in the build-up to Compass's public offering
The results
We launched the CMA and the adjustment builder experience to the Compass Real Estate platform in 2020. Following the launch, CMA creation tripled and realtors responded positively. The tool played a key role in the build-up to Compass's public offering
The results
We launched the CMA and the adjustment builder experience to the Compass Real Estate platform in 2020. Following the launch, CMA creation tripled and realtors responded positively. The tool played a key role in the build-up to Compass's public offering
▲
3x increase in new CMAs
3x increase in new CMAs
3x increase in new CMAs
In CMA docs created by Compass' Real estate agents
▲
2X faster CMA creation time
2X faster CMA creation time
2X faster CMA creation time
Due to streamlined workflows and suggested adjustments