Improving seller quality on Pinterest
Improving seller quality on Pinterest
Improving seller quality on Pinterest
Improving seller quality on Pinterest
Intro
Pinterest is a marketplace that helps shoppers discover sellers with trendy fashion items or aesthetic home decor pieces. As an Apprentice Product Manager, I helped lead the product team in charge of controlling seller quality on this marketplace.
My team's goal was simple: keep bad actors off the platform and make it easy for high quality sellers to find success. To this end, I defined and shipped updated requirements for joining Pinterest's Verified Merchant Program. We shipped to millions of business users worldwide, and saw notable improvements to average seller quality.
Highlights
asdfasdfasdfasfawef
Leveraged in-house ML model to define and quantify "seller quality"
Conducted SQL analysis to identify traits most correlated with high quality sellers
Led ramp plan that shipped new experience to millions of business users
The Problem
The Problem
The Problem
The Problem
Relatively low quality sellers were being admitted to Pinterest's seller recognition program. As a result, users would often report having bad experiences with sellers Pinterest claimed to be "Verified Merchants."
The root cause of this problem was that we didn't have a rigorous definition of what makes a high or low quality seller. The requirements to become a Verified Merchant reflected this, and thus the program struggled to keep out low quality sellers.
Relatively low quality sellers were being admitted to Pinterest's seller recognition program. As a result, users would often report having bad experiences with sellers Pinterest claimed to be "Verified Merchants."
The root cause of this problem was that we didn't have a rigorous definition of what makes a high or low quality seller. The requirements to become a Verified Merchant reflected this, and thus the program struggled to keep out low quality sellers.
Relatively low quality sellers were being admitted to Pinterest's seller recognition program. As a result, users would often report having bad experiences with sellers Pinterest claimed to be "Verified Merchants."
The root cause of this problem was that we didn't have a rigorous definition of what makes a high or low quality seller. The requirements to become a Verified Merchant reflected this, and thus the program struggled to keep out low quality sellers.
Relatively low quality sellers were being admitted to Pinterest's seller recognition program. As a result, users would often report having bad experiences with sellers Pinterest claimed to be "Verified Merchants."
The root cause of this problem was that we didn't have a rigorous definition of what makes a high or low quality seller. The requirements to become a Verified Merchant reflected this, and thus the program struggled to keep out low quality sellers.
Proposed Solution
Proposed Solution
Proposed Solution
Proposed Solution
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
Proposed Solution
Proposed Solution
Proposed Solution
Proposed Solution
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
I chose to tackle this problem in two steps:
1) Develop a rigorous way of defining and measuring seller quality
2) Use that definition to propose new requirements for becoming a Verified Merchant
By making an effort to quantify this abstract notion of seller quality,
I chose to tackle this problem at the root cause—not having a rigorous definition of seller quality—so that we have a testable, iterative foundation on which
to ensure we wouldn't have the same problem again.
To enact the first, I worked with XFN engineering teams on a multi-head ML model that assigns a numeric quality score to sellers. This model was trained on large sets of internal seller data, which enabled it to accurately distinguish high and low quality sellers.
Once we had a rigorous way to measure seller quality (which we cross-checked with other quality signals), my task was to identify tangible traits that correlated with high quality sellers. To do this, I conducted an extensive SQL analysis.
Results
Results
Results
Results
asdfasdfasdfasdfasdf
asdfasdfasdfasdfasdf
asdfasdfasdfasdfasdf
asdfasdfasdfasdfasdf