How does Pinterest know what you want before you do? The answer is simple: Machine learning. While on the surface Pinterest may seem like a simple virtual bulletin board, Pinterest is actually a cutting-edge machine learning platform that is powered by the world’s largest image rich data set. Machine learning can be found throughout Pinterest’s business functions: from its Black Ops team detecting spam content and users to its Discovery team recognizing images and providing recommendations to its Growth Analytics team determining what emails to send to users and to prevent churn.[i] Most paramount to Pinterest is its ability to recognize objects accurately and provide recommendations and related content to users. That’s why, in 2015, Pinterest acquired Kosei, a machine learning commerce recommendation engine.[ii] With over 250 million people using Pinterest every month[iii], it’s crucial that Pinterest be able to respond to searches with relevant images and recommend related content to users.
Given the importance of machine learning to Pinterest’s business model, Pinterest has continued investing in artificial intelligence. In February 2017, Pinterest announced the launch of Pinterest Labs, which brings together top researchers, scientists, and engineers from around the world to tackle the most challenging problems in machine learning and artificial intelligence. [iv] Through Pinterest Labs, Pinterest has been able to incorporate machine learning into their product development process. For example, Pinterest Labs launched PinSage, a process that uses contextual information from surrounding Pins to provide more accurate recommendations for additional matches, as opposed to using just the image or the keywords.[v] With PinSage, Pinterest is now showing users better matches in relation to what they are looking for, which has led to a 25 percent increase in impressions for Shop the Look, a feature that lets Pinterest users buy clothes seen in Pins.[vi],[vii] By embedding machine learning into Pinterest’s product development process, it has promoted innovation within the organization and in turn improved Pinterest’s business performance.
While Pinterest Labs provides short-term product development and innovation with machine learning, Pinterest Labs also looks and positions Pinterest into the medium-term of artificial intelligence. Through Pinterest Labs, Pinterest is partnering with the broader research community – sharing their learnings along the way on their website, through research papers published at conferences and by releasing datasets to further academic research.[viii] Pinterest’s machine learning, and machine learning more broadly, is improved by having larger datasets and getting more feedback fed into the dataset. By engaging with the external machine learning world, Pinterest is able to further enhance and improve its recommendation model, improving the recommendation process and thus the product development flow.
In addition to improving Pinterest’s image rich data set with internal and external machine learning partners, I think that Pinterest’s management team should think about how they can improve recommendations that may not be visually the same but still “go together”. For example, with PinSage and Shop the Look, users are able to get very accurate image results for items they may want to purchase. But, what if Pinterest could take it a step further, and provide recommendations based on captions about items looking good together, or based on items that have been shopped together frequently. If Pinterest was not only able to help you find the perfect navy wrap dress that you are looking for, but also the perfect pair of nude shoes that it goes with the dress, Pinterest’s recommendation engine could be even more powerful.
As Pinterest continues to invest in machine learning, I think they need to consider what returns they are getting on this investment. Is this type of investment sustainable in the longer-term? If not, what are other ways that Pinterest can continue to innovate in product development, but without taking on such large scale and long lead time projects?
(Word Count: 749)
[i] Josh Constine, “Pinterest Acquires Machine Learning Commerce Recommendation Engine Kosei,” TechCrunch.com, 2014, https://techcrunch.com/2015/01/21/facebook-past-google-present-pinterest-future/, accessed November 2018.
[ii] Michael Lopp, “The future of machine learning at Pinterest,” medium.com, January 21, 2015, https://medium.com/@Pinterest_Engineering/the-future-of-machine-learning-at-pinterest-88e6d4bf1968, accessed November 2018.
[iii] Khari Johnson, “Pinterest surpasses 250 million monthly active users,” venturebeat.com, September 10, 2018, https://venturebeat.com/2018/09/10/pinterest-passes-250-million-monthly-active-users/, accessed November 2018.
[iv] Jure Leskovec, “Introducing Pinterst Labs,” medium.com, February 18, 2017, https://medium.com/@Pinterest_Engineering/introducing-pinterest-labs-2e230bdcd1fc, accessed November 2018.
[v] Andrew Hutchinson, “Pinterst Improves Related Pin Recommendations, Increasing Engagement and Activity” socialmediatoday.com, August 16, 2017, https://www.socialmediatoday.com/news/pinterest-improves-related-pin-recommendations-increasing-engagement-and-a/530198/, Accessed November 2018.
[vi] Khari Johnson, “Pinterest surpasses 250 million monthly active users,” venturebeat.com, September 10, 2018, https://venturebeat.com/2018/09/10/pinterest-passes-250-million-monthly-active-users/, accessed November 2018.
[vii] Andrew Shai et. al, “Visual Discovery at Pinterest”, Pinterest Labs, https://arxiv.org/pdf/1702.04680.pdf, accessed November 2018.
[viii] Jure Leskovec, “Introducing Pinterst Labs,” medium.com, February 18, 2017, https://medium.com/@Pinterest_Engineering/introducing-pinterest-labs-2e230bdcd1fc, accessed November 2018.