Rent The Runway wants to predict your fashion choices and give you a virtual closet, will you let them?
Rent the Runway believes through machine learning they can accurately identify your fashion preferences and customize your closet with the latest trends, but only if you trust their insights enough to take a chance on renting their limited selection of used clothing.
How important are new fashion items in your wardrobe? So important you’d be willing to take suggestions from a computer to update your closet? So important you’d be willing to rent used clothing to stay on-trend? Rent The Runway (RTR) thinks so, and they’re utilizing machine learning insights to give consumers a unique shopping experience.
In retail personalization has become increasingly important. In the crowded fashion market a generic experience is no longer acceptable. Companies have to differentiate their interactions, and they can utilize machine learning driven off the recent explosion of customer data to create highly personalized interactions. [1] and improve satisfaction, loyalty and ultimately customer lifetime value. In addition to enhancing revenues, machine learning can decrease costs. Machine learning can analyze customer data to identify trends, predict preferences, and optimize inventory planning.
Rent The Runway is a relatively new entrant in retail and they’re aiming to significantly change consumer behavior. RTR wants you to rent clothing from their selection of designer items rather than purchasing it directly from designers. [2] Without existing brand-value RTR needs to create a seamless and memorable experience so that consumers are willing to take a risk on both the RTR brand and the experience of clothing rentals. I think machine learning is essential for RTR to establish itself amongst shoppers.
RTR founders (and HBS graduates) Jennifer Hyman and Jennifer Fleiss recognized the depth of data they gather on consumers and have continually put it to use. As a technology-enabled rental application RTR collects a wealth of internal customer data on rental selections, designers, and shopper tendencies (regional, age-specific) then utilizes this data to create a unique shopping experience. RTR interprets supervised learning outputs with the aid of stylists to predict consumer preferences, suggest rentals, and push add-ons. [3] In addition, RTR uses reinforced learning to analyze unreported customer behavior on their interface. Through unreported and reported customer data RTR can identify trends to influence their future assortment decisions, focused on increasing both the quantity and quality of items available to renters, and glean insights for innovations that could improve their back-end interface.
In the short-term RTR is also focused on minimizing their risk in the rental market. They use machine learning to identify correlations amongst consumers that have lost, stolen, or damaged rentals that allows them to screen for potential abuses and minimize their exposure going forward. [4] In the long-term machine learning insights drive RTR’s core projects and new product offerings. Chief Analytic’s Officer Vijay Subramanian highlighted the role of machine learning and data at RTR “Giving you insights, giving you a pulse of what’s happening and why, and ascertaining how to change what we’re doing right now, and what we need to do longer-term…data has a very, very big influence on what we actually work on as a company.” [3] The Company’s introduction of a subscription-based option and their recent move into brick-and-mortar, in which customers can exchange rentals, were both driven by machine learning enabled customer-insights and are continually improved upon utilizing data [5].
Since inception RTR has strategically leveraged data and their machine learning insights have improved over time as more users have entered the platform and higher quality customer data has become available [1]. However, in my prior experience in business development in the retail industry I witnessed the unique approach of stylists and creative directors and their value in shaping both the brand and product offerings. Ultimately, I believe machine learning has to be interpreted by a human to draw out the most relevant actionable insights. I worry RTR may rely too heavily on data and lose the human touch in their rental model. I would caution them to be wary of collecting too much irrelevant data and allowing past consumer choices to over-influence their decisions. RTR needs to keep a forward-looking eye on the fashion industry.
As RTR grows their brick-and-mortar footprint I think they should expand their use of analytics to customize their instore experience. Different regions and consumer psychographics heavily affect retail preferences. RTR can benefit from relying on data analytics to tweak their retail layouts, in-store inventory, and the tenets of their subscription model as they continue to roll out these initiatives.
I was fascinated by RTR’s use of machine learning to predict rental abuses. They certainly are at risk by lending out expensive products. How do you feel about them making potentially unfair conclusions about consumers to protect their downside? Is this different than banks approving or rejecting individuals for loans?
So much of brand-value and consumer behavior remains a mystery and is influenced in a multitude of ways – how valuable do you think the ‘human touch’ remains in creative industries (such as retail)? Do you think machine learning can replace creatives? (790 words)
[1] E. Brynjolfsson and A. McAfee. What’s driving the machine learning explosion? Harvard Business Review Digital Articles (July 18, 2017).
[2] Bertoni, S. (2018). The Secret Mojo Behind Rent The Runway’s Rental Machine. [online] Forbes. Available at: https://www.forbes.com/sites/stevenbertoni/2014/08/26/the-secret-mojo-behind-rent-the-runways-rental-machine/#3a587c4138bf [Accessed Nov. 2018].
[3] Ferguson, R. (2018). Rent The Runway: Organizing Around Analytics. [online] MIT Sloan Management Review. Available at: https://sloanreview.mit.edu/article/rent-the-runway-organizing-around-analytics/ [Accessed Nov. 2018].
[4] Dutcher, J. (2018). Ask a Data Scientist at Rent the Runway. [online] Datascience.berkeley.edu. Available at: https://datascience.berkeley.edu/ask-a-data-scientist-rent-the-runway/ [Accessed Nov. 2018].
[5] Milnes, H. (2018). Live from NRF: How Rent the Runway’s Unlimited subscription model changed its in-store strategy – Digiday. [online] Digiday. Available at: https://digiday.com/marketing/live-nrf-rent-runways-unlimited-subscription-model-changed-store-strategy/ [Accessed Nov. 2018].
I think machine learning makes a lot of sense for this type of business model but, as you pointed out, there is a considerable risk when fashion companies try to use past preferences to predict future tastes. I would also be concerned about how the algorithms are factoring the differences of style between different locations, and the shift in styles from season to season (I wonder if this would make the learning part of the algorithm back to moment zero). Because of this, I believe that the concept of augmented human makes a lot of sense when deciding future decisions of dress purchases and predictions of rentals.
Where I do believe machine learning can have a lot of impact is on supply chain management, specifically on the estimation of returns, time adjustments for cleaning the dresses, and period of time that one thinks the dress will be amortized (how many rentals before the dress losses all its value).
One last thought is that if RTR pushes more and more brick and mortar spaces it will lose its capacity to gather more data (doesn´t have a history of clicks, for example) and train the algorithm to make better predictions.
As you noted, I agree machine learning has a lot of potential in the e-commerce industry, providing a more personalized and customized experience as many RTR customers are likely looking for. I’m thinking a chatbot might help the service tremendously, both on the customer side as well as on the company side. I too agree that machine learning has its limitations. As we discussed with Gap and Big Data, I think the best use of machine learning will be to complement humans rather than replace them in terms of creative design and the next big fashion. Fashion is forward and I am not convinced a machine knows that. I thought it was interesting you noted that the ethical implications of machine learning categorizing a customer as RFR thinks about rentals and likelihood of return. In my mind, as long as a variety of data sources are used to come to the conclusion, it is a fair mechanism to use by the company to prevent issues. It is not dissimilar to how a bank thinks about lending.
Great piece. I agree with your take on the need for both machine learning and human to best determine product assortment. I’m glad to see that RTR is already doing this in a way – as you point out, RTR is interpreting supervised learning outputs with the aid of stylists to determine product assortment. In an industry where ratchet effects can quickly change the course of what’s in vogue, the “stylist’s touch” will remain critical to ensure that RTR remains relevant and appealing to its core consumer. Ultimately, I don’t think that machine learning can completely replace the creative visionary for this very reason – fashion is too personal and too tactile (with too unpredictable a life cycle) to fully entrust to the machines. I do remain curious, though, about whether these ratchet effects will be as important at RTR, as formalwear and women’s dress fashion hasn’t seen as drastic a change as streetwear or sportswear has.
To echo the point above, I do believe as well that RTR must continue to rely on outputs from the data to influence human decisions on what outfits should be recommended to people. Humans at RTR should take the data from the supervised learning outputs and determine what to recommend people, but also to your point, they need humans leading creative work as well to continually find new pieces from designers that can be added to their options.
How do you think RTR will continue to ensure that the recommendations they provide to consumers are unique enough? I have heard anectotally about Stich Fix recommending the same outfits to people who show up to work on the same day wearing the same thing. I know that RTR is different than Stich Fix, but what can they do to ensure that their algorithm provides people unique enough options so that this does not happen?
The point you raise about being able to discriminate against consumers based the machine learning output on predicting abuse is super interesting. At the end of the day RTR will not make money if they lose expensive products on a regular basis, so I feel as though they are very smart to use data in this manner to discriminate among which products they will offer to their consumers.
For me the above article raises two interesting points — one around how humans are still needed in conjunction with machine learning, and the other around rental abuses.
Specifically with fashion I think there needs to be an intersection between a designer and machine learning. I don’t think machine learning alone can determine the best outfit to send to a consumer. There is something more intuitive and less concrete with fashion (especially a fancy one-time dress) which requires at least a small human touch.
Lastly, I found the point relating to rental abuses illuminating. While I do see a possible scenario where RTR incorrectly identifies someone who is a valuable customer – I think using data to help mitigate risk is something they absolutely need to do.
Great read that resonated strongly with me, especially your point on using machine learning to underwrite risk (at my previous company, we were renting out much more expensive items – luxury watches – and underwriting was always top of mind for us). While the use of AI to predict the “quality” of the rentees may lead to unfair practices, I believe it is a necessary step that RTR needs to take in order to protect their downside. RTR can also structure its product offerings in ways to allow riskier newcomers to the business to participate while mitigating their risk (e.g., allowing them to start only in the lower-level tiers and then adjust their privileges as they demonstrate patterns of good behavior). This can also lead to additional data for their AI machine to crunch and evolve with.