During the first weeks as a Business Information Management (BIM) student at Rotterdam School of Management (RSM) my three colleagues and I were tasked in the class Information Strategy, taught by Professor Ting Li, to solve a real business problem with a new emerging technology. From Artificial Intelligence (AI) to Augmented Reality (AR); the choice was completely ours.
The business problem
We started with a brainstorm, by pitching in with our different perspectives and thoughts of business problems ourselves. Fairly soon we stumbled upon the clothing industry.
Namely: online clothing returns are very costly for online retailers.
We liked this problem a lot, because it was easily quantifiable. The costs related to online returns are simply enormous, so there is a lot to be gained by solving this problem (McKinsey, 2021).
Now it was time to think of a fitting solution, and this proved to be a little bit more difficult.
The initial solution
The first solution we thought of was to develop return handling software. The software would provide online clothing retailers with a dashboard showcasing all their return data. Moreover, the AI model would give insights regarding all clothing items offered and recommendations for handling individual returns.
We were semi-contented with this solution, and we proposed it to our professor. However, our professor gave us feedback to be more specific with the solution: what is the real problem that is solved by implementing this solution?
So, we decided it would be wise to have one more brainstorm before continuing.
The final solution
So, we were pondering the question where the real value is to be gained on the problem of return-related costs for online clothing retailers.
Soon we concluded that real value lies in preventing returns compared to better return handling. Namely: if returns are prevented, no costs are being made.
Then we had to find a way to realize this. We found that a lot of returns are related to choosing the wrong size (McKinsey, 2021). Therefore, we were thinking of a way to prevent returns by increasing the right size choice for the first time. This would not only decrease costs for the online retailer, but it would also increase customer satisfaction.
We concluded that the best way to prevent size related returns was to build an AI model that would recommend customers the size that would best fit them based on their weight, height, and body type.
Finally, we thought it best to build the model as internal consultants for Adidas instead of creating our own software company. The reasoning behind this choice: Adidas is a more interesting case due to the sheer size, high competition, and public records of the company.
Now we could each focus on a single part of the business case to further work out.
Competitive landscape
My main contribution to the project was to develop the competitive landscape and come up with a handful of Key Performance Indicators (KPIs).
To develop the competitive landscape, I did a SWOT Analysis by listing all the major strengths and weaknesses of our solution and the opportunities and threats that it faced.
Further, I made a competition table and filled it in for all the main competing software solutions of the sizing tool we proposed to Adidas. The conclusion: only our model pairs its own data (Adidas’ data) and catalogue with full coverage of its assortment.
We also found some ways to learn from the competition: Nike implemented AR as a solution but ultimately failed. AR proved to be too difficult to use for customers and therefore prone to errors. Therefore, we decided it wise to not use AR ourselves.
Difficulties choosing financial KPIs
It was difficult to come up with a valid KPI to measure the financial performance of the solution. The main ‘obstacle’ deals with the growing sales of Adidas, which is a driver for increasing returns.
Our solution is supposed to decrease returns, leading to cost savings. So: first I came up with a percentage decrease in overall return costs. However, this figure does not account for the growing sales of Adidas, which drives returns.
So, I thought of a way to account for this sales growth: the returns to sales ratio. This ratio divides costs related to returns by sales. The resulting KPI: annual returns-to-sales ratio falls 25% within three years.
Exciting Opportunities
We also came up with a couple of interesting opportunities, which were incorporated into the SWOT analysis that I made.
Firstly, data driven marketing. This entails recommending products based on combining previous personal order data and product catalogue data of Adidas. For example: recommending a previous customer a certain new sweater with similar dimensions as a sweater that the customer previously ordered and fits well.
Secondly, letting online retailers, such as Zalando and Wehkamp, use our solution for more data points and better customer experience. By partnering with online retailers, we will gather more data points, which will help increase the accuracy of our model.
This leads me to the key learning I had during this project, which is understanding the power of the data flywheel.
Key learning: the data flywheel
Before this project, I viewed AI primarily through a technical lens, focusing on the quality of the model itself. However, after finishing up my competitive analysis I came to a crucial conclusion: Data is more important than the AI model itself.
This is illustrated in the power of the data flywheel: the more datapoints, the better the AI agent functions, which improves predictions and in turn increases the usage by customers, which then again increases data points, and so on.
This positive feedback loop is super powerful and introduces a first mover advantage. That is, the first mover enters the data flywheel first and benefits from the positive feedback loop early on, which fights competition and attracts more users.
Relating these learnings back to the Adidas case, led me to the conclusion that Adidas already has a large user base from their existing app, which will help jump start the data flywheel and reach critical mass early. This significantly decreases the risk of the project.
Collaboration
During the project I worked together in a group of four. We all had a similar background (Dutch), which made it easy to work together. However, it can be a downside if there is no disagreement or different ideas proposed.
Fortunately, our team was sufficiently critical when needed. Especially while forming the solution during brainstorming, this mixture of perspectives led to important breakthroughs.
Therefore, I can also conclude that brainstorming is a highly effective tool in collaborating on a group project. In this setting, members first think of ideas themselves for a limited amount of time, after which everybody shares their ideas and exchanges perspectives and opinions. This led to very fruitful discussions in our case.
Conclusion
All in all, I sincerely enjoyed working on this project. Moreover, while working on the project I developed a new perspective on what is needed in building an AI agent and the successful implementation of them.
References
McKinsey & Company. (2021, May 25). Returning to order: Improving returns management for apparel companies. mckinsey.com
