Washing machine learning: the data revolution at Jeff
Jeff is on a mission, and that mission is to become the app that will let you live the good, good life. How are we going to do that? With a single platform to combine everything that will help you enjoy the things you love to enjoy. Be that by saving you time on those boring day to day things you don’t want to do, or by giving you the best means to look, and feel good, wherever you are and whenever you want. On the other side of the coin, by becoming a Jeff user, you are helping local entrepreneurs who have their very own Jeff franchise. In other words, we help you to live better, and, in return, you help your local community to live better.
Things are going well so far, and we have managed to open over 1000 franchises in 30 countries, covering 3 (for now…) different lines of business — laundry, beauty, and fitness. But this is only the start, and to reach our goal of becoming the first super app, we have to keep pushing forward. How are we going to do this? In many different ways, but you are here for one — data.
In the last few years data has been one of the hot topics, and at Jeff it’s no different. Across our different lines of business we see how customers and franchise partners subscribe to services, handle orders, deliver laundry… And all of these interactions generate huge amounts of information that help us improve both the business and make the good, good life even better. It is us in the data team who will be the protagonists of this part of the story.
We are the new kids on the block, with most of us having joined in the last year. We are split into your typical analysts, engineers, and scientists — but from time to time we get our hands dirty in all sorts of ways. Each of the teams have their own challenges — whether it is setting up the new Tableau service, building our new data warehouse, or getting the company started on the data science journey. What we all have in common is the goal to transform Jeff using data.
Specifically for data science, we have the vision of becoming the in-house experts when it comes to using data to understand how our business works. We want to make sure that as part of the company’s culture, no decision is made without asking the data — and that we can provide the most profound answers to these questions. The constantly evolving environment at Jeff means that there are not only plenty of opportunities but plenty of challenges too.
On the customer side, one of the biggest problems that faces Jeff is not only how to grow, but how to sustain this growth, and how to develop many independent Jeff services at the same time. There are many ways data science can get involved in this, from understanding which users are the ones that stay with us long term, experimenting with the best ways to encourage different user behaviour, to building intelligent systems that recommend new Jeff services to our customers depending on their needs. Combine all of these with the fact that Jeffers interact with us both online, through the app, and offline, in person, and you have some really interesting opportunities for data science.
So far, we have seen that some of our customers behave one way, and many behave another way. Whether it is carrying out a marketing campaign or tweaking the app, these different groups are sure to react differently. We always want to make sure that any action has the right impact and best results possible, and we saw that we could help that become a reality by building up the company’s in-depth understanding of these different behaviours. As part of this work, we found that using a user’s order history, we uncovered several distinct profiles, which we want to start using to better define the customer strategy moving forward.
On the other hand, we have our franchise partners. Normally, when entrepreneurs open their own franchise, they have to rely on their “business know-how” to make sure they are successful. They rely on the limited information available to them, their gut feeling, and hand crafted solutions to make the right decisions and take the best course of action. As data scientists at Jeff, we want to change this, and give our franchise partners the sort of help that others in their situation don’t normally have. Maybe they want to know which customers are the least satisfied, how to choose the best delivery route, or what processes they need to improve when compared to other partners in similar situations. Our job is to make sure that they have access to these answers, and do that in a flexible, objective way.
An example of a simple but important question we asked ourselves was — what is the best place to open up a new Jeff franchise? This is quite complicated, and hard to answer for most potential entrepreneurs. It probably depends a lot on the “flow” or concentration of people in that area — but it would be a tall order for future partners to understand, on their own, how this information relates to potential success.
Now consider that today we might help someone in Argentina, but tomorrow it might be Poland, or the Philippines. We have to be careful in our approach to make sure that we can scale our solutions and provide the best help for all our partners around the world. One of the interesting (and maybe obvious) results was that the rent is a great predictor for the quality of location. This makes sense, since the rent more or less directly reflects how desirable that area is for a business. On the other hand, it isn’t perfect, because all of our businesses have their own margins, business models, and requirements. With this and a few other variables, we combined a simple model with a dashboard to help stakeholders predict which potential new location will be the best.
It’s clear that exciting and challenging times lie ahead, and as a team we will need to work together to overcome them. We are working hard to define and improve our methodology, but we are clear that it has to be based on simple approaches, rapid testing and iterating, and a transparent, collaborative environment.
Simple solutions are amazing, because they are easy to understand, implement, and to improve. One of the hardest parts of data science in the real world is getting your solutions out of notebooks and into the product. Simple approaches are proven ways to get this done faster, while being more confident of the potential effects and scope of your work. Getting the big guns out once in a while is always necessary, but focusing on the easy wins first is a great way to make a guaranteed, huge impact in the shortest time possible.
Getting things into production is also important to really understand the performance of your work. You can only analyze cold data so much and the scientific method demands experimentation. This is the only way that you can be sure of your correlation vs. causation dilemma. Once you know how real people react to your solutions, you improve on it to fix the weak points. Doing this quickly and often is how you can ensure that you are providing maximum value. Simple approaches get bonus points here.
Finally, as data scientists we all come from different backgrounds and walks of life — all with a different point of view and value to add. To take advantage of this, we have to encourage a culture of communication, transparency, and collaboration. This will be the recipe for success long term, and ensure our growth and development as a team, and as data scientists.
We are excited and optimistic about data science in Jeff, and we are sure that some time soon we will be back to share some of our success stories. We hope that you have enjoyed hearing about the exciting challenges that lie ahead, and our vision for how to overcome them. If you want to find out more or have any comments, don’t hesitate to get in touch!
Follow us at @JeffApp_Tech
