Backgrounds to the Project
Miso wanted to hire more computer engineers, data analysts and data scientists. The demand for those positions was booming in Seoul, so the company struggled to attract good candidates. Many Miso engineers were highly satisfied with the level of autonomy and opportunities to participate in the companyโs decision-making process. The generous stock option was also a plus.
These upsides, however, were not communicated well to potential candidates, and the HR team asked me to write blog posts that promoting them. I wanted to find ways to deliver these messages that donโt sound like straight up HR speak, and decided to go talk to the engineers who genuinely love the company. The articles tried to find out the why.
What I Focused on While Writing the Articles
Unlike other works that I had done in Miso, these pieces were meant to be read by engineers in the tech industry. I had to fully understand what exactly these Miso engineers were doing to provide interesting details to potential applicants.
At the same time, I wanted it to be accessible to people without computer programming backgrounds. Many details mentioned were about the company culture, and I wanted the blog visitors to feel how agile and open the team is.
Links to the Articles
Curated, Translated Quotes from the Interviews
ย Jack Lim, a data scientist with a business administration degreeYes, in all areas. I feel I have learned both the right attitude for problem solving and technical capabilities as a data scientist here at Miso. The first thing I worked on was the job recommendation algorithm. At first, I thought adopting a better technologyโdeep learning modelsโwould be enough for a twelve-fold jump in KPIs.
But there was nothing. I spent two months developing a new model, but the conversion rate didnโt budge. The insight I got then was that a data scientist can never make a dent in the business without fully understanding how it operates. Thatโs why I talked to a bunch of Miso suppliers on the phone. I spent more than an hour each day to find the suppliersโ issue behind the existing job recommendation algorithm. It helped me tremendously.
Through the phone calls, I realized it is important to lower the dependency on the deep learning model and to focus on profiling the relationship between suppliers and customer orders. The entire team grabbed our phones and talk to suppliers, striving to find the comparative importance of factors such as distance, payment & bonus amount, starting hours and frequency of regular orders. Then, we released a very simple model focusing on profiling and filtering.
The results were hard to believe. We saw a 2.5x increase in match success rate. A model that took two months to build yield nothing, while the one we built in two weeks provided instant results.
Before working at Miso, I had a romanticized misconception for data science. I thought a perfect model could always lead to a remarkable business solution. My experiences here at Miso allowed me to realize understanding the business and its customers is way more important than a fancy, complicated data science model.
What made you join Miso? I heard that you had a few options when looking for a new job.
I had a great interview experience with the CEO, Victor. Usually, the second interview with the CEO after the first often takes places as a formality. However, Victor tried his best to persuade me to join, explaining all of Misoโs strengths, weaknesses and opportunities.
I was told on how big the home service market is, which market segment Miso occupies, the amount of the VC funding Miso had spent so far and the potential value of my stock option when the company achieve its next stage goals. It was the first time I heard such a transparent explanation from the CEO during hiring process, and I wanted to work at a place that was this open.
I wonder whether Miso lived up to your expectations. Was Miso a lean startup that knew how to take advantage of agile methodology?
Yes, it did! I experience it all the time. When the operations team files a development request, engineers arenโt expected to come up with all the functions but only a few thatโs critical and essential for service operations. Then, we see the usersโ reaction and improve the service little by little. The entire company understand this agile way of working. Rather than sticking to the interests of each team, we prioritize tasks that would yield the highest output in the time given. To truly work in agile, a project needs to be divided into small pieces suitable for short sprints. Iโm proud that Miso is one of a few places where agile way of developing does not stay in theory.
This becomes visible when we need to change how we deliver services or implement a new service within the app. It took us only three weeks to switch our move-out cleaning service from instant-book to request-for-quotes. Here in Miso we donโt sweat to come up with the perfect product from the beginning, but rather try to come up with a minimum viable product ASAP.
During your three years in Miso, which project are you most proud of? Which one was the most memorable?
There are a lot that Iโm proud of. A few weeks ago, I led a team developing features that allow cancellations and changes of our regular subscriptions. It was only possible through chats or phone calls with a customer service representative. People outside Miso might wonder how could a service be functional without those basic features, but that could be done manually through a CS agent, so it was not our priority. Considering that itโs a service platform connecting customers and suppliers, I found so many challenging exceptions that made algorithm building extremely difficult.
For example, the same algorithm could not service customers who are canceling before ever getting a cleaning and those who are canceling after having received service a number of times. Since we do not have future service data all saved up in our system, it was very difficult to find ways to cancel or modify services that have not been created in the system. Those requests could be processed accurately only when the order processing algorithm itself is altered. I tried to come up with a few rules that encompass all exceptions, editing the codes as new cases arose.
This took a chunk of my time and energy, but Iโm super glad that it made a huge difference. Inquiries per booking reduced rapidly, and our CS agents tell me they can also feel the reduced workload.
