Working In The 'Central Analytics and Science Team'

What we learn in 'CAST' while working with analysts, decision scientists, and data scientists.

Working In The 'Central Analytics and Science Team'

By Wanda Kinasih

The Central Analytics and Science Team (CAST) enables multiple products within the Gojek ecosystem to efficiently use the abundance of data involved in the working of the app. The aim of the team is to focus on developing self service capabilities, deep analytics, and machine learning systems.

CAST consists of several small pods such as communication platforms, customer platforms, third-party platforms, groceries, marketing, entertainment, etc. Each pod has 1–3 team members. Although, larger teams such as food and transport have their own separate data teams.

The team has analysts, decision scientists, and data scientists — where each of these roles have different skillsets and responsibilities.

Analysts are experts in data analysis, data visualisation, business guide, and product decisions to address business problems. Data scientists use machine learning for recommendations, price optimisation, and real-time driver matches. In addition, a position of decision scientist has been established to deal with modelling and statistical skills. Decision scientists also focus on causal statistics, not just on the correlation.

What we do

Classification of app review data: The team works on classification of app review data by monitoring the issues reported by users or the user satisfaction level on the Gojek app. Further, all the manual work involved in labelling and reporting the issue is eliminated through automated classification. This enables us to be more scalable, with immediate action items. If you’re wondering… Yes, we read our app reviews thoroughly.

Consumer Platform data team

Smart chat reply: The canned messages feature eases the booking process by many folds. These are suggestion messages which can be used to chat on the Gojek app (generally with a driver partner regarding a booking). CAST develops the ‘smart chat reply’ by using pre-trained CartoBERT model, trained on transport services chat corpus. Using this, we get the word and sentence embedding vectors from historical chat data, which will be feature inputs to the downstream model later on.

Smart Chat by Chat Data

Multi-Touch Attribution: Let’s move to the Marketing Data team. The current method at Gojek for attributing marketing performance is last touch attribution. Here, only the last interaction will be counted for attribution purposes. This method does not take into account that there may have been multiple interactions and all of them may have impacted the decision process of the user for joining Gojek. We did Multi-Touch Attribution Analysis to demystify the conversion (installation) path to see the propensity of the new installers, towards our marketing channels.

Pros of being in CAST

With teammates working with different teams and projects, everyone has varied skills we can learn from. The opportunity to work with different pods has enables us to learn new technical skills.

Being in a pond with a variety of fish certainly brings its own benefits. Previously we have technical skills, which are not much different. In this team, I learned a lot from friends with different skills. I can learn something that is beyond my expertise, I can even work with other pods. I learned more new technical skills in the last 8 months than in the previous 1 year.

Usually, the other analysts in the team and I do a lot of ETL (Extract, Transform, Load) and visualisation as part of our everyday jobs. Now, we have worked on many statistics and machine learning projects such as NSFW classification, chat intent classification, app review analysis, CLV (Customer Lifetime Value), and many more.

Studying statistics, machine learning, or NLP is one thing. But using them on real use cases helps us provides insightful solutions which can be used to further better our products.

But it’s not always sunshine

Working in multiple products tend to make team members very project-oriented. Product ownership is a challenge in itself. There is lack of medium or long-term strategic data plan for the product handled.

Due to the pandemic, we have team members who’ve never met each other. We have to work harder to build trust with fellow data, product, and business teams.

Where we are now

We had a big data & product reorganisation, and hired new team members. We work as product-embedded data team where CAST is divided to be embedded in each product team. The entire team has get-togethers sometimes to discuss insights, learnings, and exchange knowledge that helps all of us as a team.

Say cheese!

Here are more stories about how we do what we do. To view open job positions, click below: