Background

Developing Mobile Digital Ecosystem to Win The Market

Compounding the challenge in the last five years was fierce competition in Broadband Service and Data Service monetization. Despite these challenges, Telkomsel continued to advance towards its goal of transforming Telkomsel into a Digital Telco Company and the leader in Mobile Digital Business. This goal demands the transformation of the entire organization end-to-end, including business, people, organization and corporate culture, as well as Telkomsel larger digital ecosystem. All of these are critical elements as the aim to expand the capabilities to create a gateway to the digital world for the people of Indonesia. 

In order to transform into a Digital Company and leading Mobile Digital provider, Telkomsel established a number of programs under three major mobile transformation categories, with the overall goal of building up the digital business, while defending legacy and expand broadband connectivity. Including in first transformation category is building Digital Lifestyle (DLS) platforms and capabilities to accelerate hockey-stick growth.

The DLS platform building focused on developing digital music in the form of LangitMusik as well as digital games. In 2018, Telkomsel plays in almost all value chains of the gaming ecosystem including media and distribution, payment, e-sport, and publishing. Dunia Games is the leading game portal with more than 5 million monthly visitors in Indonesia. In October 2018, Telkomsel through Dunia Games launched its first game, ShellFire, a smartphone platform application in the action genre within the FPS (First Person Shooter) and MOBA (Multiplayer Online Battle Arena) sub-genres which have various game modes, hero characters, maps, and in-app items.

A Personalized Advertising to Reach The Right Customer

As one of Telkomsel’s objectives this year is to identify cross-sell opportunities, predict customer LTV, and serving relevant ads for remarketing campaigns, the audiences are built-in GA manually using historical data based on multiple definitions. These audiences used in remarketing campaigns to target a specific set of users for specific campaigns. This approach has limited capability in predicting the customer LTV, and forecasting the sale for the future period.

“As our goal to transform into a Digital Telco Company and the leader in Mobile Digital Business demands the transformation of our entire organization end-to-end, including business, people, organization and corporate culture, as well as our larger digital ecosystem.”  - Telkomsel Annual Report 2018 -

Solution

Telkomsel worked with Merkle Indonesia, a Google Analytics 360 Certified Partner. The process went from an advanced Google Analytics 360 implementation to use audience segmentation using machine learning algorithm, which can be used for Google Ads (previously Google Adwords)

In BigQuery, three months of historical data both for data packages and voice packages separately has been taken into consideration to build an audience based on machine learning algorithms. Telkomsel has set the target ROAS as 15+ for this campaign. Next, segmenting the audience based on their behavior on the Android and iOS apps will bring more value to the audience configuration and spot the high-value users. Nearly 20 features were selected based on GA 360 data with hypothesis & industry knowledge to segment audience behavior. Here’re the required inputs for running the model:

• Recency: When was the last transaction?
• Frequency: How often do they usually transact?
• Monetary: What is the average value of transactions?

A binary classification model using ‘AutoML Tables’ on Google cloud has been created based on the selected features to identify the high-value customers which can be targeted for the selected Telkomsel internet and voice products generating ROAS more than 15%. Finally, the audiences for Android and iOS have been exported from BigQuery and uploaded to Google Ads for running campaigns.

Result

The impact of the machine learning audience from BigQuery onto the Google Ads performance was beyond great. With the additional layer of audience targeting, Telkomsel uncovered more than 10 times improvements in buying packages conversion after running the campaign with the BQML Audience. Though BQML Audience is time sensitive, so it should be timely set and should see a drop in performance time over time because of the audience relevancy.

We ran a Universal Campaign with the BQML Audience and the result were surprisingly satisfying. By looking at the performance, we could use it on different types of campaign, like Search or Display campaigns. Scaling up the campaign can also be done with higher budget and longer period, but we must make sure that the audience will be refreshed accordingly.