Grab has built an in-house self-service consumer data solution C360 using tools from US enterprise software company Databricks to achieve a single source of truth for consumer-centric attributes that helps it to offer personalised experiences to customers.
It has also developed advanced customer lifetime value (CLV) prediction models on Databricks Lakehouse Platform to understand consumers across various product segments and functional teams. The CLV is an estimation of total future revenues from a particular customer.
Grab’s Lead Data Scientist, Zulfikar Lazuardi Maulana told the Destination Lakehouse ASEAN forum that the firm could now take advantage of consumer data and build a consistent understanding of customers with C360. “The internal portal could help different teams to collaborate and explore data attributes.”
He added that the company can now "effectively" make more personalised recommendations based on consumer preferences to improve in-app experiences.
Grab, with millions of users across 480 cities in eight countries, has generated over 25 billion transactions, for transport, food and grocery delivery and digital payments purposes.
Maulana said the company needed consolidated data to build a unified understanding of consumer behaviour. It has federated teams working on various departments like campaign segments, personalisation, need-based systems and upselling/cross-selling services.
The firm has suffered from inconsistent customer views as teams developed their own custom definition and built their own models without leveraging the full set of user attributes available, thus requiring heavy maintenance, he noted.
With Delta Lake, the company can now ingest and optimise user-generated signals and data sources from their websites and applications within hours, which had usually taken a few weeks.
Along with an Attribute Discovery platform and self-service API portal, Grab has also been able to build new consumer features quickly.
For instance, at production, the company engineered a new feature for the contact centre team to make available the consumer segment information to agents, when consumers called in.
“We are also experimenting to try things like OTP (one-time password) flow to avoid frauds,” he said.
Grab is using Databricks as a main compute engine to create, extract, transform, and load (ETL) pipelines together with Azure Stack like the Data Lake, ML flow and Data Factory.
"Through this, we could provide multiple channels and tools for our team to perform 64 different use cases", he added.
CLV models
Grab has worked on CLV models to provide the right investment into each customer in order to create personalised offers, save tactics, and experiences.
Grab’s head of Data and Analytics, Nikhil Dwarakanath, said the company could address two major challenges of churn prediction and lifetime value prediction by setting these models up in different iterations.
We are running many model refreshes over two years now and (we) also store CLV inferences back in our data platform, he added.
These inferences are available for engineering and data science teams enabling them to query for insights and other use cases. For example, this system could recommend an investor withdrawal action at the user level for a particular marketing campaign.
While the solution is helping Grab’s several teams within the organisation with different kinds of use cases in offline and production, Dwarakanath said they have reached the spot after going through the process of experimenting and validating consistently.
“If you have two models, the one with fewer features is probably better, assuming they both have the same level of accuracy.”
He also recommends starting with simple solutions in the first iteration rather than building sophisticated data science workflows.
