2018 BAI Global Innovation Award Finalist
Innovation in Marketing
Isbank – Turkey
Isbank’s Self-Learning Marketing Hub
Isbank formed a self-learning marketing hub which creates relevant, contextual, consistent, and personalized interactions with customers across all channels by employing advanced analytics, machine learning and real-time data.
The hub analyzes a complex set of structured and unstructured customer data, derives crucial customer insights that drive coordinated real-time decisions, and intelligently provides the right offer to the right customer at the right time despite of the number of our customers, complexity of channels and the number of offers presented. It also integrates adaptive decisioning models and propensity-to-buy models which were developed in-house and learn from customer responses, becoming more intelligent about how to interact with customers effectively.
Over the years, we have developed an integrated campaign management infrastructure which employs Isbank’s online banking, call center, branch network, ATM, IVR, teller, e-mail and SMS channels. This infrastructure allows us to deliver a seamless customer experience through integrated multi-step, omnichannel, personalized outbound and inbound offers. However, in today’s marketing environment, we face with an explosive growth of data variety, volume and complexity. In order to incorporate valuable customer insights, traditional structured data is not enough. Now, the key is to leverage unstructured data from various sources such as location, social media, websites and many more. And above all, data is fast. Taking real-time actions in a contextual manner is one of the key elements in today’s successful marketing.
Aforementioned issues are the main drivers of our desire to transform our sophisticated marketing and campaign management infrastructure into a decision hub which unifies marketing decision management consistently and acts like a brain with memory and predictive intelligence. In the meantime, we wanted to participate in even more intelligent, contextual and personalized interactions with our customers by employing advanced analytics, machine learning and real-time data.
With our self-learning marketing hub, we can manage marketing strategies and handle all of inbound and outbound offers by employing only one tool. The hub creates relevant, contextual, consistent, and personalized interactions with customers across all channels. Complex decisions are defined with dynamic rules, and strategies are run automatically. The hub connects every channel and touch point, and provides a smooth customer experience by handling omnichannel and multistep offers efficiently.
Despite the number of our customers, complexity of channels and the number of offers presented, our hub intelligently provides the right offer to the right customer at the right time. It incorporates state-of-the-art machine learning algorithms which allow for adaptive, real-time decision-making. It learns from customer responses and becomes more intelligent about how to interact with customers effectively. For every offer and every channel, adaptive decision-making models are created automatically, and these models update themselves with every customer response to offers presented.
Along with structured data, the hub also listens to big data, decodes it, finds meaningful patterns, and acts in real-time. These big data streams may come from customers’ digital footprints on our website or credit card transactions, indicating products or services they buy or may be interested in; from customers’ mobile devices indicating their location; or from social networks indicating their life events or their interests. Whatever the source is, it turns streams of big data into valuable business decisions, offers and actions. Apart from being an important element for capturing the context of the customer, big data is utilized in customer analytical record, predictive models and offer prioritization.
The hub employs Hadoop ecosystem products for capturing, storing and processing massive amounts of unstructured data, and event based action engine for transforming these big data into action and creating an insight about the customer context.
Our self-learning marketing hub facilitated the sale of 1.3 million products in 2017.
Before the launch of the first phase of our project, average monthly positive soft response was 750,000. After the launch, the percentage of correct guesses of a customer’s propensity to buy for adaptive models exceeded 80 percent and average monthly positive soft response rose up to 960,000, with a 32 percent increase in performance.
We have recently undertaken an analysis to compare the results of adaptive decision-making models vs propensity-to-buy models. With an accuracy rate of 87 percent, adaptive decision-making models proved to be more accurate than propensity-to-buy models, which are 72 percent accurate. In addition, recall rate of adaptive decision-making models is 24 percent, whereas that of propensity-to-buy models is 21 percent.