Inanc Cakiroglu, Director, Artıfıcıal Intellıgence, Data Analytıcs and Data Scıence at Turkcell
Analytics has advanced rapidly in the last decade in terms of technology behind, vast increase in volume of data created-captured with more in digitalized data sources and use cases spread through various business areas. Almost all pillars of advanced analytics applications have their own evolution process in their life cycles however these evolutionary paths and level of advancements may differ based on value created on customer side or marginal contribution on operational excellence of the organization. Evolvement of an analytical application is not only limited to application’s own function space; all integrated and correlated technologies and services also exposes changes through fast varying trends and needs. When we look at most decisive dimensions of changes that shape analytical use cases and application fields, following factors appear on top of the list:
• Digitalization level of the application area
• Availability of diverse variables on data
• Integration of analytics with industrial AI services
• Smart blend of deterministic rules with predictive models
• Streaming data process and real-time capabilities
It is not a surprise for any analytical expert to see the time dimension as one of important agents in fact increasingly scaling up in the field. People lives are consisting on series of corelated events so called decision moments which are consumed in real-time. Therefore, it is not astounding thatanalytical applications appear to create higher value in those captured real time moments. Let’s take deeper dive into that topic.
DATA IN MOTION: REAL TIME ANALYTICS
Early analytic applications were designed to process batch cold data due to absence of streaming data and lack of necessary technologies to get benefit of data in motion. Complex event processors were first examples of real time consumer interaction management concept. Advances in mobility and internet technologies lead creation and capture of more digital events which basically constitute main fuel of complex event handler applications. Application fields are various, any kind of trigger: download of an app or usage of it, designated consuming amount on data package of a tariff plan, change in location, purchasing, changes in consuming behavior… any kind of interaction event can be tracked and transformed into action by complex event processors. Main idea is basic, people tend to response more -and response positive- when interaction is set in real time of associated event. We see this concept to prevail in some business fields even more such as purchasing of an insurance.The beauty of engineering takes place when multiple interaction points are designed to run in a scenario. Again, take an example of triggering purchasing insurance: It is the exact right point to promote household appliance insurance offer when you detect that a consumer calls TV brand technical service number (event3) for installation appointment after navigate around multiple TV brands’ portals(event2) and e-commerce web sites. (event1)
Decision points which shape mechanism of real time analytical action scenarios are determined by intuition, existing knowledge or set as a result of descriptive analysis at the beginning. Now, machine learning models are started to take active role establishing perfect action triggers and configuring them. Which is more effective to promote additional data package, whether when a customer consumed 75% of its current tariff plan or when she ends up 90% of it? Is it same for all your customer base? A simple machine learning algorithm can answer that question for you even differentiate your customers based on their propensities and establish different rule sets for clusters in your consumer base.
The better return can be achieved when your OLTP based complex event processor described above, starts to interact with an OLAP platform which is commonly called “Real Time Interaction Management” (RTIM) application. Real Time Interaction Management concept is a beautiful tool for your growth managers. In the existence of multiple customer interaction channels (face-to-face shops, call center, web site, app etc…), RTIM promises perfect orchestration of your offers across those channels with consistency and differentiation if needed. It answers very crucial question of “Which channel is the most effective to contact for that particular customer and when this contact should be established?” Even more, what is the right tone of communication approach for that customer? You may differ your promotion letters based on personality trait of that customer.
Real time modelling is another concept which some of real time interaction platforms offer. They can even utilize batch data models alongside of real time models parallelly and based on success factors of modelling KPIs (accept rate of best offers for example) they can prefer and shift which models output will be set for the consumer. Determining and utilizing the most popular offer is an effective example of real time interaction management modelling which learns from instant demand in the field.
Best customer – best agent matching use case is also another interesting use case of real time analytics. The idea is based on a hypothesis that intelligent matching of certain customers with certain customer services representatives gives the best result in terms of customer satisfaction and successful resolution of interactions. Best customer-best agent model uses demographic data of consumer and service representatives, transaction history, reason of call or interaction and learns from performance KPIs of previous interactions to create perfect agent match for that particular customer and specified reason of call in real time.
Best customer-best agent modelling is somehow a way to create digital persona of your customers and your service representatives. Then we come to another emerging and promising topic of advanced analytics, which is The Digital Twin.
Digital Twin is a concept supported by machine learning models, based on creating digital version of your customer base in a simulation. This digital copy of the customer may include demographic data, propensities, risk scores (churn, fraud, credit risk), personality info (i.e. OCEAN scale), predicted best offers, reason to contact predictions, satisfaction and digital experience scores, transaction history, channel and contact preferences and so on. Models learn from transactions and activities associated with customers trying to predict further activities on customer side. When you finish creating all your customer base into digital twins, you can predict not just individual actions of your customers, mass movements and reflections of your customer base as well. It is a perfect simulation environment that can shape your marketing sales strategy and field activities.
Real time analytic is one of the best approaches to get benefit of your data. The competition for customers is increasingly focused on capturing and keeping their attention which is largely shaped by their experience. Moreover, the demand from customers is changing rapidly. Actions should be implemented quickly to meet with these rapid changing demands. Data in motion changes everything: allowing businesses to align with their customer demand patterns and creating best possible experience scenarios.