A major Italian auto insurer wanted to use a data driven approach for identifying targeted customers to market custom insurance solutions to based on their individual characteristics. In order to tailor offers and marketing campaigns that would ensure high response and conversion rates, the Marketing department relied on the insurer’s Risk Management group to use the company’s own big data and doolytic for segmenting and identifying individual target customers.
In order to support Marketing, the Risk Management group needed to update their algorithms for calculating premiums with more information than was available in standard customer profiles. With the new algorithms created with doolytic, business users can use interactive dashboards, graphics and frequent queries to observe data that has been aggregated and filtered for them. More sophisticated citizen data scientists now obtain highly granular data and drill down into it on the fly in order to discover new relations and rules. They can also explore the data matrices obtained through Data Enrichment with the Relation-Action model so that underwriting decisions can be made based on expanded risk factors. The insurer’s expert data scientists, using doolytic’s integrated notebook computing tool, can now identify customer clusters based on their driving behavior by relating their profiles to existing profiles calculated from real accident data. As a result, the insurer can now profile risk with a much more sophisticated model, enabling smarter and more profitable business decisions.
doolytic enables Marketing, by working with data scientists, to segment their customer base according to driving behavior linked to six categories: The “average” driver with no distinctive characteristics, “commuters” who drive long distances on weekdays, “weekend” drivers who drive long distances only on weekends, “occasional” drivers who drive short distances on weekdays, “Sunday” drivers and “speeders” who regularly drive at higher than average speeds. These clusters are associated with different risk profiles in order to plan offers based on the real characteristics of their customers, rather than simply age, sex and profession as traditionally calculated by insurers.
Insurance Black Boxes can generate up to 10 million records a day describing temporal, location, speed and vehicle status data. doolytic enables the data scientists supporting the Risk Management group to receive data that is clean, filtered and pre-aggregated for specific data models in order to generate the necessary datasets for their internal customers. With this data, risk managers define specific premiums and insurance clauses for use by the Marketing group in formulating tailored offers.
A large European multi-brand department store chain uses predictive forecasting from store data and external factors to stock the right styles and sizes across more than 1200 stores. With such a large number of retail outlets and products to manage group-wide, synchronizing stock to sales in real time to monitor supply chain logistics data has provided a real competitive advantage. Trends that had been impossible to identify without dynamic forecasting can now be detected. Assortment planning and logistics are handled in completely new ways that would not have been possible before.
A multinational drugstore chain uses loyalty card big data for end to end marketing analytics, enabling inventory forecasting across all their stores. doolytic supports the supply chain and logistic processes, furnishing accurate estimates for varied scenarios of customers and product categories, ensuring that each store has the right inventory with greatly reduced stock risk for the entire chain.
A national retail group uses big data from loyalty cards for product level forecasting with associated marketing campaign planning. With doolytic, planning for stock levels now takes into account both the calendar and the seasonality of the products themselves together with historic response to direct marketing. Targeted seasonal campaigns now bring the right, high margin, customers into stores to shop for specific in-stock products.
A large European retailer has been able to analyze historic marketability curves of their huge product range. This clearly indicates the speed at which an item sells in a certain store and measures the sales performance of a particular store, and allows for analysis against different variables. Such extrapolations have enabled a huge qualitative leap.
doolytic dynamic forecasting complements traditional demand and supply chain forecasting, making predictions of future observations based on past and present business data. Retail specific predictive analytics takes additional external factors into account – for example traffic conditions, weather forecasts, calendar events, sensor data – allowing retailers to ensure that the right stock and personnel are at the right place at the right time. doolytic leverages "ensemble and consensus methods" algorithms for ensuring that the best model is used for each specific combination of input factors. Every prediction is cross-validated and labelled by rating quality and forecast accuracy, ensuring that doolytic provides the fewest possible forecasting errors. Multidimensional, automatic, high granular forecasts can be explored interactively or batch orchestrated with high throughput. Future predicted observations serve demand planning processes and provide a solid decision support. Calendar and promotional scheduling effects are analyzed to provide what-if scenarios for marketing planning. Each forecast shot is stored for debug and investigational purposes in order to fine tune data-driven models and learn how seasonal and cyclic patterns impact business operations.
Recommendations & Market Basket
Familiar to anyone who has purchased online, recommendations propose products based on extrapolations from what other customers have bought. Sometimes called "best next offer" recommendation, they can benefit from a much broader context, not only analyzing the most likely combinations, but also, based on a very fine-grained "graph analysis," identifying a closely related peer consumer group based on a multitude of factors. In the social context, recommendations also work with people, such as when Facebook or LinkedIn recommend other people to connect to based on existing social networks.
An extension of Recommendations, Market Basket analysis traditionally matches products that are usually purchased together. Big data adds more context to these recommendations, including time of day, calendar, store ambience variables (music, lighting), store visit duration, weather, staffing levels, length of queue and similar.
Loyalty management in the retail space benefits from big data by extending channel reach from point of sale, web and call centers to include mobile and social capabilities. Rewards may be accrued through purchases and other methods, including acting as product or brand ambassadors and leveraging social relationships (descriptive analytics), which logically relates to the most widely adopted big data use case in the retail sector, Recommendations, as previously described.
The main goal of loyalty marketing is the growth and retention of existing customers through the use of incentives. One of the most valuable and strategic aspects of this type of marketing is analysis of available data to reveal a clear picture of consumers’ behavior, to estimate their value and their potential value, and to evaluate their responses to marketing campaigns. Predictive analytics allows better estimation of some key customer value metrics:
Network planning departments at mobile operators need to optimize bandwidth allocation among users and by the various types of traffic traveling over the network. Typical network equipment generates billions of log records daily for traffic and bandwidth usage, which is key data for creating effective network optimization rules. However, the existing technology stack limits data scientists to utilizing high aggregation levels and small fractions of data while performing statistical analyses. doolytic solves this problem by automatically ingesting and indexing network monitoring big data into the doolytic back-end on a daily basis. This enables business users to keep track of frequently used queries through interactive dashboards and responsive visualizations. Citizen data scientists are able to independently drill down into big data with maximum granularity on-the-the-fly at user, device and traffic package levels. They can discover new paths and rules for network optimization through doolytic’s Relation-Action model, which allows them to use wider and deeper sets of inputs to enable more accurate and effective network optimization algorithms. Telcos are further aided by doolytic’s Anomaly Detection capabilities, which enable business users and citizen data scientists alike in analyzing network flux for identifying anomalous traffic usage patterns.
A global telco’s data science team needed to identify clusters of users based on quantities of correlated network traffic. The data would be used to create new dashboards for the telco’s business users. Subsequent feedback from the customer service department was that several customers had made claims related to the “Spotify” and “WhatsApp” services over the past month. The data science team needed to understand if these claims were related to an identifiable network issue.
The network team also had a query for the data scientists. They wanted to identify which users had disproportionately huge amounts of network traffic over the past month, and which specific services generated the traffic. The data science team built on the original query in order to create a technical join with the traffic details table so that they could reuse it for future big data discovery.
Now, doolytic allows users a deeper dive discovery than previously possible with preaggregated data. They can iterate on data analysis to the highest level of detail with unprecedented speed and ease while a more accessible timespan enhances statistical support for analysis and algorithms
Network Planning ROI
Internet of Things (IoT) is one of the main technological trends in the recent years. It allows real-time machine-to-machine communication over the internet. Its application in the industrial domain – the so-called Industry 4.0 and Smart Manufacturing – is at the core of the innovation agenda for those who are committed to redesign the organizational processes of the industry of tomorrow. For these reasons, doolytic roadmap and features are designed to support the needed transformation throughout the entire data life cycle, from the ingestion of the data to the predictive analytics. One of the major goals of the Industrial IoT is the automatic monitoring and detection of abnormal events, changes and drifts on the collected data. Anomaly detection includes all the techniques aimed at the identification of data patterns who deviate from the expected behaviour (the norm). There are several approaches and they are typically linked to specific domains. They are the natural counterparts of the traditional dashboarding feature.
Anomaly detection features are pre-packaged and can be invoked in an interactive manner. Alternatively, they can be set up to monitor hundreds of metrics (sensors, systems, processes) and to react to each detected anomaly. To reduce the number of singleton, doolytic use supervised machine learning techniques. Analysis pipeline and models who are able to classify true anomalies thanks to the joint observations of all the relevant process-related metrics.