Big Data challenges for business Email: Giles.Pavey@dunnhumby.com Twitter: @GilesPavey © dunnhumby 2013 | confidential 2 © dunnhumby.com 2013 | Confidential dunnhumby is a world wide big data company 3 3,000 660 MILLION $500 BILLION EXPERTS CUSTOMERS GLOBALLY RETAIL SPEND © dunnhumby.com 2013 | Confidential dunnhumby’s blueprint for business success through customer centricity c.2005 DM responses Customer KPI's Prospecting models Customer journey Capture changes in customer behaviour via data and begin the journey again 4 © dunnhumby.com 2013 | Confidential Product development Pricing Strategy ATL Advertising Retention programmes Retail strategy Changes in Customer behaviour Call centre data Propensity models Action Research results Customer segmentation link Demographics link Data Transaction history Insight Acquisition DM 5 © dunnhumby.com 2013 | Confidential RISE OF THE MACHINES 66 © dunnhumby.com © dunnhumby 2014 2013 | Confidential Sensory overload? Accelerometer GPS Gyroscope Wi-Fi Magnetometer Bluetooth Barometer Microphone Proximity NFC: Near Field Light sensor Camera (front) Touch screen Camera (back) 14 Sensors! 7 © dunnhumby.com 2013 | Confidential dunnhumby’s blueprint for business success through customer centricity TODAY DM responses Click Stream GPS Social Media Mobile Apps + many more Customer KPI's Prospecting models Customer journey Machine Learning Graph Theory Neuroscience Agent Based Models + many others Capture changes in customer behaviour via data and begin the journey again 8 © dunnhumby.com 2013 | Confidential Product development Pricing Strategy ATL Advertising Retention programmes Changes in Customer behaviour Call centre data Propensity models Action Research results Customer segmentation link Demographics link Data Transaction history Insight Acquisition DM Retail strategy Recommendation Personalisation Mobile offers Multi-channel + many others Customer advocacy 9 © dunnhumby.com 2013 | Confidential 10 Complementing Customer Data Insight with Customer Data Science 10 ● Historic Performance ● Prediction ● Batch Processes ● Real Time ● Structured data ● Polystructured data ● Reports & segments ● Data Products ● Classical Statistics ● Machine Learning ● DIY ● Crowd-sourced ● Known unknowns ● Unknown unknowns © dunnhumby.com 2013 | Confidential Big data opportunities © dunnhumby 2013 | confidential 12 Sampling: ● Rigorous random sampling is increasingly too hard and expensive to do. ● Large biased samples are increasingly plentiful – e.g. followers on Twitter; Friends on Facebook. ● How can the bias be accounted for? 12 © dunnhumby.com 2013 | Confidential 13 Forecasting: ● Tens of thousands of new products and variants are launched in the UK every year. ● Fast diagnosis of a potential hit or miss can make or save £1,000,000’s. ● Big Data provides unprecedented potential for exploration and prediction. Performance of similar products; customer sentiment; repeat behaviour ● Can we quickly and reliably predict future success? 13 © dunnhumby.com 2013 | Confidential 14 Recommendation: ● We can build predictive models of customer behaviour by looking at both their individual history and those of customer “like them”. ● Within the history will be some time dependent predictive purchases. – e.g. iPad and iPad cover; charcoal and BBQ burgers ● These models can be further improved by combining in additional “real time” contextual data. ● How can these factors be combined to build the best recommendation? 14 © dunnhumby.com 2013 | Confidential 15 Simulation: ● dunnhumby have partnered with Sandtable and are developing Agent Based models of human behaviour ● The models have 3 elements: – Environment – Agent characteristics – Agent behavioural rules ● The models can become incredibly complex, incredibly fast ● How can we most efficiently compute the simulations? ● How can we best validate the models? ● For a validated model, how can we best optimise the environmental factors to give the best future results e.g. Marketing Mix Optimisation? 15 © dunnhumby.com 2013 | Confidential 16 Online anomaly detection: ● Not only is Big Data large it is also often streaming and instantly accessible ● There is potential for building models that detect anomalies in these streams of data. Therefore enabling us to intervene and rectify an unwanted situation ● One example, how can we use real time sales data to identify when certain products within a store have sold out? 16 ● Or, can we identify the signs that a customer’s behaviour in the store or online indicates that they are likely to stop shopping with us? © dunnhumby.com 2013 | Confidential 17 Network flow: ● Customers are increasingly well digitally connected to each other. – e.g. Facebook, Mumsnet ● There is a great opportunity to use these networks to increase retail efficiency by better matching customers with relevant products. ● However, there is a risk of offers going “too viral” ● How can we predict the final reach of an offer across a network from an initial communication? 17 © dunnhumby.com 2013 | Confidential Questions? © dunnhumby 2013 | confidential