Guest post written byOlly DownsChief data scientist at Globys; his team builds marketing products and employs advanced machine learning and data science techniques.
Although gut and intuition are foreign terms in the field of data science, we can appreciate the curiosity that is commonly exhibited by effective marketers. That's one thing that scientists and marketers have in common - an eagerness to learn, try new things, and adapt as new information emerges.
What marketers have lacked is the ability to both test their intuition in a systematic way and to learn about customer behavior and response at a scale that matches the evolution of their thinking and of their business. Knowing that a small segment of targeted customers showed initial response to an isolated marketing offer is interesting. Knowing how thousands of unique marketing treatments are impacting the long-term goals of your business across an entire customer base of millions is transformational.
By fusing the creative nature of marketing with the methodical nature of science, we're able to rapidly evaluate and optimize existing strategies which are otherwise the result of assumptions and partial results. Offering an even greater advantage is the organic discovery and communication to the marketer of customer behavioral learnings which drive the development of new strategies that otherwise would never be perceived. Today these learnings are garnered by the marketer having their Business Intelligence team work on one-off projects to look at low-dimensional behaviors of their customers. This new approach is about learning, testing, and optimizing over time - continually uncovering and acting upon new insights - not just finding something that works and running it on auto-pilot.
So what are the core elements that are being used with this scientific approach to marketing? Here's a quick snap shot of what they are, how they work, and the benefits they provide.
1. Scalable data architecture - Enable exploration2. Application of advanced data science - Understand customer behaviors
Big data technologies play a key role in establishing a framework that is capable of scaling horizontally - consider a petabyte of data for a hundred million mobile subscribers, flowing at 5 billion transactions a day, and thousands of machines - and capable of supporting the interactive process required for programmatic data science exploration. The tried and true yellow elephant and the Hadoop ecosystem enable simple integration points and handle the fast batch processing required for executing against the marketing strategies while also outputting learnings for ongoing exploration.
Automating the computational discovery of a variety of behavioral attributes, as simple as outgoing voice usage and as complicated as the Fourier-coefficients for someone's usage and billing behavior over time, enables marketers to expand their understanding of how individual customers behave.
3. Dynamic experimental design - Determine customer motivators
This deepening of customer profiles also enables rich behavioral clustering schemes that correlate customers based on like behaviors to determine 'what is common' and 'what is not common' for certain behavioral types.
For example, in the telecom industry, mobile operators are keen to understand the entire history of behavior that ultimately leads to churn. Although common assumptions would focus on a significant drop in usage at a given point in time, science indicates otherwise, guiding marketers to hone in on several multi-dimensional behaviors, such as who a customer interacts with, at what frequency, and in which direction, and contexts, such as when a customer uses voice versus SMS versus third-party messenger. The goal of this exploration is not to identify 'at-risk' customers to save, but to understand what behavior to look for and when in order to prevent. This understanding becomes core to determining how to engage with customers in the most effective way at an individual level.
Clean measurement is at the heart of the design of every science experiment. Marketers are used to having to design A/B tests of their intuition by hand, limited to the level of complexity at which a human can think, that are then executed serially. Dynamic experimental design is enabled by new technologies that make hypothesis testing obsolete.
Taking into consideration previous events and a holistic understanding of internal and external influences - this approach leverages dynamically-assigned target and in-context control groups to ensure a measurement from the comparison of equally-qualified treated and untreated individuals in an experiment.
4. Stochastic offer decisioning algorithms - Optimize marketing performance beyond 'Next Best Offer'
As opposed to traditional campaigns, which are typically measured by simple, short-term KPIs, such as total response rates or global metrics, such as the total revenue on the days the promotion was running, this approach allows the coordination of target and control assignments over time, across multiple experiments and the measurement of delayed KPIs that get to true ROI for the business - incremental non-cannibalized revenue and extended customer retention.
Surprisingly, for many marketers, even the notion of using contextually-selected target and control audiences is a leap forward in their marketing practices. In the retail space, for example, marketers have fully embraced a cross-channel engagement approach yet the lack of sophistication required to discern the impact of a specific sequence of marketing treatments (message + offer + context) on customer behavior often leaves marketers relying on assumptions.
Dynamic experimentation coupled with robust measurement ensures that a retailer can accurately determine what motivates - or doesn't motivate - customers to respond or take action. Was the 'hot trends' email message enough to prompt purchase or was a 10% off SMS message necessary to spur action? Did the 'our picks for you' message increase the time spent browsing the website this month or was it the 'thank you for your order' that resulted in the additional visit? Which drives long-term purchase behavior - a forwarded promotion from a friend or a direct message from the company? These are the types of questions that data-driven marketers who are turning to science can answer - for each individual, across millions of customers.
The adaptive nature of this approach enables marketers to simultaneously exploit and learn the optimal marketing interactions for maximizing customer lifetime value. This so-called 'multi-armed bandit' approach refines its expectation for the performance of a given interaction with a given individual over time, as marketing treatments are tested and measured. This enables marketers to quickly hone in on the best message to send to a particular user in a specific context and their expected future contexts, and for that 'best' to continually evolve over time due to customer, product, competitive or economic dynamics in the business.
As the basis for a new kind of marketing that's driven by science, these core elements are helping to accelerate the channeling of marketing creativity by providing a systematic, closed-loop process for testing and exploiting what works and eliminating what doesn't work, and exposing newly discovered insights about behavior, audience, language and interaction contents, to inform and drive further creativity. And in my experience, although marketers may enter with the mindset of proving their intuition right, they quickly change their focus when exposed to the vast array of possibilities yet to be employed.
Entities 0 Name: Business Intelligence Count: 1 1 Name: Globys Count: 1 Related 0 Url: http://ift.tt/1oBCzwL Title: The Secret to Customer Engagement? Shrink Your Big Data Description: Across the world, there is a tectonic cultural shift redefining customer behavior. Today's customers are doing more, demanding more and engaging with companies on their own terms. Businesses are being forced to rapidly change the tactics they use to reach out to and serve customers as they come to grips [...]
Post a Comment for "How Science Is Helping Marketers To Prove Their Intuition Right...Or Wrong - Forbes"