How to use data science to understand customer emotions and decisions
A few sentences jump out at me, from an interview I conducted recently with customer-experience expert Peter Dorrington.
"Behavioural economics and psychology show us that much of human decision-making is based in the world of emotion and cognitive bias, not logic," Peter observes.
Businesses desire to make use of this insight, so "We’ve moved down a path to determine how to operationalise emotion, putting core insights on emotion into practices as we design, build and operate elements of the customer experience... This is not a prediction of how the customer will feel in the future; instead, it is an explanation of what they are likely to do as a result of feeling that way."
Peter is director of customer insights at TTEC Digital, a global CX consultancy. He graciously agreed to an interview, in the run-up to the 2018 Sentiment Analysis Symposium in New York, where Peter will be speaking on March 27 on "Emotion Analytics: The Need for Artificial Emotion in Customer Engagement."
The conference premise is the link between emotion and customer satisfaction, loyalty, motivation, and advocacy; speakers will examine the application of sentiment analysis to understand and exploit market drivers, whether the markets are consumer, media, or financial. Peter's insights here are significant, relating how business can...
Use data science to understand customer emotions and decisions
Seth Grimes. You're director, customer insights at TTEC, globally one of the largest consultancies focused on customer engagement. What does your work there involve, that is, what are your day-to-day and strategic roles?
Peter Dorrington. I joined TTEC as an expert in the application of data science, especially predictive analytics, with responsibility to solve business challenges for our clients. My focus included driving outcomes for our clients - find, retain and grow more customers; improve business efficiency and effectiveness; adapt to a changing world. These common outcomes led to the development of competitive strategies to support 'commoditised' industries and sectors through enhanced customer experience and was the genesis for development of the Customer Experience Vector (CXV). Through use of this proprietary methodology and analysis, we work with enterprise organisations to develop strategies to operationalize emotions. Essentially, we are using data science to understand more about customer emotions and decision-making in a way that can be built into the company's systems and processes at the day-to-day operational level.
Industry insightsView more
Seth. TTEC works across a spectrum of industries. Are there common threads in the company's projects, and what's distinctive about TTEC's approach?
Peter. Our proprietary Humanify Customer Engagement as a Service offering is inclusive of people, process and technology. We help large global companies increase revenue and reduce costs by delivering simple and personalised customer experiences across every interaction channel and phase of the customer lifecycle as an end-to-end provider of customer engagement services, technologies, insights and innovations. We provide our outcome-based customer engagement solutions through TTEC Digital which designs and builds customer experience consulting and technology solutions and TTEC Engage which operates customer care, growth and trust and safety services. Our company purpose is to bring humanity to the customer experience.
Seth. The Customer Experience Vector is a method you devised that links data science and behavioural science. How does it work – the inputs, algorithms, and outputs? How is it applied?
Peter. As someone with a data science background, I was very familiar with the kinds of analytical techniques that are based on observable behaviour and characteristics - who, what, where, when, etc. But I was also aware of what was missing in explanations / predictions of customer behaviour - the why behind the what. Behavioural economics and psychology show us that much of human decision-making is based in the world of emotion and cognitive bias, not logic. Problematically, emotions are complex, rather tricky things to deal with and often defy traditional analytical techniques that work so well elsewhere. We’ve moved down a path to determine how to operationalise emotion, putting core insights on emotion into practices as we design, build and operate elements of the customer experience.
This is achieved by analysing customer narratives using Natural Language Processing techniques to surface emotions associated with encounters / touchpoints. Although the approach worked well for individual customers, it proved difficult to extrapolate into longer-term predictions. To adjust for this, we began to measure aspects of a fleeting 'moment' and correlate it into part of a longitudinal journey - which is where the vector maths comes into play (a way of summarising a whole customer history into a few key numbers). That removed the need to build predictive models to 'explain' the causal relationships between complex histories and outcomes (which is vital if you are dealing with millions of customers and billions of individual transactions).
Computers don't understand emotions, but they can be shown examples around an arbitrary concept (such as emotions) and be taught to recognise the weak signals and indicators that identify them.
Next, we worked out how to apply that to a whole customer population, not just the initial group analysed – using data on what has happened in the past, how that's likely to be perceived and how it will impact future decisions. This emotion data is appended to the customer record. This is not a prediction of how the customer will feel in the future; instead, it is an explanation of what they are likely to do as a result of feeling that way. The final step is to embed those insights into operational systems. For example, suppressing qualifying customers from a specific marketing campaign because they are emotionally unreceptive - switching them to a different action track - as one client said, 'stop marketing to customers who don't want to hear from us right now and focus on what they do want'.
Seth. In your Sentiment Analysis Symposium presentation description, you talk of "evolving AI from Artificial Intelligence to Artificial Emotion" and about "the science of emotion." What's the state of "the science of emotion" and what do you mean by "artificial emotion"?
Peter. To be clear, computers don't understand emotions, but they can be shown examples around an arbitrary concept (such as emotions) and be taught to recognise the weak signals and indicators that identify them. The aim is to give automated systems the ability to listen to and understand the emotional subtext in a dialogue and do so at a level of sophistication and scale that humans can't replicate. If you like, we're attempting to give computers some EQ to go with their IQ.
Of course, the natural follow-on to this is that the organisation also must be good at acting on the insights - listening is only half the equation. I like to think about it as putting the 'relationship' back into customer relationship management and we all know that the best relationships are far more than meeting each other's rational needs.
Seth. Do you automate narrative analysis? How do you handle narratives – customer journeys – that involve divers interactions over time, via multiple touchpoints?
Peter. Firstly, not all narratives are equal - it is possible to have a conversation that is completely functional (e.g. being asked to provide a factual answer to a question). Also, it is really important to listen to the authentic Voice of the Customer, not a paraphrase or a synopsis. Finally, we need to strip out 'noise' - for example, splitting the narratives of the parties engaged in conversation and analysing them separately. We do that by asking very specific questions - those designed to illicit a “customer story.” These activities can all be automated, as can be the original analytics.
The second part of your question addresses one of my early frustrations - that too many journey analytics tools are essentially linear in nature; customers start at point A, progress through B and C, ending up at D. If there are decision branches, these are just different linear paths. My experience is that, when we look at the totality of a customer's engagement, it is rarely so linear. One of the advantages of CXV is that it accommodates the concept of an 'experience landscape' - there are lots of opportunities to encounter a touchpoint or event but they can come in any order or timescale. As a result, a customer's CXV is unique to them and reflects their personal journey, no matter how long or how complex.
Seth. What technology is most critical to TTEC's success, for AI and big data and emotion analyses?
Peter. Clearly the first requirement is to be engaged with customers - whether that be in person, via digital / mobile or in store - all of which serve as potential sources of narrative or transactional records. This data fuels the analyses. Next is the ability to extract emotional meaning from narratives using advanced natural languages tools, ones that don't just categorise conversations or answer questions - this is predominantly the world of AI and Machine Learning. We then work to calculate CX vectors for every customer, every day, that are unique to them and actionable. Finally, we turn insights into action by informing operational systems in everything from choosing next best action to determining optimal channel and which associate is best positioned to manage the interaction.
Seth. My usual closing question: What's next? What will you be focused on two years hence, and five?
Peter. There is a great temptation to overthink this field. I am always at pains to explain to clients that emotions are messy, complicated and impossible to predict (unlike the decisions a customer will make a result of how they feel). 100% certainty will always be impossible to achieve and exponentially more difficult the closer you get, so the future is not going to be about an endless search for greater accuracy but about ways to operationalise these new techniques. For example, thinking about employees in the ways we think about customers or giving experience designers new tools to optimise around the (emotional) human and not just about the experience.
Meet Peter Dorrington at the 2018 Sentiment Analysis Symposium in New York, on March 27, when he will speak on "Emotion Analytics: The Need for Artificial Emotion in Customer Engagement." See you then?
You might also be interested in
Seth Grimes is the leading industry analyst covering natural language processing (NLP), text analytics, and sentiment analysis technologies and their business applications. He founded Washington DC based Alta Plana Corporation, an information technology strategy consultancy, in...