Share this content
Brain machine

Why the convergence of psychology and machine learning is the future of CX

5th Dec 2017
Share this content

Every millisecond, billions of bytes of dry, impersonal data generated by mouse clicks and typed commands are being transformed into actions enabling users to check their bank accounts, pay utility bills, find the perfect gift, or book a holiday getaway. The digital footprints left in cyber dust are already being harnessed to make recommendations, provide real-time offers, predict demand for products and more.

By throwing machine learning into the mix, these footprints can be leveraged to improve customer experience significantly, thereby generating hundreds of millions of dollars in business value. Referencing a playbook from the past – from a seemingly unrelated industry – enterprises can reap insights into how it can be done.

My background is in algo-trading - a cutthroat industry that, at its infancy, faced a challenge similar to the one many digital businesses encounter today: the perception that machines would never be able to outperform humans in certain tasks.

Today, for example, companies face the issue of trying to gain behavioural and psychological insights into their digital customers from trillions of data points to optimally serve the customer. Using available data, I developed one of the first algo-trading solutions, now used on stock exchanges around the world. It’s the process of using algorithms to follow a defined set of rules for placing a trade, at a speed and frequency impossible for a human trader.

Algo-trading was groundbreaking in its tracking of price points, financial data and real-time markets to produce actionable recommendations for traders without a human involved. As enterprises increase their data science practices and employ machine learning to answer complex questions about customer experience, I keep returning to the lessons from the algo-trading world.

Determine the evolutionary stage at which your company stands

Trading used to be done intuitively based on a manual workflow to collect data. When algo-trading was introduced, the industry transformation that followed took a number of steps. First, digitally collected data was offered to the trader, creating a feedback loop which enabled better decisions to be made by the trader. Next, the machines took over some basic decisions. Over time, technology helped process more and more of this data. And ultimately, machines were entrusted to make trading decisions.

Today, digital marketing is following a similar journey. At the outset, marketers manually analysed data and, essentially, guessed how to improve customer experience based on educated hunches. Most enterprises today have already entered phase two of the transformation, namely leveraging supervised machine learning. Yet businesses still rely heavily on experience and intuition to map out ways for improving the online journey.

As a recent article in Harvard Business Review put it: “If companies want to get value from their data, they need to focus on accelerating human understanding of data…”

Today, many businesses use generic data sets and data science practices in an attempt to draw out business-specific and even sector-specific trends. In the not-so-distant future, machines will autonomously transform the digital experience, determining and shaping the online experience with little to no human intervention. In order to get there, machines must be taught to distinguish human psychology… a task that is hardly simple.

Generally, when discussing the role of machine learning, we get hung up on instinct/intuition versus Big Data analytics. It’s key to factor in psychology when leveraging machine learning to produce the best and most streamlined customer experience possible. Otherwise, you’re only taking into account the point-of-view of the enterprise and its marketing team, and neglecting to get to the heart of your customers’ intent and mindset. 

Machine learning


Understand the data + understand the customer

Users leave behind an unfathomable amount of data. Our servers, for example can handle several trillions of customer interactions every day. The smartest move you can make for your business is to process and interpret this knowledge in order to reveal more about your customers and their needs. Just as algo-trading data included historical prices, real-time prices and fundamental financial data, these days we have access to data from mouse movements, hovers, scroll reach, pace, customer journeys, mobile gestures and countless other data points.

At its simplest stage, we enrich the data with analytics. In algo-trading, that meant – for example – moving averages and volatility analysis. For today’s digital marketers, it means experience analytics, which includes everything from form analytics to identifying the most (and least) popular paths a visitor takes on your site. When we introduce machine learning, we can abstract insights and produce actionable recommendations…initially consumed by humans and eventually by systems.

By infusing Big Data analytics with psychological insights, we can understand customers better than ever before.

The ultimate stage, which is on the horizon for digital customer experience, is closing the loop with machines that can use performance analysis and measure business outcomes – conversions, engagement and brand awareness or whatever your digital KPIs are – to calibrate models, analytics and personal experience.  

In the digital sphere, we have the ability to understand the psychology and behavior of visitors, use advanced research, and then turn their behavioral signals into insights. Optimisation cycles are getting shorter and shorter and once you take behavioral insights and put them into the feedback loop, ultimately this will be a way for brands to achieve higher levels of personalisation (segments of one).

For example, regardless of a visitor’s previous buying patterns, it is possible to sense whether he or she is goal-oriented, browsing, disoriented, or exhibiting a different mindset, or even if a particular page on your site changes the visitor’s mindset negatively or positively. This behavioural data, which I term a customer’s “Digital Body Language”, combined with unsupervised and supervised machine learning, is critical as we evolve our machines’ ability to come up with actionable insights and, eventually, execute them autonomously.

As many recent articles have pointed out, working in an agile manner, with numerous feedback cycles, is key to deriving business value from data science projects.  

Machines are not replacements

At the time algo-trading was introduced, it seemed utterly radical that we could use machines to calculate trading insights better and faster than human beings. And, naturally, there was a fear that machines would replace humans. Instead of massive layoffs in the financial sector, though, people became stronger domain experts.

It’s important to note that things occasionally went wrong with algo-trading. Most of the mistakes happened when we relied too much on technological processes at the expense of human knowledge and common sense.

By infusing Big Data analytics with psychological insights, we can understand customers better than ever before. And I say “we” with emphasis. Because at the end of the day, even if machines will increasingly become decision-makers, they should not replace humans and the experience, insight and common sense we bring to the table.


Replies (1)

Please login or register to join the discussion.

By aki33
29th Dec 2017 13:31

Most of what is called AI today is a system of statistical correlations. To some extent, humans also reason this way. If A happens at the same time as B, then A likely causes B. Trouble is, that is not always true,

Deep learning system only determine which data values are correlated (associated with) other data values (the outcome or prediction). The systems do not learn even the basics of everyday physics. A deep learning system will learn that a dropped object falls toward Earth, but will not know that the fall is caused by gravity. Most people don't know about gravity or the equations predicting objects in motion either until taught physics in school.

A correlation system cannot tell the difference between association and cause. For example, an AI may learn that zip code can be used to predict whether someone will pay back a loan or not. People, reasoning from first principles, know that the zip code is a stand in for wealth and job security. Statistical algorithms don't know the reason why zip code is a predictor of creditworthiness. Further, as a society we may decide that using zip code as an indicator is off limits, as address is often an indicator of ethnicity. The computer has no such compunctions.

As I found out when having to write my paper on AI, deep learning systems generate useful answers, but not always from connections that make sense to people. When the actual cause is represented by a chain of correlates (stand ins), it may be difficult to understand why one variable predicts another.

Thanks (0)