walking away
From data to predictive analytics, machine learning to AI — the only way brands will successfully deliver personalized experiences is by tapping all of these capabilities PHOTO: Joshua K. Jackson
"The future is already here — it's just not very evenly distributed" — William Gibson

The only way we will ever succeed at delivering personalized experiences across the broad landscape of customer experience interaction points will be thanks to the rise of the machines. 

Thanks to advertising and marketing technologies, brands can now start to conduct programmatic decision-making across the their paid, earned and owned channels, serving up what a computer algorithm believes is the right content for the person at hand at that time. 

But when it comes to understanding how computers make real-time experience delivery decisions, the hard distinctions of computer science get convoluted in a swirl of buzzword bingo. Let’s try to cut through the buzz to get clear on the layers of data-driven and programmatic decision drivers for experience delivery.

The Steps to Experience Delivery, From Data to AI

1. Data

The delivery of highly effective (and thus likely somewhat personalized) customer experiences begins with our knowledge of our customers and their concerns, and ends with how well we apply that knowledge to creating mutually beneficial experiences. 

This knowledge is built on data. Some data about customers and their concerns arrives through research methods like surveys and focus groups. Other data arrives through direct observation and measurement of behavior and transactions. Some data arrives in batch (research results, nightly point of sale transaction files), other in real-time (web interactions and conversions, social media streams, calls to customer care).  

We can code and quantify all of this data, even verbatim responses or recorded written and verbal exchanges. 

2. Analysis & Predictive Models

Once we have quantified (within nominal, ordinal and interval scales as appropriate for the variable) all the data we want to use, we can analyze it and apply that analysis to better achieve business goals. 

The fundamental analytic practice is performance evaluation. Every business has Key Performance Indicators (KPIs), and at a minimum, is applying data analysis to measure performance against those KPIs. Ideally, the performance is also being explained in some sense — an effort to define what might cause positive or negative performance against a KPI. With these causal analyses, we can also develop predictive models.

3. Optimization

One we understand what causes outcomes and can predict the outcomes we will get based on what we see in the causal variables, we can begin to optimize our situation around those causal variables. 

Some of that optimization will involve human thought process — particularly the design of content: the selection of words and images that will resonate with our segments. In other words, the “art” of marketing. Some will result in more effective timing and targeting (geographic and segment based) for large, longer-duration campaigns (accompanied by the optimized content). 

But at this stage, companies should be turning over large parts of optimization to machines (computer programs) for more immediate-term and real-time decisions from budget and forecast modeling and optimization to real-time experience delivery. 

4. Algorithms

Algorithms come in basic and advanced categories. The most basic of machine-based or “programmatic” marketing decision tools apply algorithms constructed of pre-defined step-by-step instructions to follow in sequence. 

In human terms, basic algorithms are like a child who follows a caretaker’s instructions (“please pick up the toy”) without understanding why the required behavior is necessary, or what the consequence might be of divergence from those instructions. 

5. Machine Learning

Learning however means the child begins to choose their responses to stimuli (toys on the floor, voice commands) based on an understanding of consequences. They have learned that picking up the toy proactively earns an expression of satisfaction from the caretaker, and perhaps some reward. They have also learned that if nobody is asking them to clean up, then leaving the toy on the floor (and probably creating an even bigger mess) offers a different sort of satisfaction and reward (more play-time). 

The choice of which action — which decision — the child makes is based on some learned expectation of the consequence of either action (pick up or make a bigger mess) in the context of the moment (explicit request or not) and a belief in which outcome is more desired. As they observe and learn more about schedules through the day, tones of voice, and how long “leaving in five minutes” actually takes (e.g. “when will I truly be expected to clean up after being asked?”) they also learn how to make decisions about actions (and predictions about outcomes) in those learned contexts.

Machine learning algorithms offer this ability in programmatic processing: the ability to modify and evolve the criteria considered and options available in the decision making process. Decisions are no longer a pre-defined chain established about what was known of possible inputs at the time of initial programming, but are essentially reprogrammed to meet their goal as new observations of cause and effect around that goal are made. 

6. Artificial Intelligence

If learning (by human or machine) means an increasing ability to bring greater context to decisions around known target outcomes (i.e. learning facts to pass a test), then we can think of Artificial Intelligence, like human intelligence, as the application of prior learning to identify possible outcomes beyond those directly defined through learned cause and effect. It is the ability to develop new possible chains of cause and effect that are implied or predicted by what has been learned.

Artificial Intelligence does not require (or mean) “machine consciousness” any more than human intelligence requires consciousness. It seems safe to say that in people, most intelligent thoughts and actions are not brought into being through conscious effort (e.g. one thinks about being intelligent, then is) but rather emerges from the sub-conscious (one is intelligent before/as they arrive at a conscious thought). 

What intelligence does require is an elevation of learning to an ability to create optimal outcomes within previously unrecognized patterns of cause and effect. In this way, machine learning is a foundation of artificial intelligence, and artificial intelligence is an advancement in applied machine learning. 

Set the Foundation First

My next post will go further into the distinction between machine learning and artificial intelligence. Hopefully this post provided some clarity about all of the components behind data-driven personalization to help in priority setting and decision making for anyone working on building such a personalization capability. 

Think through your capability in these steps: Is your data house in order? Do you understand performance and its causes? Do you have a culture of optimization in place? Do you know what you can turn over to algorithms?

Only when all of these are solidly in place will machine learning and artificial intelligence for experience delivery truly begin to pay off.