In a recent article in MIT Technology Review, “How AI is changing the customer experience” a survey of more than 1,000 business leaders reveals that sales and marketing is expected to lead the increase in AI deployments over the next two years, from 33 percent using it actively today to an expected 59 percent by 2022. While the survey did not directly address AI adoption in response to the coronavirus crisis, a McKinsey survey did. In its B2B Decision Maker Pulse survey, 25 percent of companies surveyed said that the pandemic is leading them to redirect and increase spend toward emerging opportunities.
The MIT article posits that personalization will largely be the driver for sales and marketing to take the leap, in a transition from more efficiency-based use cases, because AI has been shown to “transform the way companies interact with their customers … bring a deeper level of customer understanding, drive customization and create personalized journeys.”
One result of using AI to interact with customers on a more personalized basis is that it devalues the use of personas. With a deeper understanding of customers at an individual level, the traditional persona attributes – income, age, household size, location – become less effective in predicting how a customer might react to an offer or message. AI takes away the need to hazard a guess. “The idea that you can create personas, and then use them to target or serve someone, is over in my opinion,” says one CEO in the MIT article – an opinion I share.
Old School vs. New School
Marketers who rely on personas to segment customers may then rightly ask what will replace them? Will there simply be more granular groupings? Here, AI closes the loop by automating the end-to-end process; if a personalized message or offer is the result of using AI to become closer to the customer, another AI use case is to dynamically select an audience. There are numerous techniques that can be applied here such as tradition “K-means, medians, mediods”; Collaborative Filtering; or more technical approaches like Real-Time Attention Based Lookalike Model (RALM). The key in selecting an algorithm is to be able to deploy analytically based clusters as fast and easily as you would deploy your personas.
In fact, audience generation using real-time algorithms that select an audience on the fly is the second most frequent use case of Redpoint Automated Machine Learning (AML). The first is the real-time decision over what to present – the next-best action or offer. This is a completely new way for marketers to think of clustering models vs. personas, with groups determined based on certain characteristics.
Intelligent grouping of discrete data, where the data tells you what the clusters need to be, means that an audience is created only at the point in time where a next-best action will be rendered. This is because customer data – preferences, behaviors, transactions – loads in real time and at an individual level, not a group level. Multi-dimensional clustering with an algorithm groups on dimensions that are otherwise impossible to select manually – you could never figure out which or how many of 20, 50, 100 characteristics warranted inclusion in one persona over another.
Rather than setting and adjusting discrete parameters of a persona, model clustering with AML provides a far more nuanced way to bring customers together, tap into similarities and then test on the fly. It’s the difference between choosing a known algorithm vs. asking which algorithm to use based on the real-time data presented to the model; the task of having to define the model, in other words, is done by the algorithm – not the marketer.
Making Data Science Accessible
There’s a common misconception that an army of data scientists is required to build models, but one of the unique aspects to AML is that in addition to off-loading the task of defining models, the solution provides marketers with access to the modeling done in the AML tool.
Traditionally, analytics has been essentially trapped inside of a handful of data scientists who build segments and models off-line. This is a big hurdle for widespread AI adoption for sales and marketing use cases. AML removes this barrier for success because it’s built for marketers. The everyday marketer can go out and build a model without having to write a line of code, and the algorithms will dynamically select the optimal audience at the time of selection, on demand.
Since the advent of marketing, ambitious marketers have been innovating with ideas to group “birds of a feather” to more effectively engage with customers. With AML, dynamic audience selection using multi-dimensional modeling puts the feathers under a microscope. A customer is only included in an audience if the customer’s up-to-date behaviors and preferences at the very moment of the audience selection warrant inclusion.
If there’s to be a significant rise in AI sales and marketing use cases over the next few years as the MIT article predicts, I can think of few better ones than phasing out old school personas with next-level audience selection as the basis for delivering a superior personalized customer experience to a segment of one.