AI at the core. Humans at the helm.

Any marketer will tell you that applying AI to Marketing is a hot trend right now and has the potential to disrupt the industry. Just last week, Oracle announced that it is delivering artificial intelligence across its customer experience cloud. Oracle’s announcement follows a long line of press releases from major marketing clouds such as Salesforce, IBM and Adobe. The space is so hot, in fact, that there seems to be a competition to have the best name for your AI, with the likes of Einstein, Watson and Sensei regularly competing for top billing.

If you look beyond the cute names, however, you will find that there is real technology at play. AI Marketing solutions aren’t just a new way of describing the collaborative filtering recommendation engines that gained popularity last decade (e.g., Certona, MyBuys and Baynote). Nor are they the A/B/N or multi-variate testing tools that grew up earlier in this decade (e.g., Optimizely, Maxymizer and Monetate). They are, assuredly, all of that and much more, taking advantage of advances in AI technologies such as image/facial recognition, natural language processing/generation and machine learning to fundamentally change the way modern marketing is done.

Today’s marketing clouds fall short

The challenge with traditional marketing automation solutions is that they are only as scalable or as intelligent as the marketers running them. While such solutions have shown the ability to automate marketing execution for specific “if/then” scenarios, these solutions fall woefully short when trying to apply them broadly across a large B2C enterprise for three key reasons:

1. In order to get the desired targeting granularity, marketers have to write and maintain dozens of “if/then” rules across hundreds or even thousands of campaigns

2. All of the targeting rules have to be set in advance of ever having run the campaign, so initial success relies solely on the experience and “best guess” capabilities of the marketer configuring the campaign

3. Running A/B/N tests to optimize campaign efficacy remains a very manual, labor-intensive process, often requiring a data scientist to get involved and spend weeks doing uplift or propensity modeling, only to come up with a recommendation for improvement that, while helpful, only positively impacts a small portion of a marketer’s total audience.

As a result of these challenges, marketers spend their time programming campaign rules, managing holdout groups and analyzing test results instead of being strategic or creative. While it is important for today’s marketers to be “data-driven,” the pendulum has swung too far: automated and programmatic campaigns have become so focused on improving short-term opens and clicks that they miss out on optimizing the longer-term ARPU (average revenue per user) and retention KPIs (key performance indicators) that directly impact topline revenue.

AI marketing’s early days — for engineers and mathematicians only

For several years, disruptive B2C brands have recognized the role that AI can play in engaging customers and driving revenue. Amazon has developed arguably the world’s most famous recommendation engine, relying on item-to-item collaborative filtering to drive product suggestions for well over a decade now. Entertainment and media companies like Netflix and Spotify have followed suit, relying heavily upon personalization to keep customers engaged on an ongoing basis. And retail-in-a-box providers such as Stitch Fix or Trunk Club couple techniques such as natural language processing and clothing recommendations with input from a human stylist to decide which items to send to which consumer.

Until recently, however, using artificial intelligence to drive marketing and customer experience required building a team of hundreds of engineers and data scientists to spend years building and perfecting an optimized model for the particular use case at hand. With the rise of major cloud providers such as Amazon and Microsoft, as well advancements in big data tools and infrastructure, machine learning capabilities are more readily available than ever before, with off-the-shelf offerings like TensorFlow, Azure ML and Spark MLlib.

From developers to marketers with bolt-on AI

Using off-the-shelf ML tools has eliminated the need to fully code an AI from scratch; however, they still require significant customization, and one must be either a hard-core engineer or an advanced mathematician (or both) to fully take advantage of these tools. But this is changing as more and more players look to provide marketers with tools that fit seamlessly into their current marketing solutions.

This evolution to provide marketer-friendly AI tools can most easily be seen in the press releases of the major marketing clouds. In the month of April alone, Salesforce had 70 distinct press mentions talking about how Einstein is helping to power their Marketing Cloud. Point solutions are also getting into the game, as witnessed by mobile marketing company Kahuna’s recent repositioning away from being a mobile marketing company and doing an overnight transformation into an AI-powered cross-channel marketing platform.

Salesforce Einstein applies AI models to solve specific, narrow use cases across the various clouds. Image courtesy of Salesforce.com.

In the vast majority of cases, existing marketing technology providers (both major marketing clouds and point solutions) are following a similar path. They are acquiring large teams of data scientists who build and deploy models to target narrow use cases in order to improve marketing engagement and effectiveness. To see this in action, one only has to look at Salesforce and what they are doing with Einstein. Salesforce has acquired at least 10 AI companies, compiling a team of more than 175 data scientists. Those data scientists have, in turn, bolted AI models onto their existing legacy marketing cloud infrastructure to help address some of the key pain points across the marketing spectrum, such as optimizing the right time to send a consumer an email or to predict a customer’s likelihood to open or click a particular email.

Bolt-on AI solutions can add incremental value over current business-as-usual marketing, as Salesforce notes on their company blog:

“Marketing Cloud Einstein has been in beta for almost a year and we have seen some tremendous results. One of my favorite examples is the ecommerce and coupons company ShopAtHome. By redefining customer engagement around predictive scores, the company generated a 23 percent lift in email clicks, and a 30 percent increase in email opens.”

— Courtesy of Salesforce Blog Post “Welcome to the World of Intelligent Marketing”

In today’s world of the connected customer, however, simply optimizing opens and clicks isn’t enough. Today’s connected customers expect a longer-term, value added relationship with brands, so enterprises must optimize for longer-term KPIs such as 45-day average revenue per user or 60-day retention.

For good reasons, bolt-on AI solutions that are built on legacy infrastructures can’t support this type of approach. The average B2C enterprise has on average 100+ attributes for each of their customers across persona data, (name, address and so on), behavioral/usage data (e.g., game play data, voice consumption data) and marketing interaction data. By combining these attributes, there are more than 2¹⁰⁰ different permutations of assembling that data for testing and targeting purposes. That may look like a small number when written in an article like this, but when you consider the fact that the length of the universe in seconds is 2⁴⁴, it becomes apparent that bolting AI onto existing legacy infrastructure to solve narrow use cases is not going to be sufficient in today’s world of the connected customer.

Applying AI at the core of the marketing stack

In order to optimize longer-term KPI optimization for customer lifetime value metrics such as revenue and retention, one needs to re-imagine a modern day marketing cloud with AI at the very core of the tech stack.

Applying AI at the core of a modern-day marketing cloud has the potential to disrupt in a world of aging marketing cloud architectures. For example, Amplero’s own Artificial Intelligence Marketing (AIM) Platform incorporates machine learning and multi-armed bandit experimentation at the core, creating marketing automation and optimization tools that quickly run thousands of recursive tests, to continuously optimize every customer interaction and adapt to rapidly evolving consumer behaviors across nearly any marketing channel, all at a scale that is not humanly possible.

AI applied at the core is capable of driving autonomous marketing and continuous optimization:

• Auto-machine learning marketing quickly integrates and activates disparate data environments and points marketing solutions into revenue-producing campaigns.
• Continuous testing through multi-arm bandit experimentation mathematically optimizes KPI lift by exploiting pockets of KPI lift (value) while continually exploring new possibilities.
• Adaptive optimization ensures timely response to changes in customer or market dynamics without the need for human intervention.

In addition to driving improved campaign execution and optimization, marketing clouds that apply AI at the core generate a tremendous amount of insights about campaigns, customers and context which can be proactively pushed back to the marketer via alerts to help guide future marketing strategy and creative. These learnings are automatically applied back into the AIM platform to continuously optimize KPI performance.

While Amplero is an early leader in applying AI at the core of the marketing stack, other companies taking similar approaches include Optimove, Adgorithms and Motiva. Collectively, companies applying AI at the core are using machine learning to plan, personalize and optimize every interaction across the customer journey.

We are clearly entering a new era of marketing clouds — the era of Artificial Intelligence Mareting. With humans at the helm and AI at the core, the possibilities truly are endless.