Magnus Unemyr

With audience AI and hyper-dynamic targeting, machine learning can be used to help build targeted audiences automatically. This is done by using data from a variety of sources and making predictions about the characteristics of ideal buyers.

In a few seconds, you can get well-targeted, quality audiences, with a high likelihood of more conversions and sales. Audience AI learns how to find the best customers across various channels, such as social media. As more data becomes available, the targeting becomes increasingly precise.

These tools can uncover the motivations and desires of target audiences, sometimes finding passion points with interest analysis, core values, and personality traits with personality analysis, and using visual imagery analysis to work out what pictures or creative elements engage the target audience the most.

Some products performing these types of tricks are the well-known behemothslike Salesforce and Adobe, but also specialized companies like Codec, Leadspace, Toneden, and Trapica.

Of course, we can’t forget the ads platform on Facebook and other social media sites. AI can be useful in audience management, particularly for companies with large customer volumes, as often is the case in the B2C space.

While content marketers are interested in adaptive and predictive content, e-commerce marketers might be more interested in the product recommendations that optimize sales and maximize revenue.

Amazon is the master of this, and their team uses machine learning to personalize product recommendations based on digital footprint and previous purchase history—both from you and others.

Features like “You may also like this” and “Other people also bought” that were popularized by Amazon are becoming the norm in e-commerce these days.

The “Personalization in Shopping” report also notes,“Recommendations are directly linked to longer shopping visits. Shoppers that clicked a product recommendation spent an average of 12.9 minutes on-site vs. 2.9 minutes for those that didn’t click recommendations.”

Of course, when they stay longer on the site, they spend more too.

The report states that “visits where the shopper clicked a recommendation comprise just 7% of all visits, but 24% of orders and 26% of revenue.” Shoppers that click recommendations view 4.8 times more products and spend five times more per visit.

Furthermore, 52% of orders from recommendation-clickers include one of the recommended items. These figures seem to be a good argument for the use of recommendation engines in e-commerce.

Luckily, nearly all e-commerce site, even small ones, can now use AI-driven product recommendations that are optimized and adapted for each individual person. The suggestions can come in many forms, but generally, they adapt the products each specific person is offered on websites or in email.

There are many popular e-commerce platforms in widespread use including Shopify, OpenCart, WooCommerce, PrestaShop, and Magento. Wouldn’t it be nice to get a product recommendation engine for the system that drives your business?  

You can use companies like BlueShift, Dynamic Yield, Perzonalization, Reflektion, RetentionScience, and Zen to help increase sales by personalized product recommendations, including the display of similar products, trending (most popular) products, top picks for you, recently viewed, often bought together, and the like.

Up-selling and cross-selling is also a common feature here. 

Do you want to extend recommendations beyond products and e-commerce?

The same type of predictive technology can be used to recommend what content (such as CTAs, banner ads, lead magnets, and other downloadable resources) to display on your website for a particular visitor. Certona, Marketo, and Uberflip do this effectively, helping to promote your content library in a smarter way.

Recommendation engines have many more uses. For example, they can be integrated into the products themselves. Netflix is a great example, and their recommendations for movies you might want to watch are a contributing factor for their huge success.

You might be able to replicate some of that magic by using recommendation engines appropriately in your business.

Segmentation taken to the extreme eventually creates segments of one. This personalization allows for dramatically more effective campaigns. It allows you to send content and offers that are uniquely optimized for every individual person.

This is called the segment of one (or audience of one), denoting a hyper-personalized outreach where the message is uniquely modified and optimized for each person. This would certainly make marketing more relevant!

In fact, personalization is the opposite of segmentation. With segmentation, we create groups of people with similar attributes and send the same messages to all of them. By definition, this is not personalized for each individual, even with micro-segmentation.

With personalization, we want each message to be truly unique. It is actually adjusted and optimized for each individual using prediction algorithms. There are already many vendors offering AI-based personalization systems.

The majority of these tools can personalize the content on web pages or email content. They do this by using machine learning algorithms to predict what customers want and deliver relevant content or product recommendations even before they ask.

The “Personalization in Shopping” report notes that “Shopper spend soars with personalization. Purchases where a recommendation was clicked saw a 10% higher average order value, and the per-visit spend of a shopper who clicks a recommendation is five times higher.”

The report further states, “Providing your customers with a personalized shopping experience is now the cost of entry to retail.” It reports that 50% of customers say they are likely to switch vendor if a company doesn’t foresee their needs, and 58% of consumers say technology has considerably altered their expectations of what customer experience companies shall give them.

Personalization is now a requirement for any company wanting to stay with the times. I think we can conclude that it isn’t optional anymore, but rather a pre-requisite to matching your competitors.

However, it isn’t just about recommending content, products or offers. It is about creating customer experiences that build engagement and drive retention. It’s about appearing less robotic and being more personal and relevant.

Personalization will be huge in the marketing industry. In fact, the future of marketing technology and the vendor landscape might well be an arms race for better personalization.

Many companies have developed AI-based technology that uses user behavior and content analysis to deliver the content that is most likely to resonate with a particular person. Most of them offer solutions for the personalization of web page and email content.

Some go beyond that to support additional communication channels. A few prominent companies in this space include Zeta (previously BoomTrain), Adobe Marketing Cloud, Emarsys, and Perzonalization.

I asked Lindsay Tjepkema, Global Head of Content for Emarsys, to explain their product. She noted that “Emarsys is a marketing cloud which provides AI based personalization across email, SMS, mobile, web, offline and IoT devices.”

The platform consolidates all web, mobile, email, and purchase information into a unified customer profile. This profile includes preferences, behavior trends, predicted behaviors, propensities, and affinities. The unified profile is the platform’s foundation—the “single source of truth”—which enables hyper-personalization across all channels.

Tjepkema continues: “As the real-time interactions scale, the need for content increases exponentially. Marketers will have to offload the technology tasks such as identifying segments, crafting journeys, and creating campaigns. They will have to concentrate purely on creating content and training machines on marketing strategies.”

Other companies avoid developing their own proprietary AI by building their products on the back of existing engines. OpenTopic takes this route, as their tool is built on top of IBM’s AI engine, Watson. The computing power of Watson is used to predict the most engaging assets that guide individuals through the customer journey.

In addition to the more traditional content recommendation functionalities, Dynamic Yield, Klevu, PureClarity, and Similar add personalization to search results as well. Dependent on what you have done before, you will get different search results on an e-commerce site, compared to your neighbor searching for the same thing.

I spoke with Mike Mallazzo, the head of content at Dynamic Yield. He points out that in its annual “What’s Hot in Digital Commerce” report, Gartner cited personalization as the number one strategic investment for brands in 2017.

McKinsey, he said, refers to digital personalization at scale as “marketing’s holy grail,” and the Boston Consulting Group predicts that personalization will push an $800 billion revenue shift to the 15% of brands that get it right in the next five years.

The message from the marketplace is clear: personalization is no longer just nice to have; it is the single most important strategy for boosting revenue and brand affinity online. Expect predictive content and hyper-personalization to be a major deal going forward.

Malazzo also discussed what helps set his company’s product apart: “Marketers can turn control over to the machines or insert guardrails in the form of merchandising rules to control which recommendations unit each visitor sees. For example, a high-end fashion brand may want our AI to recommend the products delivering the most revenue but may not want to place certain brands next to each other. In Dynamic Yield, it is possible to set this condition, allowing the machines to go to work within controls set by the marketer.”

With machine learning, it becomes feasible to personalize content and outreach strategies to each individual contact, which helps make marketing more relevant. In fact, personalization is one of the most important uses for artificial intelligence in marketing.

The time of spammy “spray and pray” mass-marketing is over. Today’s customers require brands—and expect marketing outreach—to be more relevant in both time and content.

The time of spammy “spray and pray” mass-marketing is over.

Relevancy is becoming more important than reach; get the right content, to the right person, at the right time.

Marketers tried to become more relevant by inventing the concept of segmentation. Groups of customers or potential customers are placed into distinct sub-groups, thus making a rough classification by similarity or common needs.

Segmentation made it possible for marketers to be at least a bit more relevant, compared to sending the same marketing emails or offers to everyone.

I know a clothing retail chain that did this rather poorly, sending marketing emails pitching women’s clothes to men and vice versa. They sent exactly the same email to everyone. Another retail chain sent direct marketing information on gardening products to people living in an apartment, thus not having a garden.

These are significant blunders in today’s era, and brands just have to do better.

When you send irrelevant marketing content to your leads, you train them to avoid reading it in the future, or even make them dislike you. That in turn can harm your email sender score, thus damaging the delivery rate of future emails.

Even the most trivial segmentation—like sending different product offers to men or women—can make a major difference. Still, men are not a homogenous group, and they have different needs and interests, as do women.

A more granular segmentation would thus result in even more relevance.

With AI and machine learning, it is possible to segment the audience automatically. If this is done with sufficient granularity, we get micro-segmentation. Done correctly, this is much better than no segmentation, or rudimentary segmentation.

With AI and machine learning, it is possible to segment the audience automatically

AI-based automatic segmentation can be implemented using clustering algorithms, and can identify the attributes of ideal buyers, those most likely to convert or defect, and more.

In effect, you can ask the system to provide you with buckets of people that are similar or share characteristics in some way, even if you don’t know what you are looking for. Thus, machine learning tools can help improve segmentation, and many do.

Having too many leads may appear to be a good thing. In fact, most CMOs would love to have this problem. However, being flooded by leads unlikely to buy is a problem in itself, as they bog down your sales time, thus damaging sales activities towards potential customers more likely to buy.

There is a significant difference in being flooded by poor leads versus having quality leads who are just a bit too early in the funnel.

Being flooded by leads unlikely to buy is a problem in itself

There are two solutions to the first situation: either you get rid of the low-quality leads altogether, or you filter them out. In the second situation, you have quality leads worth nurturing, but they aren’t yet ready to purchase.

With a stretched sales organization, both situations lead to the same conclusion: your valuable sales reps ought to focus on the right leads.

In practice, this means prioritizing the leads most likely to purchase soon. The other leads might be nurtured using traditional marketing automation systems until they become warmer.

How do you do this filtering and prioritization?

Traditionally, marketing automation systems allowed manually configured lead scoring algorithms to rate the leads versus each other. This is done by attaching credits to various activities. Recurring website visits and certain page views increased the lead score.

For example, it might be +1 for each page view, but visiting the pricing page gives +10, and a visit to the job opportunities page gives -15 (since the user’s interest isn’t in the product). Every PDF download may give +10, and watching a video on the website gives +5, and so on.

You get the point.

Various activities add or subtract to the lead score, and the total lead score gives a rating of how “hot” a particular lead is estimated to be. This is great in theory, but is subject to certain problems.

For example, the manually designed scoring formula might give the wrong perception of how likely a lead is to purchase. You may assume that certain types of page views are more or less important than they really are.

Manually designed scoring formula might give the wrong perception of how likely a lead is to purchase

You may think that downloading PDF documents is a stronger purchase signal than something else. Sometimes, these assumptions are wrong. 

This is where machine learning and predictive lead scoring comes to the rescue.

With predictive lead scoring, a software algorithm compares the past behavior of those who later became customers with a particular new lead who is not yet a customer. In effect, the digital behaviors of a lead are compared to the behaviors of actual customers before they became customers.

The digital behaviors of a lead are compared to the behaviors of actual customers before they became customers.

Depending on the similarity in behavior, a lead score (usually in the range 0-100%) is calculated to assess how likely a new lead is to become a paying customer.

For leads passing the threshold of a lead score about 25%, a marketing automation system may send a sequence of nurturing emails with a soft sales message. Leads having a lead score over 50% may get emails with a harder sales message. Those with a lead score about 75% may be put in the queue for manual follow-up by a sales rep.

Several software tools perform predictive lead scoring like this. Many marketing automation systems and CRM systems, including HubSpot, Marketo, and SalesForce can do this too.

Infer (now acquired by Ignite Technologies) is one of the vendors who offer dedicated predictive lead scoring solutions, thus helping the sales team focus on opportunities with the highest probability of closing.

They use a combination of data gathering and machine learning to classify and rank leads and accounts on the likelihood that they will convert. I asked Rob Franks, Vice President of Professional Services at Ignite Technologies, how this happens.

He says, “Infer works by gathering data, called signals, about the lead from over 4000 different sources. The sources include syndicated data providers, proprietary data gathered from web crawling and the inputs and behaviors of the customer. Initial models are built by taking a history of data, classifying the signals and using win loss history to train a predictive model.”

In essence, the prediction algorithm learns what lead behaviors and attributes have led to a won or lost deal in the past, and uses that to assess new leads. Franks explains, “The models are reviewed and optimized in Infer’s proprietary workbench tool by experienced data analysts and scientists. When the model is ready, it is deployed to score all active leads and new leads as they are created. Scores can change as signals evolve so the models are periodically re-built with a more recent history of wins.”

Predictive leads scoring works well in combination with marketing automation workflows

I’ve used predictive lead scoring myself for a few years, and it is indeed a handy solution to filter out what are predicted to be low-quality leads. It is also great to use different lead scoring thresholds in marketing automation workflow trigger conditions.

As mentioned above, a marketing automation system may send different types of messages to clients with different leads scores.

Many other vendors offer predictive lead scoring systems, including Lattice, Maroon, and Mintigo, and several CRM system suppliers we discussed above. You will have to assess what system works best for your organization.

In my experience, predictive leads scoring works well in combination with marketing automation workflows. You can use lead scoring alone to prioritize leads for manual outreach or as a handover threshold from marketing to sales, but automated email sequences or other workflow logic makes this capability shine.

Good marketers always listen to their customers. This can be done in many shapes and forms. The traditional way is to talk to them in person or over phone and collect direct or indirect feedback.

It can also be done using surveys and questionnaires. In the digital age, there are many more ways to listen to the voice of the customer, and with artificial intelligence, we can do things we never thought possible a decade ago.

Why is it important to listen to the customer?

Because by the time they leave, it is too late to rectify the situation. Customer churn is the final cry. Unhappy customers often show signs that they are unhappy long before they leave you. How can you detect this?

There are many ways, including measuring:

• The number of website and blog returns and visitor behavior
• The number of web shop visits and shopping behavior
• Email opens and clicks
• Open rates for push notifications
• Social media engagement
• Customer service interactions
• Recommendations and reviews
• Sentiment analysis of interactions
• Conversion rates
• The emotional state of your customers

While many such measurements are not based on AI per se, the accuracy of your reporting can be improved by using predictive marketing and machine learning, thus providing new ways to hear the customer’s voice.

You can then measure the difference in behavior for highly personalized content versus static mass-audience content. The personalized outreach is likely more efficient and can lead to deeper relationships with your customers.

With modern tools, you can get insights never before imaginable.

With modern tools, you can get insights never before imaginable.

Sentiment analysis is one of those areas. This is about detecting emotions in text to better understand how customers feel about your products, and their opinions and needs.

With sentiment analysis (also known as emotional AI, or opinion mining), you can measure if someone is positive, negative, or neutral towards your brand or product, and what emotional state they may be in.

In effect, you get to know how they feel when they write about your brand or products. This information is gathered by analyzing and understanding text in email, social media posts, product reviews, customer service tickets, chatbot conversations, and more using natural language processing and natural language understanding.

In short, you can see the polarity of their opinions, and how strong those opinions are.

You can see the polarity of their opinions, and how strong those opinions are.

A company in the text analytics space is Lexalytics. They have offered commercial sentiment analytics software since 2004 and their system processes billions of documents per day. Their solution analyzes and provides insights around a company’s text data from surveys, call logs, social media posts, message boards, and comments.

In addition to the more common cloud-based or on-premises solutions for text and sentiment analysis, they also offer an add-in for Microsoft Excel to easily analyze both structured and unstructured survey data, generate insights, and create reports and data visualizations.  

I spoke with Seth Redmore, CMO of Lexalytics. He argues there’s simply too much human communication out there to understand, analyze, and act on without AI assistance.

To address this, he says, “The Lexalytics Intelligence Platform helps businesses work better with text. This could be as part of a decision support system, or an analytics offering, or predictions. It could be looking backwards, or peering forwards.”

Redmore explained that “Text is language. To understand text is to understand meaning, in a way, it is to understand the nature of being human, by being able to untangle how we communicate.  We focus on conversational text—as examples, social media or customer feedback.”

We will see businesses harvesting more insights from text and sentiment analysis in the years to come. There are many vendors in the field, all with different strengths and weaknesses.

In general, they all use natural language processing algorithms to enable their magic. Depending what solution you chose, your experience may differ in terms of languages supported, or how advanced the language understanding is.

Some products might only support English, which may be a problem to companies in non-English speaking countries.

Talkwalker, a social media analytics company, launched an AI-based sentiment analysis tool in the autumn of 2017 that claims to detect customer sentiment with 90% accuracy, even understanding irony and sarcasm.

With machine learning, the tool understands the meaning of full sentences and determines the attitude in social media posts or other content. Their platform monitors and analyzes online conversations in real time across social networks, news websites, blogs, and forums—all in 187 languages.

The analytics engine analyzes the data across a wide variety of categories including social engagement level, sentiment, demographics, location, linked themes, and more.

It also features image analytics technology and sentiment analysis.Christophe Folshette, co-founder of Talkwalker, explains that, “All of Talkwalker’s AI technology is proprietary and is currently focused on image analytics capabilities, sentiment analysis and data classification. Talkwalker’s image analysis is able to detect not only brand logos, but also scenes and objects within images to give clients greater clarity on the context these images are used in.”

Folshette goes on to say, “We’ve rapidly increased our AI-based image applications, adding scenery, objects, gender and age detection. We’re seeing clients use this technology for a wide range of purposes from understanding the impact of sponsorship to analyzing Instagram data to create maps of customer tastes and preferences and direct business expansion.”

Companies with the right AI-powered tools certainly seem to have better insights to base their decisions upon, but that’s not the end of the possibilities. Sentiment analysis can be applied to audio as well.

Sentiment analysis can be applied to audio as well.

One example is the phone service company Nexmo, which provides real-time customer sentiment analysis of voice calls. A voice-based AI-bot monitors the conversation and estimates the emotions of calling customers.

The service agent can watch the estimated customer sentiment live as the call progresses, then react to change the course of the call as needed.

 Another machine learning solution that goes beyond words is Affectiva. They can analyze the emotional state of facial expressions in images, while Beyond Verbal does something similar thing voice recordings. These types of tools can help analyze how people feel based on images or audio recordings.

Given the vast amounts of footprints leads and customers leave in text, audio, and images, there is no shortage of data to harvest for emotional intelligence. As these systems develop and become ubiquitous, we as a society will need to think and act clearly on the legal and ethical aspects of these systems.