There has been no shortage of promises recently about how deeply the application of machine-driven artificial intelligence is expected to change our society in the next decade. Microsoft announced its ambition to conquer cancer using natural language processing to analyze research papers in close to real time. Google—now a self-proclaimed AI-first company—will make computers sound just like humans by applying machine learning to a vast corpus of human voice samples. Facebook is using AI to analyze satellite footage to locate all human life and fulfill its promise to bring the internet to the world’s entire population. And the list goes on.
While all these ambitions deserve serious attention, they portray AI as an abstract toolkit for the world’s largest tech companies to solve generational challenges, rather than a foundation for market-ready technology that can instantly improve an organization’s marketing right now.
Many CMOs express a desire to leverage AI-based technology and solutions, but often lack a deeper understanding of the foundations of AI, as well as any starting points to introduce AI to their operations. A recent study of CMOs from the U.S., U.K., and China from organizations with $500 million+ in revenue found that about two-thirds believe AI will play an essential in the future of their marketing operations. But only a third say they have a solid understanding of how to apply AI to their operations.
This article aims to provide a starter guide to market-ready AI technologies in the digital marketing space that CMOs can and must consider today to remain competitive in the years to come.
Long gone are the days when blasting universal marketing messages, ignorant of consumers’ demographics or preferences, was an acceptable communications strategy. Today’s marketing requires engaging prospective and existing customers through highly personalized and relevant conversations. These should be based on myriad factors, such as observed individual behaviors and preferences, contextual awareness like the time of day or location, as well as extracted insights from data gathered from the customer base. Companies like Spotify, Amazon, and Netflix lead the way and set expectations in this regard. Observed user behavior and choices are leveraged instantly to optimize the experience and offering, which yields a more personal brand relationship and improved loyalty.
However, personalized, tailored marketing at scale quickly reaches a breaking point when relying on a manual workforce. Activities like identifying existing customer segments that require separate messaging, running social advertising targeted to the right demographics, or curating product recommendations based on a user’s other selections are time consuming for humans and often beyond the scale of what a marketing department can accomplish.
Enter artificial intelligence. To perform increasingly personalized marketing at scale, marketing workforces will need to rely on the input and efficiency machine-driven AI applications can provide across all cycles of the customer journey. This does not necessarily imply a hands-off approach, where AI-based applications operate on their own without human involvement. Indeed, the performance of AI-based algorithms improves when they are reviewed, sanitized, and operationalized by real, live humans.
AI’s potential will soon allow existing and new market competitors to disrupt the landscape with a force similar to that of the mobile revolution; putting off the adoption of AI will set marketing back. Non-AI companies will inevitably engage with customers in a less personalized and impactful way, miss opportunities to increase the efficiency of day-to-day marketing tasks, spend advertising budgets in suboptimal ways, and miss out on data-driven insights that can directly impact sales.
A Marketplace Overview.
Based on the promise of AI, it is no surprise that new AI-based marketing platforms are surfacing almost daily, which leads to a murky and complicated market landscape. Broadly, today’s relevant applications can be classified as vision, language, insights extraction, and anticipation and predictions.
Vision-driven marketing AI applications.
Many emerging technological advancements, like self-driving cars and earlier cancer detection, rely on vision-based AI. The mechanics often involve providing an algorithm with a training data set with predetermined output, then recognizing patterns in the data that allow a high probability of identifying the accurate output of a new, unseen data corpus. Following are three AI-based technologies in the vision space that are market-ready today:
Visual retail analysis and targeting.
Organizations with brick-and-mortar stores can benefit today from AI technologies through visual retail analysis and analytics providers. Based on comprehensive continuous camera footage inside and outside stores, machine learning algorithms can identify business-critical aspects such as the behavioral patterns of repeat visitors to the stores using facial recognition, or the efficacy of different product layouts. These input factors can be used to optimize store operations and design. However, AI-driven visual retail analysis tools don’t need to be confined to the in-store experience. For example, quick-service restaurants today are able to use vision-based AI to reliably read license plates of passing cars to their franchise locations, to then use public third-party data to relate the license plate information to an individual and based on that input and their recognized movement patterns create hyper-personalized marketing communications. Note that retailers need to be cognizant of aspects related to customer privacy when applying these technologies.
- Difficulty of adoption: medium; requires hardware investment and installation
- Providers: ShopperTrak, RetailNext, OpenAPLR
- Impact: Increase sales volume and operational efficiency in brick-and-mortar locations; increase optimization of marketing spend through hyper-personalized marketing
Image and video recognition for user-generated content.
Brands have a strong interest in engaging and reacting to both positive and negative conversations that involve them in any online and offline channels. With visual-driven social channels such as Snapchat, Instagram, Pinterest and YouTube in wide adoption, many digital conversations and mentions however no longer occur in a more readily processable textual manner, but instead through the means of user-generated images and video content. Using deep-learning techniques based on training data, technology solutions have been developed based on Vision AI that allow brands to monitor social channels at scale and recognize any relevant visual patterns such as a brand logo or individual products. These discoveries are then served to the social teams for review and direct consumer engagement.
- Difficulty of adoption: low; mostly stand-alone software with API integrations to the industry-leading social management tools
- Providers: ClarifAI, Cloudsight, Indico, Dextro
- Impact: Increase consumer engagement and customer satisfaction; improve operational efficiency within social marketing teams
Smart digital asset management.
Leveraging a similar technological approach as UGC media processing, modern DAM solutions leverage AI technology to apply semantical metadata to marketing assets automatically, eliminating the need for exhaustive manual tag curation. Many large-scale marketing organizations suffer from a lack of organization and tagging structure in their digital asset operations. Since machine-based tagging increases discoverability of already produced and licensed media assets, media production costs and licensing costs can be reduced significantly.
AI-based auto-tagging features are able to recognize and tag assets with advanced concepts such as “family on a beach” or “smiling kid eating ice cream”.
- Difficulty of adoption: medium; may require DAM replatform or extensions
- Providers: Adobe Smart Tags, Asset Bank, Google Cloud Vision API
- Impact: Increase operational efficiency by reducing the need for manual curation of asset tags; optimize marketing spend by decrease licensing and media production spend
Language-driven marketing AI applications.
Language-based AI has the longest history in Academia; researchers have studied for decades how to - both in a verbal and written context - accurately understand and semantically process the human language (Natural Language Processing) as well as how to generate it (Natural Language Generation and Text-to-Speech). Amazon’s Alexa, Apple’s Siri, and IBM’s Watson have become household names to tech-savvy consumers and represent the speech-enabled side of digital products that support our daily lives. The recent maturation of language understanding, processing, and generation has opened up a whole new channel of consumer interaction.
Conversational interfaces and chatbots.
A large share of the younger generation’s online time is spent in written message-based communication channels such as Facebook Messenger, Whatsapp, or text-messaging apps. These communication platforms are more and more evolving to full eco systems by opening themselves up to brands through SDK and API integrations. Leveraging the Facebook platform, for example, Burger King is currently testing a chatbot that will allow users to order food from a nearby branch and pay from within the Facebook Messenger app. Conversational UIs provide the promise of a more direct and natural conversation with a customer and platform-as-a-service providers that allow the quick stand-up of such conversation applications are on the rise. The scale of maturity of these application hereby is wide; a successful adoption strategy for an organization can be to start with a common but narrow service need that can already significantly reduce human service support need e.g. traditionally managed through a call center. Following an iterative delivery model, the bot’s domain knowledge and service function can be gradually expanded from there.
- Difficulty of adoption: low for an initial concrete use case up to high for advanced applications; established SDKs and APIs allow for quick go-to-market
- Providers: Facebook Messenger, Alexa Skills API, Converse.AI, IBM Watson
- Impact: Improve customer relationships through quick access to human-like support; increase operational efficiency by reducing human support bandwidth needs
Marketing messaging generation.
Another maturing application use case of AI in Marketing is Machine-driven language generation. This technology offers the promise of truly communicating with every single customer and prospect in a truly individual and personalized manner, with the goal to make a fully relevant and emotional connection to trigger a particular action. Using a continuous test and measure approach artificial content generation platforms are able to autonomously learn a communication strategy that generates the most personalized and conversion-optimized marketing message for each individual customer. Based of a general vocabulary, marketing communication templates as well as the customer’s known preferences and observed behavior, these systems are optimizing your marketing communication strategy through machine learning to to interact with every customer at the right time, through the right channel, with the right tone, and the most relevant content.
- Difficulty of Adoption: medium; typically requires several integration with systems driving direct consumer communications such as ESP
- Providers: Persado, IPSoft, Automated Insights
- Impact: Improved customer relationships through more personalized messaging; increased sales through optimized conversions for direct consumer communication channels
Automatically detecting the sentiment of a customer communication at scale is a powerful marketing tool for both written and verbal communication. Natural language processing technology has evolved to a point where it can fairly accurately recognize sentiments in a customer voice, such as the degree of frustration. Such classification through AI-based system can be leveraged in call-center applications. For instance, the technology can automatically trigger an escalation from an automated interactive voice system to a human representative to ensure customer satisfaction when tensions on the customer side are rising. Similarly, natural language technology can detect sentiments at scale for written user-generated content in social channels such as Facebook, Twitter, or community comments. Such tools can assess and benchmark brand perception e.g. against marketing campaigns or dynamically moderate UGC content.
- Difficulty of adoption: low; mostly stand-alone technology
- Providers: Lexalytics, Sysomos, Crimson Hexagon
- Impact: Increase customer awareness and insights and with that relationships; increase operational efficiency by reducing the need for manual social media monitoring or UGC content moderation
Insights and discovery-driven AI applications.
A raising application of AI techniques and particularly machine learning has been around mining large sets of data and deriving actionable insights from it; this approach can be exercised in a supervised manner where the systems mostly serve insight to inform decisions and optimization strategies developed by humans as well as in an autonomous unsupervised say were the AI system directly leverages its machine-driven insights to optimize a certain machine-driven behavior. Below we introduce the three most common applications of such approaches in today’s marketing world.
Programmatic advertising is among the most adopted and mature digital marketing technologies that make heavy use of AI today. Based on machine learning algorithms, programmatic advertising tools are able to learn an optimized decision-making strategy for buying advertising space as it relates to which audience, demographics, and keywords to target and at what price. Given the bidding-based and real-time sales mechanism of the industry-leading paid advertising platforms, sophisticated programmatic approaches to media buying usually far outperform the strategies applied by humans. Hence programmatic advertising can be considered a necessity for any organization to optimize their online media spend and improve the performance of their campaigns.
- Difficulty of adoption: low; mostly stand-alone technology
- Providers: Adobe Media Optimizer, Rocket Fuel, Kenshoo
- Impact: Optimize marketing spend through learnt and continuously improved buying strategies; increase operational efficiency by reducing human management needs for paid media campaigns; increase direct sales by increased conversions
Lookalike audience modeling.
Lookalike modeling is another trend in marketing technology that’s being rapidly adopted. Usually, this technique is baked into so-called data management platform tools (DMP) that allow organizations to aggregate first, second, and third-party data to determine and manage target segments and consolidate user profile information. Lookalike modeling features are based on machine learning algorithms and automatically discover new target segments based on a significant overlap of characteristics with existing customers. For example, a retailer’s learning algorithm may discover that the characteristics of recently converted customers who purchased a winter jacket have a significant overlap in demographical and socio-economical characteristics with individuals who visit the websites of skiing portals in the U.S. The later segment can now be targeted specifically e.g. through display advertising campaign on ski portal web sites to extend the customer base with a high probability of above-average ROI for the media spend.
- Difficulty of adoption: low to medium; usually stand-alone platforms; requires integrations with systems that hold important user data; requires development of an Enterprise-wide data model
- Providers: Oracle Bluekai, Adobe Audience Manager, DoubleClick by Google
- Impact: Optimize marketing spend through highly targeted advertising; increase sales by discovering new market segments with a high-probability to convert
Algorithmic real-time personalization.
Most of the personalization performed on the web today is driven by of a set of manually curated rules that look for certain contextual data points such as a user’s location, customer status, or projected household income to then deliver varying content and messaging based on a content author’s assessment of relevancy. Algorithmic personalization aims to use machine learning techniques to dynamically personalize a website in real time within a user’s browsing session. In a commerce context, this can involve dynamically offering a discount to the user when the probability of a session and cart abandonment reaches a certain threshold based on an established probability models. The machine learning techniques used by these providers are often based on an unsupervised learning approach called reinforcement learning. In this approach, the algorithm optimizes its actions against a fixed reward function, such as e.g. the converted cart size of a user. Every time a positive outcome is reached (e.g. a customer buys a set of products) the AI system looks back at its taken decision - in this case the applied personalization actions - and applies a reward to it. In the opposite case, where a user session is abandoned without checkout, it discounts the taken personalization actions for future sessions. The algorithm constantly explores alternative actions at a small scale so it can adapt to contextual changes such as shopping behavior during the holiday season. This is referred to as the balance of exploitation - leveraging past data of which personalization action performed well - and exploration - constantly trying out new actions and observing their success.
- Difficulty of adoption: medium to high; often requires deep integration with existing commerce or content platform
- Providers: NeoWize, Bloomreach, Sentinent
- Impact: Increase sales by providing relevant offers and reduce cart abandonment; improved constumer relationships through a more personalized and relevant experience
Anticipation and decisioning-based marketing AI applications.
The last but not least category of applications of AI in marketing are tools that focus on anticipating user actions. While there is overlap to the presented insight-based applications, this category of systems usually look ahead and try to predict a future customer action and optimize the experience accordingly with the aim to reduce friction and narrow or entirely eliminate choice.
AI-driven product recommendation engines.
While production recommendation engines in eCommerce have been around for a long time, a new generation of tools heavily driven by AI have surfaced more recently. Traditional recommendation engines primarily have used a technique called collaborative filtering which identifies product recommendations based on overlap in buying behaviors between customers. Those approaches, however, were prone to the cold-start problem, the scenario when no purchase data of a new customer was known. Newer, AI-driven big data approaches consider a much broader set of data including contextual information such as the used device, time of day, or off-site activity fed through third-party providers. As an example, clothing retailer Under Armour is using IBM Watson to generate higher performing product recommendations by analyzing customer purchase data along with third-party information on fitness and nutrition. These products can also learn when a positive return is expected from engaging a customer in a quiz around their preferences, and what precise questions to ask in those scenarios. An example for such approach is Northface’s “Find the perfect jacket” experience that through an interactive dialogue narrows down its product recommendations.
- Difficulty of adoption: medium; commonly requires product domain modeling and external data integrations
- Providers: Amazon DSSTNE, LiftIgniter, IBM Watson
- Impact: Increase sales through strong, relevant product recommendations even for new customers; build strong customer relationship from the start by given the customer a sense of being understood and properly served
Predictive analytics and anticipatory design.
In layman’s terms, predictive analytics are platform that provide predictions about the future and thereby expand the traditional digital analytics approach focused on analyzing and reporting on historical data such as visits to a website, average time of engagement etc. Its most widely adopted applications in marketing are with CRM applications. Using machine learning and correlation techniques, the Salesforce Einstein product for example aims at predicting and alerting marketers about expected business-critical pivot moments such as a files customer being particularly receptive to a sales conversation based on observed online research activities, or an existing customer carrying a high-likelihood of churning when a contract expires based on his usage behavior. Such predictions are based on myriad data points that span from general market trends to personal micro data. However, predictive analytics isn’t exclusive to sales operations. The technology can also be used to improve CX (customer experience) by predicting a user’s next action and choices. Using those predictions experience designers can narrow down, sometimes even eliminate, the selection of options for users and with that address the paradox of choice. This design approach is often referred to as anticipatory design.
- Difficulty of adoption: medium to high; commonly requires data integrations across system and custom algorithm tuning for prediction engine
- Providers: Salesforce Einstein, IBM Predictive Analytics, Marketo
- Impact: Increase sales by acting on machine-driven predictions; improve consumer satisfaction and customer relationships by an improved, personalized user experience
The Way Forward.
Marketing leaders who are not yet leveraging AI-based platforms in their day-to-day operations are urged to identify and prioritize opportunities as they fit their organization needs. The above list can provide a starting point and inspiration for mature impactful applications. We recommend Marketing teams develop a prioritized roadmap of AI-based initiatives based on the collective value as represented by the expected user value, business impact, operational readiness, and implementation effort.
Marketing teams are advised to include their IT departments closely in this exercise, as often AI solutions rely on consuming existing enterprise data as well as integrating in existing marketing technology applications. Further, often AI solution providers in a particular category have similar functional offerings, but the compatibility with your existing technology footprint may vary, which can be a deciding purchase factor and fuels the need for a close marketing and technology collaboration.
Before investing in and rolling out Artificial Intelligence solutions, organizations should identify and measure a clear benchmark of key metrics the new technology is supposed to improve. This is a prerequisite to develop a total cost of ownership model that should inform any purchase decision. Given that most AI technology providers offer a SaaS software licensing model, it is often possible to try technology solutions for a certain period and measure their potential impact ahead of purchase. What needs to be considered is that many machine learning algorithms require some time to be either trained in a supervised or unsupervised manner to show their full potential. Hence enough time and data needs to be allocated to trials to extract meaningful results and projections.
In conclusion, artificial intelligence technology is well aligned with its literal meaning: computational algorithms that intend to replicate human intelligence and, ideally, improve upon it. In the context of marketing, this can be translated to the emotional intelligence and strong understanding marketers need to develop toward their customers to be successful. And just like in real life, this is a learning process based off past experiences and observed behaviors. In the end, customers will continue to be emotionally drawn and connected to brands that give them a sense of being understood in their individuality; independent of whether that understanding has been developed manually or artificially.