An ounce of information is worth a pound of data.
An ounce of knowledge is worth a pound of information.
An ounce of understanding is worth a pound of knowledge.
Russel Ackoff
If big data is the new oil, analytics is the combustion engine.
Gartner
Complex algorithms are considered the driving force of the digital world. They are said to open unprecedented opportunities and give competitive advantage if applied to the right business models. Success stories make it to widespread media. For example, Amazon gets 35% of its revenue from the recommendations generated by their AI algorithms, and has set up a customer support system that is 90% automated which is estimated to save the company costs by some 3 to 4 percent. Netflix has estimated that its Recommender Systems influence around 80% of the hours that are streamed on their platform. Alone on their homepage, which constitutes the main place for their recommendations, 2 out of 3 hours of consumed content are discovered.
Why are AI Applications so widespread in marketing? Marketers have long known the importance of knowing their audiences to obtain the desired changes in their behavior. The key lies in the revolution that digitalization has brought in terms of what is now possible to know about customers. The famous 4 v’s model of the characteristics of big data, is very helpful to apprehend why the match AI-marketing has been so successful:
– Volume refers to the quantity of information that is now possible to store about a person (preferences and behavior). With this new behavioral information, it is possible to build much more accurate response models that predict the acceptance probabilities of a given marketing measure.
– Variety refers to the increase of diversity in the sources of information that now serve to generate insights. In particular, new forms of unstructured data (text, images, videos), coming from sources like social media and product reviews, or through the usage of intelligent devices (Internet of Things) are now giving light to much more specific and accurate knowledge about people attitudes and preferences.
– Velocity means that digital media has increased the speed of reception of new information, which in turn means that it is more readily available to enter the decision-making process.
– Value refers to the well-documented fact that data-driven decision-making processes perform better than traditional ways.
There is an underlying phenomenon in the diversity of applications of Artificial Intelligence and Machine Learning: personalization. Successful implementation of AI requires a general mind shift in the way marketing is done and what the role of the marketing specialist is. In this new paradigm, traditional rigid customer segmentation is substituted by dynamic client profiles. This means that marketers now are responsible for giving the general guidelines of the marketing strategies, but it is the machines (and the algorithms) who decide what the potential customer gets recommended, how they are approached, what channel is used for communication and when is the right timing). Becoming a data-driven company is like being in a different state of mind.
AI & Marketing – the Applications
It is time to dive deep into the applications of AI in marketing. To give a well-rounded overview, we explore different steps of the traditional customer lifecycle. Within each one, state-of-the-art applications of Machine Learning and trends in research are explored.
1. Customer Acquisition
The first step in a customer lifecycle is finding potential customers. Artificial Intelligence has brought unprecedented possibilities in this domain. Nicely summarized by Peter Gentsch
It is impossible for humans to tap the 70 trillion data points available on the Internet or unstructured interconnectedness of companies and economic actors without suitable tools. AI can, for example, automate the process of customer acquisition and the observation of competition so that the employees can concentrate on contacting identified new customers and on deriving competitive strategies.
The applications of Machine Learning to acquire customers are manifold. For example, in a joint research program funded by the U.S. Department of Energy’s Solar Energy Evolution and Diffusion (SEEDS), demonstrated that Machine Learning methods were an important source to decrease costs in the residential solar industry because they help optimize and personalize several points of customer acquisition: targeting for advertisement, gaining understanding of what motivates customers and giving more timely responses to competitors.
Targeting the right prospective clients is a decisive factor that determines the success in customer acquisition. The most common way to frame this business problem is to use response models that, taking into account information about the customer behavior, attitude and characteristics, will estimate acceptance probabilities of a given marketing campaign or marketing measure. For example, the German company Otto uses an interesting approach. They use a targeting system is based on attribution modeling. This means that they estimate the value proposition or attribute of each possible touchpoint with a customer. Like this, a personalized mix of communication channels is chosen for each client, and the invitation to make a purchase is done at those touchpoints where the likelihood of conversion is maximized.
AI for Customer acquisition has also proven relevant in Business-to-business (B2B) scenarios. For example, Meire et al. (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3130661) made a real-world experiment in which they analyze the impact that including information retrieved from social media for the acquisition of new customers (selling points) for Coca Cola Refreshments USA. In their experiments, introducing the extra information from social media led to models that were able to achieve an increase up to 6.7 percentage points the response rate of potential selling points.
2. Maximization of customer engagement
Engagement can mean a multiplicity of things in the marketing setting. A lot of times it refers to making customers perform and hopefully continuously repeat an action (e.g. make a purchase, consume some content). A very interesting class of algorithms that addresses this business problem, has gained widespread attention: Recommender Systems (RS). This technology changes how the market works, nicely summarized by Jeffrey M. O’Brien (https://www.press.umich.edu/296888/best_of_technology_writing_2007) as follows:
The Web, they say, is leaving the era of search and entering one of discovery. What’s the difference? Search is what you do when you’re looking for something. Discovery is when something wonderful that you didn’t know existed, or didn’t know how to ask for, finds you.
The state-of-the-art research on Recommender Systems takes personalization to a new level: technology allows for algorithms to make different recommendations to the same person taking into account the context (people will probably want to watch a different movie on a rainy Sunday afternoon than on a Wednesday evening), location (take into account the place where people are to improve the relevance of recommendations, for example restaurants or bars nearby), and even intention (the models seek to understand what people’s concrete intention is with each new session –are people just searching, looking to buy, satisfying curiosity).
Recommender Systems have gained a lot of popularity and are now used in very varied applications. One very interesting example is JustGiving, an online donation platform from the UK. Together with demographics and transactional data, they used a multidisciplinary approach (Biology, Behavioral Economics, Psychology) to pinpoint the most effective ways to unlock generosity (i.e. influence people’s decision to donate). Their Chief Data Scientist, Mike Bugembe, describes that (https://www.youtube.com/watch?v=EBGp4oLUk8o&t=1189s) , according to their research, a key element for donations is trust. They used this insight to build a Recommender System that is based on network analysis. The premise is that you trust people that you know, therefore if a donation cause can be linked in some way to someone you trust (for example, because they also donated for it), the probability that you donate will be higher. They also declared to perform experiments to see how people react at the hormonal level when presented with different recommendations.
3. Customer Satisfaction
Keeping customers happy is one of the main goals of marketing. Customer Support is one of the most important touchpoints to determine how happy a customer will be. This is a critical moment that might determine the loyalty that they might show towards a company: if they feel like their problem was solved, they will be more willing to stay. Enterprises have a lot of interest in making this experience as personalized and quick as possible, however, the costs of having a lot of employees dedicated to this task can outweigh the benefits.
One of the most successful technologies since 2014 has been the development of conversational platforms and chatbots. These technologies use a special branch of Data Science called Natural Language Processing (NLP) which focuses on using text and speech as inputs for the model. Due to the special characteristics of language (for example, the fact that humans can differentiate different tones and feelings from the same series of words), this area has been particularly challenging. But applications (https://dl.acm.org/doi/10.1145/3064663.3064672) of these models are at a stage where real-time conversations between humans and machines are possible.
This technology is a good example of the trend of democratization of AI for Marketing Applications being a new business model. Big companies like Google (DialogFlow), Microsoft (Azure Bot Service) and IBM (Watson Assistant) as well as completely new players like Cleverbot or ManyChat are now creating the overall technology for people and companies to personalize their conversational agents to their needs. This means that a company does not need an NLP-specialist to start from scratch. The best technology is available for widespread use. It is already possible to have multi-platformed solutions, that take both text and speech as input and that can even detect the sentiment of the conversation they are having with the human being.
4. Identification of Customer’s Lifetime and Churn
Situations where customers have repeated interactions with a company are the perfect setting for Machine Learning algorithms to flourish. The behavior that a customer has, can help companies predict the customer’s lifetime, the value that he/she will bring to that company (CLV) and identify the moment when the customer is showing signs of defecting the relationship. This, in turn, helps determine the best way to approach the client to try to stop the churn.
One of the most exciting trends in research of optimization of marketing campaigns is the field of profit analytics. This line of research seeks to breach the gap that the most advanced Machine Learning models tend to have with business goals by explicitly taking them into account in the modeling stages. This is especially important for data-driven situations where targeting decisions operate in automatic or semi-automatic form (e.g. when they occur in real-time). For example, a group of researchers (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3130661) from the Humboldt-Universität zu Berlin, propose a two-stage profit-conscious model in which relevant information in campaign planning (like budget constraints, other parallel campaigns, and customer lifetime value) is taken into account. This new modeling approach was tested in 25 real-world data sets from widespread sectors and the results are striking. They estimate that the profit analytic approach would increase the profit for a given campaign by 5% compared to the traditional AI model with the best performance. A 14% increase in profit was expected when compared to its performance with the random forest Machine Learning model, one of the most widespread models that is known for its high accuracy.
Concluding Remarks
Artificial Intelligence is one of the most exciting scientific journeys of our lifetime, one that will have deep impacts on how companies relate to their customers. Data is allowing unprecedented knowledge which has the potential to be used to influence the behavior of people. But with great power, comes great responsibility. One of the biggest challenges that societies face is to set the right legal framework that prevents companies from abusing their power. The lack of flexibility and domain-expertise in governments is delaying this urgent matter. Any ethical company needs to develop a funded data governance strategy together with their marketing actions, that makes it clear for their customers, how exactly their data is being used.
A lot of ground was covered. Some of the most important take-aways:
– Digital footprint, that is, all the information that people leave with the usage of internet and intelligent devices (Internet of Things), has become the new raw material that fuels the flourishing of AI models in marketing.
– Marketing, being traditionally inclined to influence people’s behavior through understanding of their characteristics, has gone through a revolution with all the new possibilities that data and data analytics has brought. Big data has 4 main characteristics that help frame the success of AI implementations in marketing: 1) volume, 2) variety, 3) velocity, and 4) value.
– Applications of AI in marketing are becoming ubiquitous. Whether it is to attract new customer though personalized targeted campaigns that know how to attract people (which channels to use for the contact, what offer is most likely to work and what the right timing is); maximize customer engagement thought recommendations that will not only take into account tastes and preferences, but also what the context of the recommendation is (when is it made, which location and what intent the customer might have at a given moment); or maybe to create systems that will raise flags in a company when customers are about to quit.
– In terms of the fields of application, AI is extending its applications to all imaginable aspects of life. Traditional pioneer fields like retail, social media platforms and media streaming services (films, TV-Shows, music), have been the first ones to take advantage of this technology. But other markets are quickly catching up. Now, extremely interesting applications of AI in marketing can be found in renewable energy marketing, charitable donations, and much more.
– Machine Learning algorithms usually focus on applications for Business-to-customer (B2C) companies. However, there is also a lot of potential for Business-to-business (B2B) companies to use AI implementations in their marketing campaigns.
– The relationship between man and company and also between man and machine is going to have interesting developments in the next years. People will get more information through conversations with platforms. This opens new possibilities for companies to create a relationship with people and enhance their experiences. Applications in customer support are very important.
– There is a lot to gain from the cooperation between business and AI methods. The direction in which methods help optimize business outcomes is quite clear. But the study from Lessmann et al., also shows that the other direction is equally important. Including business context variables (for example in campaigns), can lead to significant profit gains.
In Data Talent we follow closely the scientific developments of this important field. We believe that AI applications will play a key role in the big changes that will come in the next years. Years of experience give us the necessary knowledge to help you find the perfect match for you and ensure that you take your Marketing strategy or any other area of speciality to the next level.