The days of not knowing what 50% of marketing spend is money thrown out (Henry Ford) are largely over thanks to Big Data and AI. The following chapters will explain and illustrate this.
The automation of marketing processes has been common practice since about 2001, when the collection of Big Data became more important. The data sets consist, for example, of customer databases or clickstream data, which is a record of the customer’s navigation between different websites. However, the amount of data has increased explosively, with 90% of all data created in the previous twelve months at the beginning of 2016.
Since many companies do not know how to use these data volumes with their existing database systems and software solutions, the full potential of Big Data is far from being realized. In addition, traditional methods of marketing automation do not provide deep insight into the data, do not suggest actions, do not predict the impact of actions, and do not influence customers in real time. However, when algorithms are used for marketing, the data sets can be processed more efficiently.
Algorithms can analyze and break down large data sets and identify patterns and trends. They can monitor changes and make recommendations for actions in real time, i.e. during the interaction with the customer. In addition, by using algorithms, marketers can address more sophisticated tasks, which can result in a more efficient and cost-effective marketing process. In the long run, a company can gain a competitive advantage by using algorithms in marketing and achieve higher customer loyalty by increasing customer proximity.
AI Marketing Matrix
Today, there is already a large number of potential applications for marketing based on artificial intelligence. In principle, these potentials can be divided into the dimensions “automation” and “augmentation” and according to the respective business impact. Augment applications are particularly concerned with the intelligent support and enrichment of complex and creative marketing tasks that are currently still usually performed by human actors. For example, artificial intelligence can support the marketing team in media planning or the generation of customer insights.
The augmentation potential is already more pronounced in those companies that have a high level of AI maturity in the AI Maturity Model. Here planning and decision-making processes are also supported or already executed by artificial intelligence. With regard to automation applications, it is not surprising that both the degree of maturity and the distribution of these applications are much higher in comparison. Thus, there are many automation applications that already have a high degree of maturity and use in practice. These include marketing automation or real-time bidding, for example.
However, there are also applications which, despite their high degree of maturity and high business impact, are still used comparatively little in practice today. One application area to which this phenomenon applies is the principle of lookalikes, which can be used for Leas Prediction and Audience Profiling. In the B-to-C area this can be implemented well with Facebook Audiences.
This principle can also be successfully applied in the B-to-B sector. Behind this is the possibility of strategically identifying new potential customers who are similar to the best and most attractive existing customers of a company. It is easy to understand how this works: Customers – in the B2B sector these are companies – can be characterised by various aspects. These include not only classic firmographics such as location, economic sector and turnover of the companies, but also information about their development, digitality and their thematic relevance. In times of Big Data, this enormous amount of available information can be obtained primarily from the companies’ websites, because every day, up-to-date articles on new products, changes within the company and many other topics are published on the website and in social networks. Based on these aspects, a comprehensive characterisation of all companies can then be carried out, on the basis of which a generic customer DNA is generated. In a subsequent step, this generated generic customer DNA can be used to identify other companies that have the same DNA – the so-called lookalikes. The result: a pool of potential new customers whose approach offers promising opportunities.
Ultimately, the use of AI-based, automated applications can significantly increase the conversion rate in both marketing and sales. Practical examples show an increase in the conversion rate of up to 70 %. It is therefore clear that the principle of lead prediction and the determination of so-called lookalikes is an application area with considerable potential and a major business impact for marketing and sales.
Advantages of algorithmic marketing
Efficient analysis of data sets
Grouping the data
Recognizing patterns and trends
Monitor changes in real time
Reaction to changes in real time
More efficient and cost-effective marketing process
More time for creativity
Long-term competitive advantage and higher customer loyalty
customer journey intelligence
Based on Big Data Tracking, the customer’s “customer journey” can be systematically measured via various touch points such as search, social media and advertisements. This way, media and marketing planning can be optimized with the help of so-called attribution modeling based on the data thus obtained. The data mining model calculates the optimal channel mix from a variety of data and time points by automatically calculating the value contribution of each touchpoint in the overall channel context. This makes it possible to say exactly which touchpoints have a direct conversion function and which have more of an assistant function. It is also possible to draw conclusions about the temporal cause-and-effect chains.
Interesting and important for companies to store is above all the customer data, from the pre-acquisition phase to the conclusion of the customer relationship – in a sense, the entire so-called “customer journey”. By combining this customer data with other billing information, with customer service aspects and other sales and marketing aspects, intelligent algorithms can make business decisions, derive recommendations to the entrepreneur and conduct market research.
The customer’s path to the purchase of a product already provides strategically valuable information. This customer journey to the purchase decision usually takes place in several cycles, ideally in six stages: Identifying demand, research, obtaining an offer, negotiation and purchase, post-purchase and word-of-mouth communication. The touchpoints are the starting points at which data is collected and analysed, such as tracking data or clickstreams. In this way, predictions can be made about future customer-journey patterns. Networked contact points can be prioritized as part of a digital strategy.
The advantage of this data- and analytics-driven approach is the empirical grounding. Data are neutral and objective, and they give the same message on Monday morning as on Friday just before closing time. The digital “giants” such as Apple, Google, Facebook and Amazon demonstrate how much corporate success is determined by data sovereignty, data quality and data diversity. The information is more up-to-date, faster and more available than an annually recurring internal campaign to “once again maintain the CRM system better”.
Data protection and data sovereignty
In principle, a distinction must be made between personal data and data relating to companies. As soon as it becomes possible to draw conclusions about a specific individual person and work is also carried out on individual data record levels, it is necessary to pause for a moment before processing and to consider: What is processed? Does a business relationship already exist? Are we before or after a business relationship? Which consents or legal permissions exist?
Customer data may not be collected without consent and may not be resold.
Anyone who acts too recklessly in this regard will quickly become liable to prosecution. In general, however, the following applies: With the explicit consent of the customer, almost anything is possible. That’s why Facebook, for example, is able to use the data to such an extent. This is because the consent of the users is present, even if only a few of them may have read and understood the terms and conditions in full. Similarly, relatively extensive data processing is possible and permitted within the framework of an ongoing customer relationship under the keyword “for own purposes”. This may include market research, acquisition activities and advertising. In connection with digitisation, the keyword “data sovereignty” is often used: It is indeed existential for entrepreneurs, because nobody can or wants to disclose more data on the Internet than is absolutely necessary.
Data sovereignty means nothing other than knowing exactly what happens to one’s own data and only sharing as much data as is absolutely necessary. This also includes critically examining the use of one’s own data and the online services, portals and databases used. Service providers should be particularly careful to check how they handle the company data entrusted to them.
For entrepreneurs, data sovereignty therefore means deciding for themselves who, when and where may find, use and pass on which data. The following chapters will first address the use of algorithms in all four steps of the marketing process. Afterwards, practical examples and suggestions for the correct handling of algorithmic marketing are given. Finally, the anticipated effects of algorithmic marketing on the economy as a whole are briefly described.
Algorithms in the marketing process
Algorithms, e.g. in the form of bots, can be used in all four steps within the marketing process: in situation analysis, in marketing strategy, in marketing mix decisions and in implementation and control. The situation analysis should identify the unfulfilled wishes of the customers. In the internal situation analysis, bots can be used to determine the Key Performance Indicator, which provides information about the strengths and weaknesses of the company. In an external situation analysis, bots can search the Internet for specific keywords to learn more about customers and competitors. Consumer behavior can be observed and evaluated with the help of bots.
If companies use chat bots in customer service, bots can observe and evaluate the course of conversations to obtain more information about the market and customers. Bots can also conduct interviews with specific customers or trend experts to perform qualitative analysis. This can save both money and time, as the interviews can be conducted in different locations simultaneously. Algorithms that can make predictions about the influencing factors and effects on marketing activities (Predictive Modelling Algorithms) can be used to explore future demand. In the second step of the marketing process, the creation of the marketing strategy, target groups can be identified using bots that segment the customer base and analyze it according to various characteristics.
However, defining the product’s value proposition requires both creative and analytical skills, and therefore this task is less suitable for automation. A widely used tool to implement strategic decisions is the marketing mix with the four Ps: Product, Price, Promotion and Placement.
Algorithms can be used in the different areas as follows:
Product: Chatbots can be used e.g. in customer service. Furthermore, algorithms enable companies to develop new and innovative products and services that are tailored to the needs of the customer.
Price: algorithms can be used to automatically change product prices according to demand, availability, and competitive pricing. Examples of companies using this dynamic pricing are airlines, Amazon and Uber.
Promotion: Algorithms using AI can learn the buying behavior and needs of customers and therefore display individualized content and product recommendations to customers. For the company, this is more efficient, cheaper than mass advertising and can be done in real time. In addition, sophisticated self-directed recommendation systems can increase the chances of cross-selling, the offering and sale of additional products.
Place: Bots facilitate electronic business transactions, also known as e-commerce. If payment information and delivery address are available, the entire transaction can be carried out by bots. Based on previous purchasing behavior, a personal butler can also autonomously decide where a product is purchased. However, this can also be problematic, as it is no longer possible to measure the customer’s purchasing behavior in the long term. The question also arises of how to proceed in the future with regard to brand management.
Many aspects in the last step of the marketing process, the implementation and control, can be taken over by algorithms. Examples for the implementation of marketing strategies are e.g. the placement of advertising, the launch of a website or the sending of e-mails. As explained above, bots can display individualized Internet advertising. Bots can also create, personalize, and send marketing campaigns via e-mail. Bots can even be used to create websites, and The Grid has been offering a private beta version since 2014.
The control phase at the end of the marketing process can be carried out both qualitatively and quantitatively and is essential. Factors that should be monitored include the campaign’s reach, marketing budgets, customer satisfaction, market share and sales. Algorithms can be helpful here to measure the various factors and to make statements about the efficiency of the campaign as well as to uncover potentials, for example with regard to an increase in the customer life time value, the customer value during the whole life of the customer, or a reduction of the customer acquisition costs.
Algorithms can also improve the accuracy and efficiency of control. The evaluation and presentation of the analyzed data can be handled by smart process automation software, which is capable of training itself or being trained. It can perform more complex and subjective tasks by recognizing patterns. In addition, the data can be visually interpreted in the form of dashboards.
In some industries, the use of algorithms is already commonplace, such as in manufacturing for process control and in the financial sector for stock trading. Recently, it has also been shown that algorithmic marketing can increase the turnover of companies.
One example is Amazon, which uses algorithms and has even grown during the recession. It is striking that the company has invested comparatively large sums in IT (5.3% of sales revenues), while competitors Target and Best-Buy only spent 1.3% and 0.5% respectively. Amazon’s dynamic price calculation responds to competitors’ prices and current inventory levels. Investment in complex recommendation algorithms has automated 35 % of sales and 90 % of customer support. This reduced Amazon’s costs by three to four percent.
The Otto Group uses Big Data and AI for marketing and media controlling. On the basis of Customer Touchpoint Tracking, a customer’s activities can be systematically measured via various touchpoints such as search engines, social media and online advertising. With the help of so-called attribution modelling, the Otto mail-order company has optimised its media and marketing planning on the basis of the data thus obtained. The model calculates the optimal mix of communication channels from a variety of data and time points by automatically determining the value contribution – the attribution – of each touchpoint. This makes it possible to say exactly at which touchpoints the customer is directly encouraged to buy, i.e. which have a direct conversion function and which have more of an assistance function.
The temporal cause-effect chains can also be derived. From this, Otto systematically derives marketing measures and media budgets. The large number of digital touch points and devices and their extremely variable use by the customer can no longer be optimised by experience and gut feeling alone. This empirical grounding and objectification of marketing helps to question the opinions and barriers in marketing often shaped by the respective channel and contributes to a significant increase in its effectiveness.
BOSCH SIEMENS HOUSEHOLD APPLIANCES (B/S/H)
Classic market research has an extensive range of instruments at its disposal to obtain consumers’ assessments of products. The main disadvantage of these methods is often the effort involved. On the Internet, thousands of product evaluations can be analyzed automatically at any time.
This cannot be achieved systematically without Big Data: Ratings and reviews, which are distributed globally via various Internet platforms, must be intelligently recorded and integrated. In order to be able to react quickly to the product evaluations, this data must also be quickly recorded, analysed and measures implemented. Negative ratings can be answered by companies so quickly. Positive evaluations can be implemented in marketing communication via websites, social presences or other product advertisements. Based on a big-data infrastructure as Software as a Service (SaaS), BSH manages the entire process from the generation, collection, analysis and use of ratings and reviews. These automatic rating and review analyses allow customer ratings to be examined both qualitatively and quantitatively and used to achieve a sustainable increase in sales.
BSH’s internal evaluations show, for example, that products with positive ratings achieve a sales increase of up to 30 %. These product rating and review analyses thus become the new Stiftung Warentest, a modern gold digging tool.
The logistics company UPS has also set itself the goal of saving up to US$400 million a year by using an algorithm to determine the most efficient transport route. The taxi company Uber uses an algorithm to bring driver and passenger together. When a trip is requested, the algorithm offers the trip to a driver who is nearby. This corresponds to the supplier pick model, that is, the supplier selects. Similar to Amazon, the company uses dynamic pricing: If demand for rides in a certain region is high, the price increases by a certain factor that is known to the driver but not to the customer.
Netflix, the online service for playing (streaming) movies and TV series, uses algorithmic marketing to personalize content for users and recommend titles. A total of 800 developers are working on the algorithms with the aim of retaining viewers. The social networks Facebook and Twitter and the online video channel YouTube use algorithms that select the posts that are displayed to the user. For Facebook, for example, the visibility of a (advertising) post is determined by various factors, such as the popularity of the company’s site, the success of past posts, the form of the content (videos are preferred over photos) and the time since the post was created.
However, there are also application examples of algorithms that demonstrate the dangers and limitations of algorithmic marketing. For example, Coca Cola maintained a Twitter account that converted negative tweets into cute ASCII images if they were marked with the #MakeItHappy hashtag. The US magazine Gauker then created a Twitter bot that published lines from Hitler’s “Mein Kampf” and added the hashtag. Coca Cola also converted these into pictures of dogs and palm trees without further verification.
BANK OF AMERICA
Bank of America operated a bot to help customers with complaints via Twitter. When an angry Occupy activist contacted the bank’s Twitter account, the bank sent out the same prompt and standardized responses that are intended for customer help requests. However, Bank of America assured that people, not bots, were behind the responses.
The correct use of algorithms in marketing
As indicated by the negative examples mentioned above, certain dangers lurk for companies that use algorithms in marketing. It is therefore essential for companies that the algorithms used and their limitations are fully understood and that the algorithms are used wisely. Algorithms must also be supervised and controlled so that they are in line with the principles of the company and the face of the brand.
Another aspect is the growing concern of customers about their privacy, which can raise mistrust about the use of algorithms. If consumers see too much personalized advertising, it can be perceived as creepy, especially if the advertising is based on very deep insights into private information. This is also called overkill targeting and can reduce the success of the marketing strategy.
The described creepiness, which the customer can experience, is caused by an imbalance in the distribution of information: The advertising company knows more about the customer than vice versa. Companies should also be aware that the collected and evaluated data gives them an advantage over the customer and that they can manipulate and misdirect their perception. If consumers are only shown pre-sorted information, they have no chance of getting a complete overview. There is therefore a danger that individuals will exploit algorithmic marketing without observing certain ethical aspects. In order to gain consumer confidence, marketers must ensure that algorithms comply with the code of digital ethics and respect privacy, manipulation and selection of information and communication behaviour.
For a successful use of algorithms in marketing, it must also be remembered that not all factors are always analysed in context. For example, the mood of the customer, the weather or the presence of other people can influence the customer’s buying behavior. Therefore, an algorithm should contain as many variables as possible, but also surprising and random elements in order not to be too predictable. Another disadvantage of algorithms is that they are often limited in their ability to analyze why a customer has made a certain decision.
To avoid mistakes like those made by Bank of America, algorithms and bots should be used with caution. Ideally, algorithms should be combined with real human interaction in customer contact. In this context, a distinction is made between two cases: Either the point of contact between customer and company is random, or the customer contacts the company with certain expectations. The first case concerns advertising campaigns or recommendations on websites, through which the customer can be positively surprised if the advertising corresponds to his preferences. This can improve the brand value.
Conversely, a customer who is not interested in the advertising would ignore it without damaging the brand equity. However, if the customer has certain expectations of the company, such as when contacting the company directly about a complaint, the brand may be damaged if the company cannot meet the expectations. On the contrary, in the second case the brand value can increase if a customer is satisfied. This does not necessarily mean that no algorithms can be used in this case. However, it is important that they act under human observation and that people can intervene in the process if necessary.