Last November I was invited to give a course on Artificial Intelligence for Digital Marketing at EBS school in Geneva. EBS stands for European Business School that was created jointly with CREA, which is specialised in digital communications and digital marketing. EBS runs bachelor’s and master’s degrees in digital business, which are nowadays quite relevant. AI for digital business is also very much relevant, as marketeers all over the world are usually among the first to adopt new technologies and new gadgets.
So, how can one leverage artificial intelligence in digital marketing ? I assume that readers know already the theory of artificial intelligence and main approaches (state-based algorithms, machine learning, deep learning), so I will not talk about the ML training process, cost functions, gradient descent, kernels, neural networks and other techniques. I will start with explaining the difference among very similar expressions used around artificial intelligence, then I will describe high-value use cases, and in the end how to successfully implement all this, so that your project flies like a rocket. So, I will stay with high-level and business-relevant explanations.
Terminology and Definitions
Data warehouse and business intelligence
This is already an old concept and it emerged from the need to get insights from large data sets, stored in datawarehouses (or large databases deseigned for analytical and not transactional purposes). Data is loaded once per day from transactional ssystems, then extractions are made to several specialized data marts, and in the end we would use different tools to create reports and dashboards for ‘business intelligence’ users. End users interpret these reports and don’t need to have much knowledge about data analytical techniques.
Data mining
Data warehouses and business intelligence gave birth to data mining. This is more manual work, done by data analysts with objective to extract some additional knowlegde from these datawarehouses. Data mining is a synonym for knowledge discovery. The techniques used in data mining are statistical methods like clustering and regression. There are several standards covering data mining, like for example CRISP-DM.
Data analytics
Data analytics is broader and more professional approach to analyzing larger and more complex datasets. It is still based mostly on structured (tabular) data, but it includes also sophisticated data visualization techniques. There were three stages during the evolution of data analytics:
Descriptive analytics : Uses historical and current data to understand what happened and why it happened.
Predictive analytics : Uses statistical models to answer what will happen and provide probability for these predicted outcomes.
Prescriptive analytics : Makes predictions and suggests what to do to take advantage of the predicted situation or to prevent negative events.
Big data
The available external and internal data was growing, largely due to internet, smartphones, social media, and lower storage prices, so it became valuable to analyze huge datasets reaching petabytes (aka Big Data). This data includes also unstructured data (text, images, videos, sound) and we can characterize big data through high volume, velocity and variability. In order to perform analytical tasks on such data, it is necessary to break it down in partitions, and distribute the processing to multiple worker nodes, coordinate the work and summarize the results by a master node. Hadoop and Apache Spark are the technologies to use, and the data should be stored on Object Storage devices (or cloud service).
Data Science
Data science is more recent expression and even more advanced domain. It is based on sophisticated statistical methods and is coded algorithmically mostly in programming languages R and python, with the objective to learn from data, to develop models and use these models for predictions. Still, there is quite a lot of manual processing of structured data sets, where data scientists need to understand the attributes, to clean the data, to perform data pre-processing, to determine and eventually create additonal features, to experiment with data, use machine learning to develop statistical models and create relevant visualizations. Behind data science there is always a data scientist with multidisciplinary skills in statistics, computer science and business, supervising the details of the whole process.
Artificial intelligence
As much as data science is about data, AI is more about algorithms. Modern AI as we know it nowadays is based on deep learning, which is able to ingest unstructured data and automatically create hierarchical knowledge representation models for future predictions and inferences. Behind this kind of AI there is a computer with GPU cards, advanced deep learning-based frameworks like Pytorch, TensorFlow and models trained, tuned ad deployed by AI developers and deep learning engineers.
Robotic process automation
Even though there is ‘robot’ in its name, RPA is not really artificial intelligence. It is an automation of simple administrative work done by humans. But in this era of big data and machine learning, there is a new and very interesting concept, a combination of machine learning and RPA. Some RPA solution providers gather quite a lot of data from their business customers on how particular processes should be handled, including the exceptions, so they train ML models to handle these exceptions. As the result, they are able to implement more sophisticated automations for existing and for the new customers. This is called ‘intelligent automation’ and is a very exciting new trend. The big idea is to have a completly autonomous, akind of self-driving enterprise. You may check this reference here.
Now that we know what is what, the question is : how do we actually use AI in digital marketing?
AI for Digital Marketing Frameworks
First of all, we need to use a framework so that we know exactly what we are doing, how we are doing it, how we measure success and benefits, and that we don’t forget something important while doing it. There are many frameworks like the following:
Marketing Strategy Framework and AI (see this research paper). It covers the full marketing cycle of Research, Strategy and Action and describes different kinds of AI tools and techniques that can help accelerate and automate the marketing process.
Modeling Customer Journey, where we drive potential customers from Suspect to Prospect, then to Customer, Advocate and Influencer, and how AI can support the positive transitions (and also prevent negative transitions to Churner and Criticizer instead of Advocate and Influencer).
Storytelling framework and how AI can help understand and address the underlying problem of your potential customers, and how it supports proposing your solution and incite your customers use it.
When we create our own mental mindmap based on one of these frameworks for our own digital marketing business case, we will need to identify appropriate KPIs for our project so to be able to measure and control the business success. These KPIs, including AI-specific ones are:
- Business Value of AI (positive impact on ROMI)
- Conversion rate (per digital platform) before and after AI
- Conversion rate per AI tool (how a particular AI tool works on improving the customer journey)
- Content production (number of topics, number of pieces of content per category, positive feedbacks etc.)
- etc
These KPIs will also help us guide our choice when we start deciding what is the best use case or the best project for artificial intelligence, or the best AI tool and approach.
High-value use cases
Now a fun part starts. We need to identify where in our project or in our business lies the potential of AI. In order to do it, we can consider a number of high-value use cases like the following:
1. Programmatic advertising
What is this ? The simplest way of determining what ad to show on which users’ device is through cookies, small files downloaded on your computer when you visit a web page. A special kind of advertising cookies will transmit the information about your surfing behaviour, and this allow advertisers to follow you with their spooky ads wherever you go afterwards. For privacy reasons, these advertising cookies will be abandoned in 2023, so advertisiers in the future will basically have to use more AI in their ad targeting.
Companies buy and sell space on your smartphone or computer display based on auction methods on ad exchanges like Google AdX. In China the whole process is even more complex (see the ecosystem here). Ad bidding is when you place a bid on the exchange to display your ads on people’s screens. One can pay for this service per click of per impression. AI should be used here to optimize your costs. You should use AI to understand your customers, and determine more precisely when you should display what kind of advertisement to increase your conversion rates, and to minimize the money you should pay for it.
There are specialized intermediaries and service providers offering their services to manage this programmatic advertising, but you can also use APIs or develop your own programmatic interfaces.
2. Market analysis
This is standard marketing activity like marketing segmentation. AI can do much better tuning of market segments than you can do manually, and help you create many simulations to develop the best segmentation strategy for your products.
Another is forcasting market needs. This is a classical prediction problem. Traditionally, ARIMA, or auto-regressive integrated moving averages method has been used for this, but nowadays AI algorithms can take into account much more external data, so data scientists can create several market models and explore different scenarios.
3. Customer journey analysis
This will cover lead scoring, funnel analysis and predictions, churn analysis and predictions, feedback analysis (very critical for the success of your campaigns) and brand sensibility analysis.
4. Social media monitoring and engagement
General monitoring of social media discussion topics, sentiment analysis, keywords analysis. Next, a very important domain is influencer identification : who they are, what is their level of influence, if they generate content or repost it, what market segment they can reach, their attitudes, what they think about your brand etc. AI can be also used to identify influencers who are not active on social media (like for example researchers, security experts, TV stars and opinion makers etc.) and you can apply visual recognition tools to identify automatically the objects and brands in influencers’ posts.
5. Public relations
AI can help generate and manage pitches, press releases, statements, images and similar. In certain cases, you want to develop stories around your business, campaign or a product and AI can help you generate content and present it in such a way that is very intuitive and easy to follow. Example is Smarticles from Gardian Innovation Lab, where AI can determine what your customers are mostly interested in, when they read a particular story. This can help customize the way how to display your stories, so you will improve the user experience and the readers will come for more.
6. Recommender systems
This is a must for most of the marketing use cases. You want to be able to suggest something to your customers when they express interest in a product or service : recommendation on how they could use the product, if there is another product they will need or like, with whom they can connect etc. An example: imagine someone likes very much a particular song. So, where and when will there be a concert with this artist ? Are there some interesting articles or books about that song or musician ? Are there musical partitions that one could use to learn to play that song at home ? You can identify all these things, present them to your customer and grow your business. A good recommender system can make a lot of difference.
They can be based on collaborative filtering (which will match the recommendation with other similar customers) or content-based filtering (that will match the recommendation based on the particular customer history) or a hybrid one. A nice example how it can be used effectively is clothing recommendations of Thread company.
There is much more to be said about recommender systems and how to build them, which goes beyond the scope of this article here. Who would like to know more, please contact me through LinkedIn.
7. Customer experience through chatbots
This is a tricky area and needs to be done carefully. Chatbots are according to Gartner’s Hype Cycle ranked very low in 2021, which shows that people are rather disapointed by them. But this doesn’t need to be so. Chatbots can be done in different ways: traditional rule-based chatbots where you need to code everything (example: Pandorabots), generative chatbots that use deep learning and language models, but are still not able to sustain the long conversation, or high-quality retrieval-based chatbots like Google DialogFlow that are still very much used in practice. Read about BBVA in Spain and how they leveraged this technology to build superior customer experience with their chatbot Blue. The latest trends in chatbots are:
-
-
- integration with voice assitants and voice search (it is faster than when you type)
- integrations with backend knowledge databases, knowledge graphs, calendars, booking and transactional processes etc.
-
Chatbot design is a very important topic. One should carefully collect the queries from multiple sources and define the intents for the training purposes, but also pay attention to the tone (formal / informal) and chatbot’s approach (proactive / reactive) when desiging the chatbot, and all according to the business objectives (should chatbot answer the questions or sell the products ?).
As chatbots originally emerged from China (remember Tencent and WeChat), you might be interested to read (here) how minibots can be implemented on that smartphone platform.
8. Gamification
Games that engage customers can make a big difference. You remember 10’000 steps challenge ? This is one of the best examples of how gamification can work for you. And AI can help encourage gamers, reward them, adjust the level to their capabilities, match gamers to other gamers, help them develop their skills etc. And some games can move you emotionally to take actions in the real world (like in this game: That Dragon, Cancer)
9. Content generation
We are talking here about web site design and automatic generation, generators of podcasts, images, logos, avatars, memes, videos, musics, captions etc. Most of the recent use cases are enabled by emergence of very large language models like GPT3 and BERT. You can do text summarization, story generation, question-answering systems, blog article generation and similar. All this helps accelerate your content marketing. Maybe the bext example how this could be done successfully are the stories of Heliograph and The Washington Post, and The Press Association’s use of AI to automatically create press stories.
10. Augmented Reality
This is definitely the future trend, and can’t be imagined without AI. Whether we immerse ourself in virtual worlds like Metaverse or we stay in the physical world, there are many potential use cases where we can benefit from some sort of AI-help. And marketing is everywhere. You can offer financial services, tokenization, guidance, training, recommendations, search, chat, advertisements etc. This field is huge and still very much open for imagination.
Implementation phase
After this long, but not exhaustive list of high-value use cases, we should identify the areas of our business and our marketing project with high potential for AI. Then, we need to identify what data is needed for the successful implementation of these use cases, what data we already have about our customers (online and offline) and how we are going to cover the data gaps (of what data we are missing). Very often we determine that the data is scattered accross different databases, so we need to start from the ‘Project Zero’, which is building Customer-centric Data Platform (CDP).
Implementation of these identified AI use cases can be done in different ways:
– using APIs. Most of these APIs are actually freeware for study and development purposes, so they are ideal for experimentation and prototyping. See for example this platform or IBM Watson APIs. If you had previously some good experience with IBM’s Personality Insights API (which was unfortunately recently decomissioned), you may explore similar Symanto’s APIs: Personality Traits and Communication Style. API-based solutions require minimal coding, but you need to integrate them with others to build your AI pipelines, to perform integration with your front-end applications, and to make sure that your data is clean and of good quality.
– using SaaS solutions / tools. They are developed by established marketing technology providers and can cover broad application areas. There are many companies selling AI tools or different quality and reputation, so you need to do careful due dilligence when selecting them. The disadvatage is that they will use your data to improve their algorithms, and you need to understand how you can migrate away from them, should in some future you decide to do so.
– developing your algorithms internally. This is recommended when you gain experience, and when your AI system becomes a source of your competitive advantage. Then you should hire a team of AI engineers and data scientists, and start bringing all these things in-house.
Implementation choice must be done by solution managers, considering both architectural design and business case.
After each AI project along this AI Transformation program, you need to evaluate the KPIs to understand how AI brings the value to your business. You also need to review the lessons learned – if you should invest more in education, if you should change the strategy or anything else.
Data Quality Model
In the end, I must reiterate the need for high data quality for AI in digital marketing. It is not only “garbage-in , garbage-out”. Imagine putting just one rotten apple to your juicer. This one rotten apple will make the whole juice tastes bad and cause food poisoning. So, one should make sure the data quality is close to perfect and eliminate all rotten apples. But, how do you do this ? There are two things to pay attention to:
1) Data governance. Make sure you implement a formal process like described for example in Data Management Body of Knowledge.
2) Work continuously on the data quality. Make sure that you can measure how clean, complete, credible, compact and quantifiable your data is, by sampling the data, and evaluating how your machine learning models will behave if you encounter exceptions and variations accross these dimensions.
Conclusions
All in all, Ai in digital marketing is a very exciting and dynamic field. Some estimates claim that as of 2021 there are around 10’000 AI-enabled tools for digital marketing! But, not all of them are really based on AI. very often when we dig a bit more, we will realize these are just simple automation solutions, which don’t deserve really an AI label.
AI is highly disruptive technology. It is like a magnifying glass. If implemented properly, it can magnify your abilities like a powerful telescope. In order to exploit its full power, my recommendation here is to learn what AI really is, and what it isn’t. This way you will avoid wasting time with wrong directions or wrong products. Take a structured approach like I described here, and you will avoid overcomitted or failed projects. And last but not least, find the right partner, a consultant or a coach and you will learn faster and be able to leverage accumulated experience from the industry.
In the end, I got a very positive feedback from my 6-days course. I’m also very proud of this generation of undergraduate EBS students that have completed my training, performed with excellence during the exam, and developed very interesting and highly realistic AI marketing projects.
If you like this article, feel free to share it. If you would like to learn more about this exciting topic of AI in digital marketing, I’m open to giving courses and practical workshops to your teams. Feel free to contact me through LinkedIn for a discussion.
PS: Special thanks to my guest lecturers : Dany Cerone from Botte Secrète, who talked about AI-powered website heatmaps, and Beatriz Prieto and Quentin Duborper from Accenture / Wire Stone for the presentations of their marketing solutions based on facial and sentiment recognition, and their amazing AI-powered avatar.
Thanks,
Sasha Lazarevic
December 2021