Inside this article, you will find among others, information about:
- When Machine Learning meets Marketing Automation
- Ways to use Machine Learning in Marketing Automation
- The channels of recommendation delivery
- The new face of Marketing Automation
In the blockbuster loved by millions in the world, the main character travels to the future, to 21st of October 2015. Marty McFly, as it was his name, marvells at the fact that science fiction has become reality. The society by then has been relieved by the machines and enjoyed the creature comforts provided by the advanced technology. The movie reflected high hopes of us all – hopes of living comfortably and happily ever after in the modern world. Nowadays, life may not resemble treacly reality from the “Back to the future”, but we are surely progressing in a dynamic pace. We witness extensive modernization in many fields and observe how simple processes are automated. Also, many have raised the question whether the machines will replace human resources? The question is asked with a great deal of fascination, though it reflects the fear that we, humans, will become useless. One may wonder what is it exactly that makes machines capable of performing tasks and the answer is their ability to learn.
Machine Learning together with the AI seem to be the invention to the 21st of October. It may be surprising to find out that the foundations of the two fields were laid in the late 50s. So what is the difference between Artificial Intelligence and Machine Learning? The difference may be not so obvious as the terms are often used interchangeably and the boundary between them seems to be blurry.
Artificial Intelligence: The term Artificial Intelligence was used for the first time in 1956 by John McCarthy, an American informatician and mathematician. The term itself refers to machines that demonstrate intelligence which makes them capable of performing tasks that normally require human intelligence (e.g. understanding human language, problem solving, etc.).
Machine Learning: If it comes to Machine Learning, the term itself was coined by Arthur Samuel in 1959. The definition states that machine learning is the ability to learn without being explicitly programmed. In practice it means that machine learning involves “educating” an algorithm. Educating in the sense of combining learning from experience, learning from data and following instructions. To facilitate Machine Learning, delivering an abundance of data is necessary, so the algorithm can be trained and thus continue self-improvement. Machine Learning is widely used in the world. An instance of popular application of machine learning is face recognition.
Data Science: Finally, data science, often referred to as data-driven science, is an interdisciplinary field that focuses on scientific methods, processes, algorithms and systems which are used as tools to obtain knowledge and understanding out of the data that takes various forms. Data science utilizes data for designing processes and finds correlations between data. Additionally, it offers a range of solutions from which you can benefit the most. Data science requires a great amount of data and the team of specialists who will be able to conduct the analysis.
When Machine Learning meets Marketing Automation
Machine Learning and AI are used on a daily basis and it makes our lives more comfortable. If you have ever used Google Maps to navigate yourself to the destination or if you have watched movies or series on Netflix; if you created an account on Spotify to enjoy your favorite music; if you have taken advantage of the creature comforts such as taxi service known as Uber, then your fate intertwined with the blessings of Machine Learning. Google, Uber, Netflix, Facebook and many more brands make use of AI and Machine Learning to develop and improve the services they provide. Machine Learning is implemented in the applications and programs that are designed to add up to the creature comforts. But also, or maybe, most of all Machine Learning and AI are meant for serious purposes. This advanced technology finds its place in the marketing branch called Marketing Automation.
How does Marketing Automation work?
Marketing Automation is a dynamically developing subfield of Marketing that aims at optimizing the work of sales and marketing departments. An excellent tool for that is SALESmanago. It helps to design a long business process, which begins when you acquire the contacts as they visit your website and fill in the form. The platform identifies users on the website and monitors their behavior on the basis of cookie files. SALESmanago collects contact data, behavioral and transactional data which is saved in the system. Each user receives points for every kind of demonstrated activity, it can be a visit on the website, opening an email from you, a purchase of the product, the activity on social media, etc. It is an objective rate that reflects the user’s engagement and the degree of readiness for a purchase.
Alongside the scoring system, the platform operates on the system of tags which are assigned to contacts manually or after the occurrence of a certain event. The database is segmented in this way.
All pieces of information are collected on the contact card of each user and at the same time being a rich source of knowledge and a reference point in choosing a marketing strategy towards a particular user. On the basis of this information, you can reach customer through a number of channels.
Marketing Automation gives you a range of methods of automating your marketing and sales activities. All these activities on the platform are coordinated by Marketing Automation Specialist. The specialist, as a human expert, has a vast knowledge, rich experience and knows the tool very well. You can use the tool to segment the database, define buying personas, and use the particular automation processes to reach them in a proper moment with the right content. So in conclusion, a person uses the tool`s features to programme the platform so it can reach specific customers with the right content in right time.
In practice, the expert, though well-equipped with knowledge, experience and a tool, is not always able to predict the non-standard situation in which the customers spin out of control from the typical segmentation path. Every human is an individual entity with their own will, that is why it is hard to group all people in several groups. In such case, the expert can in fact lead to conversion of only a certain part of customers and the rest of them is lost.
This is the place where Machine Learning and AI enter – technologies that can lift the burden. AI and Machine Learning allow to collect, process and utilize Big Data. They consist of the in-built mechanisms and algorithms that let us predict potential interest of customers. It allows you also to personalize the content in any communication channel in real time. Additionally, it prepares reports, analyses data and self-improves which translates into increased effects. Utilisation of these highly advanced technologies and big data can be employed in designing customer journey. The purpose of it is to guide the potential customer from the first stage that is getting him or her interested in the offer to the last stage that is purchase and maintain the contact with this customer.
In the standard expert approach, the expert designs customer journey, applies automation rules, workflow and starts the whole process. In the end of the customer journey, the expert is able to measure effects and test other scenarios by error and trial. Machine Learning and Data Science reverse this process – it focuses on analyzing the data of customers who converted and finds correlations between all their interactions leading to this conversion. It helps to recreate the most effective channels and processes which lead to the conversion – when these actions are identified, based on the reports the specialist is able to re-design the processes to make them more effective.
Data collected by SALESmanago
Data taken into the analysis:
- Website visits
- Products bought
- Products added to cart
- Conversion paths
- Conversion sources
- Buyers’ personal and demographic data
- Purchased products` attributes
- Reactions to direct marketing
- Search terms used
- Chat conversations
- Products displayed
- Cart value
- Offline behaviour
Expert Approach vs Machine Learning
Let’s bring Expert Approach and Machine Learning face to face. The two approaches differ significantly. As mentioned before, the expert approach is connected with the limited amount of the data and the source of knowledge comes from the expert’s experience, additionally being confined to the limits of the human imagination. Automation rules created by an expert are mostly designed to solve or react to a specific situation.
The advantage of the Machine Learning over the Expert Approach lies in the way it works. Machine Learning functions in a complex environment and takes into consideration many variables. The system is provided with the huge amount of data from which the algorithm needs to learn. In other words it is trained from an abundance of examples in a relatively short time, shorter than a human. While implementing particular solution, the system takes into consideration many factors which human mind may not be able to process or come up with as his or her knowledge is limited. Additionally, the database is constantly increasing and the Machine Learning technology is provided with new sets of data to be processed.
Copernicus AI & Machine Learning in SALESmanago
SALESmanago Copernicus – Machine Learning & AI is an advanced self-learning environment that analyzes the behavior of individual customers and predicts future purchases. Then it sends personalized product recommendations according to what the algorithm deems most likely to be bought. This cutting-edge Marketing Automation tool provides insight into customer purchase history, buyer’s journey, and analyzes the way products correlate in categories, allowing for highly engaging and eye-catching offers to be delivered to individual customers.
The technology of Copernicus is based on two recommendation models. Each is optimized to support a specific marketing approach. For inbound marketing – affinity analysis (or the so-called Inbound Predictive Marketing) and for outbound marketing – behavioral analysis (the so-called Predictive Outbound Channel). Used in tandem, the models enhance both inbound and outbound marketing activities.
The mechanism of affinity analysis relies on sophisticated algorithms used in association analysis. By thoroughly analyzing transaction data and correlations between specific products and in categories, they calculate the optimal combination of items in each offer. After the resulting data is parsed and modeled, a frame with product recommendations can be shown to each customer. In addition, the use of metadata makes it possible to instantly react to changes in customer preferences. The Marketing Automation system can employ machine learning to compare predictions from product association analysis for end customers on an ongoing basis. Then it assigns scoring to each given recommendation in order to indicate how likely that product is to be bought by individual customers. Moreover, by updating product exclusion grids, the algorithm ensures that products are not recommended to customers who already bought them.
Expert’s limitations resolved by Machine Learning
- Unique customer behaviours
- Number of possible customer segments
- Price sensitivity
- Unique customer preferences
- Advanced personalization
- Identification of uncommon actions leading to conversions
- Analysis of correlations between multiple variables in the same time
- Real time process adjustment to changes in customer behaviour
Stages of implementation
Before you take the full advantage of the AI & Machine Learning system, you need to go through several stages of implementation.You can use some algorithms which are ready to use or use the system in most advanced way – to create your own algorithms and processes which uses machine learning for very precise reasons. In that case, you should go through a process of designing such implementation:
- Determine objectives, metrics and constraints: This is a formative stage that will shape the way how the algorithms will work. You need to think about the objectives. One of the objectives can be assistance of the Machine Learning with servicing all B2B marketing campaigns created in Workflow. Then you need to think how you will measure the results. The selection of the metrics is of critical importance because choosing right metrics is the determinant of success. The whole model can fail, if the selected metrics is wrongly matched.
- Assessing data, data collection: it may not be obvious, but the data may not be gathered in one place. But even such scattered data as well as not pre-processed, is considered to be of great value and importance for the Machine Learning approach more than “clean data”. In the case of insufficiency of data or its lack on this stage, you need to determine which type of data will come in handy while solving the task. Once you get to know it, start collecting it. You need to specify how these data will be transferred to the MA platform, determine their format and how they will be matched in the system so it will be easy to process them for Machine Learning system.
- Model training: You need a consultation with a data analyst who will indicate various factors which may have impact on the model. This step is also very important as you need to check whether important elements have not got lost. As previously indicated, you might need the counsel of experts in the field who will work in tandem with the data analyst. The data analyst will also be responsible for training the model and will be equipped with the necessary tools. The duration of the training may vary as it depends on the intricacy of the model.
- Integration and testing: Once the model is trained It needs to be integrated into the management system. Then together with the expert, you should carry out the practical tests. While you are carrying out the tests, it is obligatory to check the model’s accuracy and the economic effect it brings.
- Model monitoring: In the last stage you will use the model that successfully passed testing. The model will require constant monitoring as well as extra training as the new data will be constantly gathered.
Ways to use Machine Learning in Marketing Automation
Segmentation is the process of partitioning database into groups. This process is not only connected with organizing customers into proper segments according to their preferences or gender, but it is also about organizing products into categories or users who have bought products from a certain category. Segmentation is used to find certain behavior pattern, similarities between products and examine correlation between groups. So later you can target your customers with better offers.
Some examples of segmentation encompass organizing database according to the similarity of the purchases during the year. The analysis can show us the dependency of the purchases between age and location. You may be also interested in the segmentation of the cart value and see how the age and the frequency of the shopping correlate with the location. Additionally, you can segment the products database on the basis of pictures. The Machine Learning system is able to convert the picture into a code and group products which are similar in order to recommend them later. If you would like to use more basic solutions, you may use regular segments such as gender and check the shopping value of men and women.
Sentiment analysis is the process of computationally identifying and categorizing opinions expressed in a piece of text, especially in order to determine whether the writer’s attitude towards a particular topic, product, etc. is positive, negative or neutral. So for instance you can research the opinion of Twitter users about Italian retail industry in New York.
You may use Machine Learning and AI to familiarise with the customers’ opinions about a product and thus you can take advantage of it by implementing the dynamic pricing. AI engine will calculate discounts based on probability of purchase thus maximising income across all customers. Sentiment analysis may prevent you from the customers’ churn as well. By analysing behavior and interactions of the users, the system indicates customers who are likely to be lost.
Natural Language Processing
Natural-language processing (also known as NLP) is an area of computer science and artificial intelligence concerned with the interactions between computers and human (natural) languages, in particular how to program computers to fruitfully process large amounts of natural language data.
AI engines are widely used with regard to natural language processing field. The most known fruit of the fusion is a chatbot that facilitates conversation between a person and a robot. They are widely used in the B2C enterprises, but also they are successfully transforming B2E and B2B sectors in their organizational aspects. According to the Oracle, 80% of businesses want chatbots by 2020. Moreover, Juniper Research forecasts that chatbots may cut business costs by $8 million by 2020. Who wouldn’t want now to make use of chatbots after such prognosis?
How such chatbots work? On the surface it looks fantastically simple. When you are not able to service your all customers and answer their queries, you don’t have to employ people because chatbot will talk to your customers and help them out with the most frequently asked questions. If you look closer, the chatbot uses highly advanced technology and bot learns the most typical scenarios of conversations. That works miracles with reducing the costs.
Unlike product recommendation engines available in most Marketing Automation platforms, AI recommendations are not based on the product data itself or adjusted to specific user’s in 1-to-1 model. By the analysis of the data about all users, you do not need to apply 1-to-1 model. The recommendation prepared by the Machine Learning system can include new users and the inactive users to which you are not able to prepare 1-to-1 recommendation model. It is possible because the system uses information about similar users.
Electronic commerce and cloud computing giant that we know as Amazon.com generates sky-high revenues and 35% of them are said to come from product recommendations.
The goal of product recommendation is to increase average order value and number of transactions. Ordinary product recommendation includes sending offers with the products a user has recently viewed or bought or the products which the user has left in the cart. Product recommendations powered with AI engine facilitate personalization of product offers for all users, no matter how much do we know about them. The system learns the behavior of each contact, learns the conversion paths and analyses other factors which may influence the purchase decision. Then makes calculations and chooses the optimal offer for the contact. Sounds like magic, but it’s real.
AI Recommendations in SALESmanago Copernicus
AI recommendations types:
- Collaborative Filtering (users and products)
- Most frequently bought after visit other
- Most frequently visited together
- Most frequently bought together
- Mixed statistics with weight
Collaborative Filtering (users and products)
There are five types of AI recommendations. The first one is called Collaborative filtering and it involves two approaches. The first one called Product-Product is connected with probability and frequency of co-occurrence of different products (not necessarily similar to each other). The second approach is called User-Product approach and it shows which products may interest a user based on the interests of other users who have similar profile to the chosen one. Generally speaking, the idea behind this type of AI recommendation is to offer products based on the similarity of users and concurrence of various products.
Most frequently bought after visit other
The second type of the AI recommendation is most frequently bought after visit other. Based on what product the customer is currently displaying on the website, the system analyzes purchases of other customers who also displayed this product and recommends the products purchased by the others to the user.
Most frequently visited together
The fourth type of AI recommendation is most frequently visited together. As the name suggests these are the products that are often viewed together by all users. The system offers products which were browsed by other users along with these products. Such type of recommendation may contribute positively to the customer experience as you can provide the users with products which were browsed together and thus save their time and energy on searching for the one and only product.
Most frequently bought together
The third type of the AI recommendation is most frequently bought together. The name of the recommendation type speaks volumes. The system analyzes the products the customer has purchased. And also the system analyzes the products which have been purchased by other users along with the same products. The user can encounter this type of recommendation while buying a product, he or she may be offered “similar products which other users bought”.
Mixed statistics with weight
The fifth and at the same time the last type of AI recommendation. The mechanism behind this recommendation type employs all previously enlisted types of recommendations and additionally assigns weight for each action. The value of the weight can be determined by you. How does it translate into practice? The system creates connections and analyses products bought by the contact, recommending in the first place several products which are probable to be bought, then products which the user wants to see and so on and so forth prioritizing the rest of the products with regard to the actions.
The channels of recommendation delivery
Now you may wonder what are the channels in which you can make use of AI and Machine Learning. You may have a quite nice scope to work as every channel available in SALESmanago supports delivering recommendations
- Website: Display dynamic content on your website, may it be a pop-up, iframe, product frame or something else in which you can present a product, then the AI-based recommendations may boost the click rate of your forms on the website.
- Web Push notifications: This is the quickest method of communicating both with the anonymous and monitored contacts. All stagers of the online marketing are familiar with the concept of a small notification showing in the corner of the screen. Empower the dynamic content by AI recommendations.
- Email marketing: You may use AI and Machine Learning technologies to adjust the email scenario to the customers’ preferences.
- Social media: With AI and Machine Learning you do not have to worry any longer if you have enough people to answer the customers’ queries
- Ad networks: Once you integrate with ad networks, you can display product recommendations outside your website.
The new face of Marketing Automation
Without doubts we are entering the new era of Marketing Automation. The SALESmanago platform has become a place for omnichannel data collection in which data-driven processes are adjusted in real time. Because of that, the omnichannel communication is going to be even more precise and concise, as each customer will receive consistent communicates on every channel. What is more, the system is going to be self-improving in Machine Learning thus increasing the effectiveness of product recommendations. Additionally, it has an impact on the skill set and role of a marketer as the stress is put more on the data than marketing itself. As the data science is starting to play the first fiddle now, some of experts’ responsibilities will be taken over by the Machine Learning. Marketing is liable to be overridden by the Machine Learning technology, so the expert will need to know more about servicing this technology. The analytical and data-processing skills are going to become an asset. Marketing will stop oscillating between the content and creation and focus more on delivering AI recommendations. However, the experience the expert have gained throughout the years will not be lost as they can be used while creating the data models for the purposes of Machine Learning work. Last of all, here is a handful of statistics to ponder about:
- 57% of executives believe that the most significant growth benefit of AI and Machine Learning will be improving customer experience and support.
- 44% believe that AI and Machine Learning will provide ability to improve on existing products and services.
- 58% of enterprises are tackling the most challenging marketing problems with AI and machine learning first, prioritizing the customer care and new product development.