Speaker 1: Hello, and welcome to our video blog. We proceed with data science. Earlier, we discussed NLP, or natural language processing, and the focus of our today's discussion is customer behavior in data science. I invited Elisaveta Kurila, a data scientist and a leading specialist in customer behavior technologies at IBA Group, to share her experience and ideas. And Mark Hillary, a writer and blogger on technology, will help her share her ideas. And I'm handing over to Mark to introduce the questions.

Speaker 2: Thank you, Marina. It's great to be here, and this is a subject that I find really interesting, you know, because I've used tools like Amazon for so many years, and they always seem to predict exactly what I want. What I was going to ask you first was about data science companies. They often claim that they can predict how customers will behave. How accurate are these claims, and does it work maybe just for specific businesses, like your mobile phone subscription, for example?

Speaker 3: Hello, thanks for inviting me to take part in this important video blog. So returning to the question, actually predicting their consumer behavior is one of the biggest challenges faced by marketers all over the world. And today, this task is even harder because consumers are constantly being exposed to new technologies, products, and even new ones. And of course, advertising campaigns have a very huge impact on the consumer behavior. But fortunately, this is exactly where data science comes into play. And nowadays, using data, marketers with the help of data scientists can find the answers to the questions such as, what is our target audience, or how effective is our advertising, and a lot of other questions. And the answers to these questions can help marketers efficiently predict consumer behavior, build marketing plans, and as a result, maximize company's ROI. And what is especially great, it doesn't matter what the specifics of the business.

Speaker 2: That's interesting. I mean, what about predicting behavior before the customer actually does something? So for example, like, if I regularly buy food for my dog from Amazon, could Amazon possibly send me products before I've even ordered them? Because they predict that I need something.

Speaker 3: Absolutely. And these predictions are connected with personalized data. You know, recently, personalization has become more and more popular in e-commerce. And for example, for a long time, marketing has used many different channels to contact customers. And these channels were not linked to a single system. But today, this approach has been replaced by Omnichannel. Omnichannel removes all boundaries between marketing channels and creates a single, integrated, fully connected system. And thanks to these systems, it became very easy to collect and store data. And now marketers are fully informed about the customer's journey to purchase. And as a result, they can make more customer-centric business decisions. So we, data scientists, can build a predictive model based on personalized historical data. And this model will determine what the customer needs at the right moment and offer him or her to make purchases just, for example, with the notification. And the customer only needs to confirm the predicted order.

Speaker 2: That's all. Okay. Okay. I mean, how much data or insight do you need then to be able to build the model you were just describing? I mean, it seems like you must need a long history of behavioral data to be able to predict what somebody is going to do next.

Speaker 3: You know, it's common knowledge that the more data you have, the more tasks you can solve. And of course, the amount of data depends on goals. But in general, data over a one-year period is enough to get quite good results. But it's worth remembering that the more historical data you have, the more accurate the results. So if we talk about what data we need to make some predictions, so discounts, promotions, advertisement budgets can help to plan the best marketing tactics and, for example, the best channel budget contributions. Tasks directly related to predicting consumer behavior can be solved using purchase data, product ratings, data on retail transactions, and so on. And in this question, I would like to talk about one of the problems that marketers can face. It's a cold start. A cold start is when a new client comes in and there is no information about him at all. So to solve this task, we need to know or to predict some information about the new customer. It may be, for example, gender or customer location and the need to predict a base recommendation list. So, you know, according to research, 68% of new customers are not profitable at all. And this task is very important. And any improvement in solving this task is very valuable.

Speaker 2: And then, I mean, if you go from a cold start, you can teach the system from real customer behavior. But can the system, the algorithm learn as well using machine learning? And if so, then how do you prevent bias or bad data getting into the system?

Speaker 3: I can say that machine learning combined with historical data analysis and some experience can solve a really huge variety of problems. And there are a lot of models which we can use to solve different tasks. For example, classification model. Classification model is best suited for answering yes or no questions. For example, they may answer the question whether the client will leave. The next one is the clustering model. This model can split all customers into similar groups based on common features. For example, age, behavior, place of living, and so on. And as a result, segmentation allows to apply marketing strategies not only for one customer, but for all group at once. And it significantly saves the company time and money. The next one is the predictive model. This model can estimate, for example, the number of customers per week or, for instance, calculate their required amount of stock in their workhouse. And if we talk about invalid data, data filtering and applier models can help clean the data and they are used to solve some specific tasks, for example, detecting anomalous data in transactions.

Speaker 2: I mean, it's really exciting, but I think that it also sounds very complex. You know, if I'm an executive working in marketing and thinking about how does this practically work for my business, what would you say? I mean, what are the sort of practical examples of solutions you can offer?

Speaker 3: Let's think about what all companies want. All companies, regardless of their orientation, want to first optimally plan their budget, second, increase key performance and third, better understand their business. And these tasks are solved differently for each type of business. For example, for a retailer, the typical questions are, is this customer about to churn? Or, for example, how much and what products need to be delivered in the near future? Or which advertising channel can attract the desired segment of customers to store? And many other questions. Then the loan provider is interested in questions such as, for example, will this loan be approved or is this applicant likely to be default? And in conclusion, I want to say that optimal budget planning for marketing tactics, customer retention, the ability to work with the client from start, identification of anomalies and trends are the tasks that will help any business reach a new level.

Speaker 2: That's great. Thanks.

Speaker 1: Thank you, Mark. Thank you, Elizaveta. Now we can see that data science can do a lot for businesses and for customers, maybe too. Thanks, everyone, for watching this video. If you're looking for more information on data science, please contact us and stay tuned to our video blogs.