Редакція dev.ua Special projects 24 September 2026, 11:27

Billions of data points and dozens of models: how Data Science works at Philip Morris Ukraine

The architecture of the labor market is changing: the line between classic IT and other fields is becoming less and less visible. In 2026, the largest categories of vacancies in the Ukrainian IT market will be analytics, marketing, HR, Sales and QA, while AI/ML, Data Science and Data Engineering remain among the areas with the highest level of employment of specialists. At the same time, 30% of Ukrainian IT professionals would be ready to change their field if they were offered better conditions.

So working with big data today does not necessarily mean a classic IT career. Analytics, machine learning, and AI are increasingly penetrating marketing, manufacturing, customer experience, and business operations. According to DOU, AI/ML/DS/Data Engineering specialists in 2026 are among the groups to whom recruiters write most often: 65% of Ukrainian specialists in these areas receive such messages several times a month.

In the public mind, Philip Morris is a conservative tobacco giant with paper packs, distribution trucks, and classic marketing from the last century. But behind the corporate facade, the company has long lived by the laws of the technology business: they work with billions of data points, build ML models, and gradually integrate generative AI into business processes.

All product and marketing hypotheses at Philip Morris are based on a foundation maintained by a team led by Dmitry Sitenko, senior manager of customer experience management. In a partner article, he told dev.ua how data shapes the company’s agenda and helps solve customer pain points.

Billions of data points and orchestras of models: what’s under the hood of Data Science

The focus on data here is not about fashion, but about necessity: the market and consumer behavior are changing rapidly, and decisions based on intuition are too expensive at this scale of business.

“We work with a full range of data. On the one hand, internal business data: from production, distribution and planning to product deliveries to the market and actual sales to the end consumer. On the other hand, consumer-centric data: demographics, geography, behavior within the IQOS ecosystem, interaction with all brand channels and programs,” says Sitenko.

According to him, the total is about billions of data points across hundreds of parameters.

A typical data science project at Philip Morris always starts not with a model, but with a business question.

"Any tool or project that begins to be developed in our department starts with defining the business need — what problem needs to be solved and what is the value of this solution. Simply put, we start with two questions: 'why do we need this' and 'what does it give us'," says Dmytro.

Next, the team analyzes the available data, forms hypotheses, and chooses an approach — from classic analytics and segmentation to machine learning or generative AI, tests solutions in limited production tests, and only then integrates them into the company’s processes.

"Data Science for us is not about models for the sake of models. A successful project is one that helps make better decisions, automate processes, or improve the customer experience — even if the solution "under the hood" is completely simple from a technological point of view."

In parallel, for over five years, the company has been developing the field of machine learning — today it has dozens of models and their "orchestras" around users: forecasting, segmentation, clustering, recommendation systems.

And since last year, generative AI has been actively integrated — both for internal needs: onboarding agents, generation of template documents, audio transcription, preparation of presentations, drafts of creative materials, analytics of large data sets, and directly for the consumer experience, for example, automated generation of a shopping cart from natural language or preparation of personalized communication.

How algorithms determine who needs help

A clear example of how data and machine learning work in PMU in practice is working with user churn.

After a company attracts a new consumer, it is important that they remain active in the brand ecosystem, and this cannot be ensured manually for everyone, as different people have different needs, products, and communication channels, and resources and budgets are limited.

When a person switches to smoke-free products, it is critical to support them at the start. But users are different, and the resources of live operators and marketing budgets are limited.

Instead of carpet bombing with pushes and calls, the team ran a set of predictive machine learning models.

"Using machine learning methods, we identify individual users who need our help, consultation, or problem solving, choose the appropriate channel and time of communication, and exclude those users who do not want to receive it. As a result, we work more precisely, spend resources where they are effective, and increase user satisfaction," says Dmytro.

In this way, algorithms identify customers who are experiencing difficulties or exhibiting patterns of rapid churn from the ecosystem in advance. The system itself determines who needs specialist advice, and selects the relevant channel and optimal timing.

And most importantly: the model automatically excludes from the newsletters those users who do not want to receive messages, so as not to generate negativity. The result is a win-win: budgets are not wasted, and customer retention increases.

Banned ChatGPT and Corporate AI

The issue of security in a corporation of this caliber is uncompromising. On work laptops, third-party neural networks are blocked by cybersecurity policies.

The only fully legitimized tool is the corporate Microsoft Copilot, locked inside the company’s local loop and connected to the internal knowledge base. It helps write texts, analyze documents, and prepare drafts of presentations and visualizations.

Dmytro specifically emphasizes the need for human control. Despite the fact that a huge number of processes are already AI-powered, the company does not consider AI as a complete replacement for human decisions. He formulates this as the human-in/on-the-loop principle: human control remains an obligatory element of working with AI.

AI at Philip Morris and prospects for specialists

Philip Morris' transformation into a knowledge-intensive company once started with the development of alternatives to cigarettes, which required fundamental research. Now this approach permeates all business processes.

This also changes the requirements for people working in a team. Dmytro believes that today, for a data scientist, it is more important not to have perfect knowledge of mathematics, SQL, or statistics, but to be able to correctly formulate a business problem, understand the value of the solution, and explain the result in a way that other teams can use:

"Artificial intelligence makes it relatively easy to compensate for the lack of deep technical expertise or hard skills, and the correct translation of business problems is increasingly becoming valued."

As for the future, the company has a strategy for developing AI until 2029 — but, according to Dmitry, it is unlikely that it will be possible to follow it step by step: the field is so dynamic that the strategy will probably have to be revised every six months. The team has already completed the stage of researching basic technologies — the first platforms, agents, working with knowledge bases, administration and governance.

The focus is then on developing reusable tools, covering entire workflows from start to finish in an autonomous manner, integrating with other corporate systems, and making decisions in real time.

The company also has a separate, yet cautious plan: to turn AI into a full-fledged “second brain” for analysts and data scientists. According to Dmitry, this is in the pipeline until the middle of next year. Today, he says, it is “a very powerful work assistant” that has already allowed teams to significantly reduce the time spent on routine tasks — and this, rather than replacing people, remains the main effect of artificial intelligence within Philip Morris.

A separate challenge for this direction is attracting people with the necessary technological expertise. The PMI HR team recognizes that today the company has to compete for specialists with technology startups, the IT sector, and international product companies.

At the same time, the company considers its advantage to be the ability to combine the stability of a global corporation with the dynamics of the technological environment. It is not only about working with data or AI, but also about the ability to see how the solutions created affect real business and can be scaled internationally.

Another emphasis is on development within the company. PMI encourages internal mobility, opens up access to international expertise, and provides the opportunity to expand competencies and move between areas.

Philip Morris invests in people development, supports internal mobility and creates a comprehensive ecosystem of well-being support that covers the physical, emotional, financial and social well-being of employees. Combined with access to international expertise and career development opportunities, this allows specialists not only to work on ambitious projects, but also to build a long-term career in a global environment. One of the company’s advantages is also the ability to combine the technological complexity of tasks with the scale of international business. Specialists work with big data, modern AI solutions and global platforms, collaborate with teams around the world and at the same time have access to extensive development opportunities and a holistic well-being ecosystem.

For digital professionals, this means the opportunity to work not only within a narrow technological specialization, but also to experiment at the intersection of data, AI, and real business. Here, you can test different approaches — from classic analytics and ML to generative AI, agents, and automation of entire processes — and see how these solutions work not in a test environment, but on a large array of real business data.

Thus, for the Data Science team, it is not just work on individual models, but the opportunity to gradually expand their own zone of influence — from an analytical task to a solution that works at the level of a large international company.