1 of 4
The tobacco industry is increasingly embracing AI in different areas. Photo credit: Tara Winstead, Pexels.
2 of 4
Diego Paccagnan, c.e.o., Aleph Digital Industry.
3 of 4
Dr. Christian Paleani, chief AI scientist and co-founder, Vernaio.
4 of 4
Henry Mozon, c.e.o. and managing director, Vernaio.
Artificial intelligence has the potential to revolutionize the way the tobacco industry operates.
Following the more visible shift sparked by the introduction of reduced-risk products, the tobacco industry is undergoing a quiet yet profound transformation. As companies embrace alternatives like heated tobacco and e-cigarettes, they are now turning to artificial intelligence (AI) to optimize production. Soon, they will also use it to enhance product innovation, refine marketing strategies, and ensure compliance with ever-evolving regulations.
In late October 2023, Vancouver, Canada-based Generative AI Solutions Corp. (GenAI) announced the development of a functional beta version of Tobacco Titan, an AI-driven large language model tailored for the tobacco industry. Tobacco Titan aims to provide AI-powered information about products, marketing, and regulations, along with health and safety insights. The product could also offer customized recommendations, such as new flavors, brands, or smoking accessories that align with the adult user’s tastes and preferences. As of February 2025, the product had not yet entered the market.
From smart manufacturing that reduces waste and improves product consistency to AI-driven analytics that shape targeted marketing strategies, machine learning and data-driven decision-making are about to reshape the industry’s landscape once again. AI is already making a significant impact, influencing production efficiency and marketing innovation.
However, there isn’t just one AI. Artificial intelligence is a broad field with a vast range of approaches and applications, each at different stages in its lifecycle. To assess these stages, American research, advisory, and information technology firm Gartner has developed the so-called hype cycles. These provide a graphical representation of the maturity and adoption of technologies and applications, illustrating their potential relevance in solving real business problems and seizing new opportunities.
In its most recent AI edition, the Hype Cycle for Artificial Intelligence in 2024, Gartner identified causal AI—the underlying technology of the solutions implemented in the tobacco industry—as a key emerging technology. Causal AI, one might say, is the smarter sister of predictive AI. Prediction-based AI models rely on correlations, forecasting outcomes based on past patterns but offering limited causal insight. They cannot answer ‘why’ and ‘what if…’ questions. By contrast, structural causal models (SCMs) are causation driven. They model interdependencies and causal pathways, identify underlying causes, and enable prediction, intervention, and decision-making. SCMs allow for real-time process control, complex problem-solving, risk mitigation, and root cause analysis, while supporting proactive adjustments (‘what would happen if…’).
Adaptive production speed
Philip Morris International (PMI) has been at the forefront of integrating artificial intelligence into its production processes, pioneering advancements at its Bologna heated tobacco product (HTP) manufacturing site. The company faced challenges with productivity falling short of targets and operators handling high workloads. A key focus was optimizing the efficiency of material buffers in high-speed production lines—these buffers hold materials from upstream machines and feed them to downstream machines to prevent production stoppages.
In recent years, PMI reduced the size of these buffers, making supply depletion dependent on machine speed. Traditionally, this speed was managed reactively, with operators adjusting settings based on buffer filling levels. Seeking a more intelligent solution, PMI initiated an improvement program for its legacy HTP production lines.
Diego Paccagnan, c.e.o. of Aleph Digital Industry, an Italy-based smart manufacturing solutions provider, recalls, “PMI required Aleph to develop an application that autonomously adjusts machine speed in real time, utilizing data beyond what operators can access. This allows operators to focus on higher-value tasks instead of continuously monitoring the machines.”
The causal AI-driven speed management controller (SMC) functions similarly to adaptive cruise control in automobiles. It ensures machines operate at an optimal pace, slowing down when material accumulates in down-stream buffers and speeding up when buffers are running low. Moreover, SMC learns from past unplanned stops and adjusts speed accordingly for future production cycles.
“The implementation of SMC led to substantial improvements, prompting a global rollout,” Paccagnan explains. “The mean time between failures [MTBF] increased significantly due to reduced downtime caused by material shortages or blockages. Furthermore, production output improved since SMC maintains machine speeds closer to optimal targets than manual operator adjustments.”
Another advantage of SMC is its flexibility—operators can configure it to prioritize either MTBF or production efficiency, akin to choosing between a smooth or sportier driving style. These benefits were realized despite implementation challenges, particularly due to the Covid-19 pandemic, which necessitated remote work. However, the most significant hurdle was ensuring stakeholder trust in AI-driven decision-making.
Since machine speed directly impacts key performance indicators (KPIs) used to evaluate both operators and managers, gaining operational buy-in was crucial. To address this, PMI involved its operations teams from the project’s inception. A Digital Twin developed by Aleph enabled side-by-side comparisons of actual operator-driven performance versus SMC-driven scenarios, demonstrating the AI’s reliability.
“By visualizing how SMC would have handled production, the operations team gained confidence in the AI’s decision-making capabilities,” says Paccagnan.
Minimal inputs required
SMC is designed to function even in data-limited environments, making it widely applicable across different manufacturing settings. It requires only minimal inputs, such as production counters, reject counters, and speed set points. In cases where buffer level data is unavailable, SMC estimates it by analyzing production differences between connected machines.
“This adaptability makes SMC suitable not just for HTP production but also for conventional tobacco manufacturing and filter production,” Paccagnan notes.
To facilitate seamless integration, Aleph developed TMC OPC UA servers, secure communication platforms based on open OPC UA technology. These servers expose the entire tobacco production model, including critical data points such as production order numbers and recipe parameters.
Importantly, they support both reading and writing data, enabling full interoperability across manufacturing systems.
The Tobacco Machine Communication (TMC) standard, authored by Aleph, was established by the TMC Working Group, which includes PMI, Japan Tobacco International, British American Tobacco, and Imperial Tobacco Group. This industry-wide standard ensures compatibility between various manufacturers and enables smoother AI integration into existing production infrastructures.
Given the vast amount of machine data generated in modern production environments, a strategic data management approach was essential. Aleph assisted PMI in implementing a two-layer data strategy: locally, all machine data is exposed via OPC UA, while only a selected subset is transmitted to PMI’s central data stack using MQTT, a lightweight messaging protocol for the Internet of Things (IoT). “Aleph developed an OPC UA to MQTT gateway that allows selective data publishing, ensuring only relevant information is transmitted for storage and analysis,” Paccagnan explains. This system has since been deployed globally, streamlining PMI’s data infrastructure.
Beyond SMC, Aleph has spearheaded a broader digitalization initiative for legacy equipment, addressing the challenges of integrating AI into older production lines. Additional AI-driven solutions developed by the company include advanced control systems for stick diameter regulation and crimping optimization in HTP manufacturing. “The principle behind these innovations is similar to SMC,” Paccagnan says. “We enhance existing machine operations by leveraging AI capabilities that were not available when the equipment was originally built.”
Causal AI for the primary
In the primary of HTP production, where raw materials coming with natural variations are being processed, challenges differ, as Henry Monzon, c.e.o. and managing director of Vernaio, a Munich, Germany-based company specializing in AI solutions for manufacturing and production processes, points out. Meeting precision requirements in the production of the tobacco cast sheet used in the HTP consumable was the top priority for his customer, a market leader in the production of HTPs. Throughout the coating and drying process, strict quality metrics must be maintained accurately with minimal deviation. Key factors include moisture content, thickness, and grammage.
“Any deviation will result in significant material waste, increased energy consumption, and reduced final product yields,” says Monzon. “The combination of these factors creates an environment where traditional control methods struggle to keep up.”
To address these complex challenges, Vernaio implemented a broad AI-powered solution, which encompassed the development of a comprehensive target KPI combining deviations of moisture and grammage profiles and the setting of precise limits on this target KPI to ensure both deviations remain within allowed ranges. The company also conducted a thorough root cause analysis to identify all key parameters affecting the target KPI and uncovered hidden patterns and correlations in the production process. Vernaio achieved this by feeding raw historical process data into its causal AI engine, aivis, which powers its process control solution, Process Booster X.
The model Vernaio implemented at their customer’s site can predict KPI behavior and facilitate proactive corrective action. “Our approach is to directly determine interdependencies between variables by leveraging time invariance. Our scalable causal AI is highly autonomous, automatically identifies causal relationships, and efficiently models non-linear interactions.”
Unlike the commonly applied silo approach, which addresses issues one by one, Vernaio’s unified master causal AI takes all the input about issues and the recommendations that come back, processes them, and delivers holistic models for end-to-end production process optimization, addressing all process KPIs. With its real-time intervention capabilities, the AI solution then provides real-time countermeasures, making it a super tool or AI co-pilot for operators, the company says.
Making machines smart
“Vernaio delivers the brains of the machine,” explains Dr. Christian Paleani, Vernaio’s chief AI scientist and co-founder. “Our solution is very flexible regarding the input data. Our client only needs to have data and internet access. You just have to provide our AI with goals, such as ‘I want to minimize waste or my CO2 footprint’ or ‘I want to maximize profit,’ and you will get an intelligent machine by definition and achieve optimal outcomes even with conflicting production priorities. Our solution is learning continuously – there is no catastrophic forgetting, i.e., overwriting of older information.”
Production implementation into the customer’s IT and operational technologies (OT) is easy and fast, Monzon relates. “The evaluation or proof of concept can be done within days. Once that is completed, Vernaio enters into a licensing agreement on an annual basis to scale its IT solution up to multiple production lines. Vernaio is the only company in the world offering a causal AI-powered solution with real-time intervention capabilities, which optimizes all KPIs simultaneously. Our solution applies to industry sectors beyond production and, according to Gartner, gives us a two- to five-year market advantage.”
For Vernaio’s HTP client, the implementation of the causal AI solution meant that the aivis model has been able to predict the target KPI with a high accuracy of 91%. It also projected a waste reduction of 50-75%, from 4% to just 1%, which translated into substantial cost savings and improved resource efficiency. Operators have gained the ability to make timely, informed decisions based on AI recommendations, an approach that prevents quality issues before they can impact production.
“The continuous learning nature of the AI system means it will keep improving over time, adapting to process changes and new challenges,” Monzon concludes. “This ensures the solution remains effective and relevant as the production environment evolves.”