Part II: Physical AI Adoption by Industry

Chapter 7: Pharma, Cosmetics, Food, Apparel — Physical AI in Manual-Labor Industries

Written: 2026-04-28 Last updated: 2026-04-28

7.1 Why Manual-Labor Industries Differ from the Automation Frontier

The previous two chapters covered Physical AI's "successful industries" — automotive and logistics. Their lines are standardized, ROI per unit task is clear, and a well-designed automation cell repeats the same job for years. The industries this chapter examines — pharmaceuticals, cosmetics, food, and apparel — are also manufacturing, but the texture of automation is fundamentally different. Even within a single factory, some lines resemble a 1990s Japanese auto plant while others look little different from a sewing workshop a century ago.

Four factors set these four industries apart from automotive and logistics.

First, material variability. Cars cut steel panels. Logistics moves corrugated boxes. Both have well-defined shape and stiffness. Cosmetics handle dozens of bases with different viscosities, food processing handles chicken legs and pickles and noodles whose shape varies every time, and apparel handles fabric that loses its form the moment you grasp it. Deformable material manipulation remains an unsolved problem in robotics research. At the ICRA 2024 deformable manipulation workshop, eleven teams competed on cloth grasping, unfolding, and folding — yet none could simultaneously satisfy real factory speed and accuracy [11].

Second, SKU diversity and small-batch production. A cosmetics ODM handles hundreds to thousands of SKUs on a single line. Food processing changes recipes by season, region, and campaign. Apparel issues fresh designs every season. This is unlike running one car model on a line for twenty years. Every SKU change requires the automation system to be reconfigured, and this reconfiguration cost creates a zone where "automation is technically possible but the ROI doesn't pencil out."

Third, hygiene, cleanliness, and regulation. Pharma needs GMP, food needs HACCP, cosmetics need ISO 22716. Every surface must be food-grade stainless steel, periodic cleaning and sanitization is mandated. A robot in a car factory can leak grease; a robot on a food line that does so condemns the entire line. These constraints double or triple robot prices and cut applicable gripper and sensor types by half [29].

Fourth, the tacit knowledge of skilled labor. A seamstress lifts the fabric slightly with one hand while positioning the stitch line with the other; a cosmetics worker spots a label tilted by half a degree at a glance; a pickle cutter adjusts the knife angle to the ingredient. These motions are not in any operations manual — and even if they were, they take months to learn. This tacit knowledge zone occupies the final 35% of automation.

The result of these four differences is simple: the same phrase "AI manufacturing transformation" describes fundamentally different work in these industries than in automotive and logistics. PR materials prefer to blur this difference, but materials Cosmax executives use to make decisions must name it precisely.

7.2 Pharma and Chemicals — Research Automation Leads the Way

Of the four industries, pharma and chemicals are most advanced. Interestingly, the leading edge sits not on production lines but in research labs.

Sanofi Modulus — Modular Bioproduction Borrowed from Automotive

Sanofi's Modulus, unveiled in 2025, rewrites the manufacturing paradigm of pharmaceuticals [Sanofi, 2025a]. The core idea is to transplant the modular production approach from the automotive industry into biologics. A single facility can produce up to four products simultaneously, and product-changeover time shrinks from months to days. Sanofi invested approximately S$800M (€558M) in the Singapore facility, with €1B+ annually flowing into modernization across the manufacturing network. Modulus was named to TIME Magazine's 2025 Best Inventions in healthcare.

Modulus matters not because of the efficiency gain alone. Biologics historically tied one facility to one product — needing a new vaccine meant building a new factory. The plug-and-produce design breaks that assumption and lets production infrastructure adapt in an era of pandemics, emerging diseases, and personalized therapies where demand shifts quickly.

Sanofi SimplY AI and Five Digital Lighthouses

If Modulus is hardware innovation, Sanofi SimplY AI is the software and digital twin counterpart [Sanofi, 2025b]. The Lyon Digital Manufacturing & Supply Accelerator (central hub) and five WEF Digital Lighthouse sites (Geel in Belgium, Waterford in Ireland, Scoppito in Italy, Toronto in Canada, Hangzhou in China) operate GenAI-driven batch yield optimization, predictive maintenance, and real-time quality control. After GenAI deployment, the development timeline shortened by 25%.

AstraZeneca Wuxi — The Lighthouse Benchmark

AstraZeneca's Wuxi plant in China is the most-cited pharmaceutical AI manufacturing case in the WEF Global Lighthouse Network [4]. The quantitative effects:

  • Output +55%
  • Lead time -44%
  • Defective batches -80%
  • Productivity +54%
  • iLab initiative: 300,000 compounds screened per day

What is striking is that these are compound effects, not single-axis ones. Output usually trades off against quality, but Wuxi improved both simultaneously. Digital twin plus AI batch analytics produced this result.

Pfizer PAXLOVID — Cycle Time -67%

Pfizer, through its NVIDIA and DISG partnership announced in October 2024, cut PAXLOVID production cycle time by 67% [21]. This figure shows the promise of digital twins plus AI process optimization, but its ceiling is also clear — the result is bounded to a small-molecule drug, and complex biologics need further validation.

Novo Nordisk + OpenAI — Enterprise-Wide Integration

Novo Nordisk has signed a strategic partnership with OpenAI to integrate AI across R&D, manufacturing, supply chain, and clinical trials [19]. Full integration is targeted for 2026, with a parallel $9B supply-chain investment. Surging semaglutide demand (Wegovy and Ozempic) is the core driver. This is one of the first cases of an LLM company and a pharmaceutical company entering an enterprise-wide partnership rather than a single-department engagement.

BASF AI Reactor — From 18 Months to 3 Weeks

In chemicals, BASF AI Reactor reports the most dramatic numbers [5]. By applying machine learning to chemical reaction optimization, catalyst discovery, and process control, BASF compressed research duration from 18 months to 3 weeks and accelerated experimental throughput 20-fold. With €2B in annual R&D investment, the company is implementing self-driving lab concepts at industrial scale.

Self-Driving Laboratories — Automating the Scientific Method

All these cases converge on a single paradigm: the Self-Driving Laboratory (SDL). A 2025 systematic review in Nature Reviews documents how SDL automates the entire scientific method — hypothesis generation, experimental design, execution, analysis, and literature update [1]. A 10x acceleration in data collection and dramatic reduction in experimental cycle time has been demonstrated in materials science, drug discovery, and chemical optimization. Crucially, robotic platforms combined with LLMs now automate even hypothesis generation.

Why Research Automation Comes Before Production Automation

A natural question follows: why is the research lab automated before the factory in pharma and chemicals? Three reasons.

First, repetition density. Chemical reaction screening repeats the same task tens of thousands of times — automation's best friend. Pharmaceutical production, by contrast, makes one SKU at large scale, and changing the SKU requires re-validating the entire line. In a GMP environment, that re-validation takes months to years, so automation ROI is paid back far faster in the lab than on the line.

Second, regulatory friction. AI freely recommends candidate molecules in discovery. But replacing even a single step on a GMP-certified production line with AI requires re-certification by FDA or EMA. In many places, regulatory cost overwhelms automation cost.

Third, data assets. Each lab experiment generates novel data; production lines generate the same data repeatedly. There is overwhelmingly more fresh, training-relevant data on the research side. The Pharmaceutical Laboratory Automation market is projected to grow from $5.97B (2024) to $9.01B (2030) at a CAGR of about 7.1%, with AI-driven drug discovery timelines compressing 30–40% as the main growth driver [9].

Dow's Different Path — Cost-Reduction-Centered Transformation

Within the same chemical industry, Dow's path runs differently. Dow's "Transform to Outperform" strategy combines AI and automation deployment with workforce restructuring [8]. In early 2024, Dow announced a -13% headcount cut (4,500 employees) and a target of +$2B EBITDA by 2028 through AI-driven process optimization. The same technology accelerates R&D in one company and contracts headcount in another — a fact to keep in mind whenever Physical AI adoption is being assessed.

7.3 Cosmetics and Beauty — The Gap Between PR and Reality

This section is most directly relevant to a Cosmax executive office. The cosmetics industry produces the loudest AI PR, and the gap between that PR and actual process automation is also the widest.

L'Oréal Operations 4.0 — What Counts as Real Automation

As the global leader in cosmetics, L'Oréal puts out the most AI marketing material. The Operations 4.0 strategy integrates AI-based formulation development, GenAI training, and logistics automation [L'Oréal, 2025a]. Headline numbers:

  • Prototype development in 24 hours
  • 42,000 employees trained in GenAI
  • IBM-partnered sustainable AI formulation platform

It is critical to separate which of these is "Physical AI automation" and which is "digital transformation." Prototype in 24 hours is in-silico formulation simulation plus AI recommendation — not the speed of a robotic filling line. 42,000 GenAI-trained employees is office-AI tool training — not factory operators handling collaborative robots. Both are valuable, but they are a different category of work from substantive process automation.

L'Oréal SMART Fulfillment Center Suzhou — What Real Automation Looks Like

Among L'Oréal's cases, the one closest to real Physical AI is the Suzhou SMART Fulfillment Center [L'Oréal, 2025b]. The 46,000㎡ facility, opened in May 2024, runs 39 ACRs (Automated Crane Robots) and is one of the largest beauty automation logistics facilities in Asia-Pacific. This is not "digital transformation" but the cosmetics application of logistics automation covered in Chapter 6. Put another way, the most aggressively automated zone in the cosmetics industry is not the factory floor but the warehouse.

P&G AI Factory — The Lights-Out Four Hours in Berlin

P&G goes one step further. The AI Factory enterprise initiative spans 80+ global sites, has accelerated AI deployment by 10x, and is backed by a $1.1B ICT investment [20]. The most discussed milestone is the four-hour lights-out night shift at the Berlin plant (started July 1, 2024).

This case must be read precisely. It does not mean "the Berlin plant runs unattended" — it means for four hours of the day, on specific product lines, unattended operation became possible. It worked because detergent and household-goods lines have a high degree of standardization, and there is no guarantee the same approach transplants to a multi-SKU cosmetics ODM line. Even so, four hours of lights-out is meaningful: shrinking the night shift from 1.0 to 0.7 already shifts labor cost, safety incidents, and union management substantially.

Shiseido VOYAGER — The Data-Integration Path

Shiseido in Japan has chosen a different route. Co-developed with Accenture, VOYAGER went into full operation in February 2024 as an AI-driven beauty manufacturing and supply-chain platform that integrates demand forecasting, production optimization, and quality management using 500,000+ data points [25]. VOYAGER is closer to data integration plus AI decision support than to robotic automation per se. It illustrates what the cosmetics industry must do before process automation — collect the data.

Amorepacific Beauty Park — K-Beauty's Data Infrastructure

Korea's Amorepacific Beauty Park is an advanced beauty manufacturing hub combining smart factory and logistics, collecting and analyzing 600 million data points per day [2]. IoT sensors, AI vision, and automated logistics converge in this representative K-Beauty case. The 600 million figure is impressive, but "collecting data" and "using data" are different categories of work. Cosmax likely already collects substantial process data; the real question is whether that data actually enters the decision loop.

Cosmax's Position — The Asset and Burden of Being the ODM Leader

Cosmax achieved 2025 revenue of ₩2.4 trillion (+10.7% YoY), reinforcing its position as the global ODM beauty leader [6]. Core AI assets include a fragrance prediction AI built on 8,600+ molecular data points, a smart color matching AI, and the HelloBiome microbiome AI partnership, all developed and operated by the CAI lab.

The Thailand new-factory investment ($43.9M, 230 million units of annual capacity = 3x current, going live September 2026) is the centerpiece of the Southeast Asia and Middle East expansion [7].

A crucial recognition follows. Cosmax's 8,600-molecule fragrance prediction AI sits in the same category as the BASF AI Reactor and Sanofi SimplY AI from Section 7.2 — Cosmax is already at the frontier of research automation in global cosmetics. The gap is in process automation and line automation, the subject of Section 7.6.

How Ingredient Diversity and Formulation Complexity Make Automation Hard

The structural reason cosmetics ODM process automation is hard is ingredient diversity. A single Cosmax line handles hundreds of bases, and counting micro-additives the count runs into the thousands. Even on the same lipstick line, color, viscosity, and fragrance differ subtly per SKU. This is not the same as a car factory running 5–10 models on one line.

Add to that the fact that mixing order, temperature, and agitation speed also vary per SKU. For an automation system to handle all combinations cleanly, every combination has to be validated. In an ODM environment where a new SKU enters every week, that is unrealistic. Therefore, cosmetics ODM automation is realistic not as "fully automate one line" but as "partially automate by process step" — starting with standardizable steps such as filling, labeling, box packing, and QC vision.

The Beauty AI Market — A Trap in the Headline Number

A final word on market data. The Beauty AI market is forecast to grow from $4.9B (2025) to $33.75B (2035) at a CAGR of 22.3% [22]. The headline is glossy, but decomposed, personalized beauty, AI skincare analysis, and virtual try-on dominate the segments. Most of the growth is B2C consumer AI, not B2B manufacturing AI. ODM manufacturing AI is a small slice within that. Cosmax should not rest on the headline growth and instead measure the actual size of its own slice.

7.4 Food — The Twin Challenge of Hygiene and Variability

Food and beverage faces a similar order of automation challenges to cosmetics, but hygiene regulation is one notch tighter.

Unilever Dubai Lighthouse — An Exceptional Success

The most cited success in food and consumer-goods automation is Unilever's Dubai Lighthouse [28]. As a WEF Global Lighthouse Network site, the quantitative results are:

  • Labor productivity +400%
  • Changeover time -85%
  • 2020-2024 cumulative productivity +27%
  • Waste -41%

Labor productivity +400% is nearly unrealistic by food industry norms. Part of it comes from the special Middle East context (low-cost expatriate labor, greenfield plant, single-category focus). Even so, waste -41% and changeover time -85% are universally applicable improvements on the environmental and operations side.

Tyson Foods — A Frontal Assault in Food Processing

In the United States, Tyson Foods invested $300M in robotic palletizing and automated sorting at its Danville, Virginia plant, and announced a target to cut chicken-processing headcount by 20-30% by 2032 [27] — one of the largest robotization cases in food processing. The case is also tied directly to social cost: thousands of jobs eliminated as a direct result of robot deployment. From an ESG perspective, this is a variable Cosmax must weigh when designing its own automation strategy.

Nestlé SAP Migration — Digital Infrastructure First

A different path is taken by Nestlé. Nestlé executed the world's largest SAP migration [17]. 335 plants, 112 countries, 50,000 employees were unified onto a single SAP platform, with AI-driven supply-chain optimization, demand forecasting, and real-time production monitoring layered on top. This is "digital infrastructure integration" rather than "factory automation," but it is a prerequisite for substantive Physical AI — without integrated data, there is no dataset for AI to train on.

Kraft Heinz "The Cookbook" — Software Is Easy, Physical Is Hard

The most candid lesson comes from Kraft Heinz. Their AI agent The Cookbook, built on Azure OpenAI in under three months, cut product development time by 50% and improved pickle production efficiency by 12% [13].

Look at these numbers carefully. The 50% reduction in development time is impressive — but that is a software-domain change. Recipe optimization, ingredient substitution, consumer customization — all happen inside a computer. The 12% improvement in pickle efficiency, by contrast, is a physical-process change, and it is single digits. The fact that the same company, in the same AI deployment, sees 50% in software vs. 12% in physical captures the reality of food-industry Physical AI more honestly than any single metric.

Food AI Market and the Weight of Hygiene Regulation

The food AI and safety market is forecast to grow from $2.7B (2024) to $13.7B (2029), a CAGR of 30.9% [16]. Defect detection, quality control, and supply-chain traceability are the core application areas. According to a 2025 WEF review of Physical AI in food manufacturing, the food and beverage sector is adopting Physical AI faster than several traditional industries — but food-grade materials and cleaning requirements push robot prices up and smaller food companies are gated out of access [29].

The same constraint applies structurally to cosmetics. Robots on cosmetics lines face hygiene demands at food-grade levels. Half the reason Cosmax cannot directly drop in robots validated on automotive lines lies right here.

7.5 Apparel and Consumer Goods — The Nightmare of Flexible Materials

The final section covers the industry where automation has progressed slowest — apparel.

SoftWear Sewbot — A Precise Case of Promise and Limit

The icon of apparel automation is SoftWear Automation's Sewbot [26]. Combining computer vision and a robotic arm, it produces a T-shirt in 22 seconds at a labor cost of $0.33 per piece, with capacity of 600 pieces per shift. The fashion company BESTSELLER is an investor.

The numbers are striking. A 22-second T-shirt is faster than any human seamstress. But the limit is equally clear — what Sewbot makes is a T-shirt. That is, only garments composed of simple shapes, few seams, and standard fabric are automatable. Shirts, jackets, and dresses — with complex seams, multi-layer fabrics, and curved stitching — still depend on human hands, and most apparel-industry revenue lives in the latter.

Nike Grabit — 20x Upper Manufacturing Speed; What Comes Next?

Nike, partnering with the electroadhesion company Grabit, achieved 20x speedup in upper manufacturing [18]. Electroadhesion is interesting — it grips fabric using electrostatic principles, sidestepping the deformation problem suction cups create. But this technique is effective only for flat parts like uppers; the rest of shoe assembly (sole bonding, stitching, lacing) remains an automation challenge.

Nike's Mexico Automation Failure — When Automation Is Not the Answer

The more important Nike case is the Mexico factory automation failure [18]. In the late 2010s Nike attempted to introduce automation lines in Mexico and ultimately kept 5,000 jobs in place. The reason is simple: on the Mexican labor-cost curve, automation ROI did not pencil out. The same automation system pays back in the United States, Japan, or Korea, but not in Mexico, Vietnam, or Bangladesh. Half the reason apparel is the slowest industry to automate is that moving production to low-labor-cost countries is cheaper than automating.

H&M + Smartex — AI Quality Control Is Possible

On the other side of automation — quality control — apparel has progress to show. The H&M + Smartex partnership introduced an AI-based real-time knitting defect detection system [10]. Computer vision detects defects during knitting and is integrated with robotic precision-cutting to minimize fabric waste. The pattern is clear: manipulating flexible materials is hard, but inspecting flexible materials is (relatively) easy — because cameras and AI never have to grip the fabric.

Inditex/Zara — AI Demand Forecasting Instead of Automation

Within the same apparel industry, Inditex (operator of Zara) succeeded with AI demand forecasting instead of automation [12]. Inditex achieved excess inventory -20%, a 2-3 week design-to-store cycle, and +7.1% revenue post-AI deployment. The interesting part is that the physical processes are largely unchanged; what changed is the data flow. Same sewing factory, but AI now decides which design to make and how much.

Why Deformable Material Manipulation Remains an AI Challenge

The ICRA 2024 cloth manipulation workshop pitched eleven teams against each other on cloth grasping, unfolding, folding, and seam preparation [11]. The result is simple: neither data-driven approaches (end-to-end learning) nor model-based approaches (cloth physics simulation) could simultaneously satisfy real factory speed and accuracy. Cloth is a deformable material with infinite degrees of freedom, and its shape changes completely depending on grip location, force, and direction. This is a different category of problem from a car panel's 6 DoF.

The same problem applies in part to cosmetics. Soft packaging materials (pouches, films, labels) and viscous-base filling on a cosmetics line are variants of the same deformable-material problem. That is why advances in apparel automation can be transplanted directly into parts of cosmetics ODM processes.

Why Apparel Automation Maturity Trails Other Manufacturing

According to a 2025 systematic review of apparel automation, of the three areas — sewing, cutting, and quality inspection — only quality inspection has reached substantive automation, and deformable material manipulation remains the largest technical barrier [3]. Small-batch, high-mix production requirements again make automation hard. The conclusion that apparel-industry diversity precludes a single solution applies equally to cosmetics ODM.

7.6 Common Lessons for Manual-Labor Industries

To pull together the four-industry tour for a Cosmax executive office, the message reduces to three lines.

Distinguishing "Automatable" from "Worth Automating"

First, every case here shows the same pattern: the set of automatable things ⊃ the set of things worth automating. SoftWear Sewbot can automate T-shirts but not shirts. P&G Berlin can run lights-out for 4 hours but not 24. Pfizer compressed PAXLOVID cycle time by 67%, but not other drugs. The boundary of automation is not drawn by technical feasibility but by ROI justification.

When Cosmax evaluates an automation tool, the first question is not "does this tool work?" but "does this tool produce ROI in our SKU, line, and labor-cost combination?"

How Not to Be Fooled by PR Articles

Second, even in the same company's same AI deployment, effect sizes differ by domain. Kraft Heinz's The Cookbook is a clean example — product development time -50% (software) vs. pickle production efficiency +12% (physical). Read this gap incorrectly, and you will miscalibrate the headline "AI delivers 50% efficiency" into your automation ROI model.

Five questions to ask when reading a PR article:

  1. Is this effect a software transformation or a physical process transformation? (Mostly the former.)
  2. Is this effect on a single line or enterprise-wide? (Mostly the former.)
  3. Is this a pilot result or a 3+ year stable run? (Mostly the former.)
  4. Is this in a greenfield plant or a brownfield retrofit? (Greenfield is overwhelmingly easier.)
  5. Are this company's labor cost, SKU count, and regulatory environment similar to ours? (Mostly not.)

Cosmax's Realistic Starting Point

Third, despite all of this, there is a clear opportunity for cosmetics ODMs. The Self-Driving Laboratory (SDL) paradigm from Section 7.2 maps directly onto Cosmax's fragrance prediction AI, color matching AI, and HelloBiome microbiome AI — Cosmax already sits at the research-automation frontier of global cosmetics. The gap is in process automation. And the starting point for that process automation is not an integrated solution like Sewbot or Sequoia, but the combination of logistics automation à la L'Oréal Suzhou SMART Center plus data integration à la Shiseido VOYAGER.

The next chapter builds on this recognition to map Cosmax's realistic starting point in concrete terms — what to start first, where, at what cost, and with whom.

References

  1. Abolhasani, M., Coley, C., et al. (2025). Self-Driving Laboratories 2025: Automating the Scientific Method. Nature Reviews / Science Review. https://www.nature.com/articles/s41570-023-00527-5
  2. Amorepacific (2025). Amorepacific Beauty Park: Data-Driven Cosmetics Manufacturing. Amorepacific Press Release. https://www.amorepacific.com/us/en/company/pressRoom.do
  3. Apparel Automation Review (2025). Apparel Manufacturing Automation: A Systematic Review 2025. arXiv / Journal of Manufacturing. https://arxiv.org/search/?searchtype=all&query=apparel+automation+robotics+2025
  4. AstraZeneca (2025). AstraZeneca Wuxi WEF Lighthouse: AI-Driven Pharmaceutical Production. WEF Global Lighthouse Network / AstraZeneca Press. https://www.weforum.org/press/2024/01/world-economic-forum-announces-21-new-global-lighthouse-network-sites/
  5. BASF (2025). BASF AI Reactor: Accelerating Chemical R&D with Machine Learning. BASF Press Release / Industry Report. https://www.basf.com/global/en/who-we-are/organization/group-companies/BASF_New-Business-GmbH/news/press-releases/2024/06-2024/basf-and-chemspeed-technologies-announce-collaboration.html
  6. Cosmax (2025). Cosmax 2025 Performance and AI Strategy: ODM Beauty Leader. Cosmax Investor Relations / Annual Report. https://www.cosmax.com/ir
  7. Cosmax (2026). Cosmax Thailand New Factory: Tripling Production Capacity. Cosmax Press Release. https://www.cosmax.com/news/thailand-factory
  8. Dow (2025). Dow 'Transform to Outperform': AI and Workforce Restructuring. Dow Investor Relations. https://investor.dow.com/news-releases/news-release-details/dow-announces-restructuring-and-transformation-drive-sustainable
  9. Grand View Research (2025). Pharmaceutical Laboratory Automation Market 2024-2030. Market Research Report. https://www.grandviewresearch.com/industry-analysis/pharmaceutical-laboratory-automation-market
  10. H&M and Smartex (2024). H&M + Smartex: AI-Powered Knitting Defect Detection. Smartex Press Release / H&M Partnership. https://www.smartex.ai/news
  11. ICRA (2024). Cloth Manipulation Robots ICRA 2024: Deformable Object Handling Challenge. ICRA 2024 Workshop on Deformable Object Manipulation. https://icra2024.org/workshops
  12. Inditex (2024). Inditex/Zara AI-Driven Demand Planning and JIT Manufacturing. Inditex Annual Report 2024. https://www.inditex.com/en/investors/investors_and_shareholders/annual_report
  13. Kraft Heinz (2024). Kraft Heinz AI Agent 'The Cookbook': Azure OpenAI-Powered Product Development. Microsoft Azure Case Study. https://customers.microsoft.com/en-us/story/1648514524718400040-kraft-heinz-consumer-goods-azure-openai-service
  14. L'Oréal (2025a). L'Oréal Operations 4.0: GenAI, Sustainable Formulation, and Speed-to-Market. L'Oréal Annual Report / IBM Partnership. https://www.loreal-finance.com/en/annual-report-2024
  15. L'Oréal (2025b). L'Oréal SMART Fulfillment Center Suzhou: Cosmetics Logistics Automation. L'Oréal Operations Press. https://www.loreal-finance.com/en/press-release/loreal-opens-smart-fulfillment-center-suzhou
  16. MarketsandMarkets (2024). AI in Food Safety and Manufacturing Market 2024-2029. Market Research Report. https://www.marketsandmarkets.com/Market-Reports/ai-in-food-and-beverages-market-53289914.html
  17. Nestlé (2025). Nestlé SAP Digital Migration: World's Largest ERP Transformation. SAP Press Release / Nestlé Annual Report. https://www.sap.com/about/company/sap-customers/nestle.html
  18. Nike (2024). Nike Factory Automation: Grabit Electroadhesion and Mexico Lessons. Nike Annual Report / Industry Analysis. https://about.nike.com/en/newsroom/releases/nike-manufacturing
  19. Novo Nordisk (2025). Novo Nordisk + OpenAI Strategic Partnership: AI Across All Business Divisions. Novo Nordisk Press Release. https://www.novonordisk.com/news-and-media/news-and-ir-materials/news-details.html?id=171835
  20. P&G (2024). P&G AI Factory: 10x Deployment Speed and Lights-Out Manufacturing. P&G Press Release / AI Factory Announcement. https://pginvestor.com/financial-reporting/press-releases/news-details/2024/PG-AI-Factory/default.aspx
  21. Pfizer (2024). Pfizer AI-Driven Manufacturing: PAXLOVID Cycle Time Reduction. Pfizer Press Release. https://www.pfizer.com/news/press-release/press-release-detail/pfizer-announces-strategic-artificial-intelligence-ai
  22. Precedence Research (2025). Beauty AI Market 2025-2035: $4.9B to $33.75B. Market Research Report. https://www.precedenceresearch.com/beauty-ai-market
  23. Sanofi (2025a). Sanofi Modulus: Modular Next-Generation Biomanufacturing Facility. TIME Best Inventions 2025 / Sanofi Press. https://time.com/collections/best-inventions-2025/7318453/sanofi-modulus/
  24. Sanofi (2025b). Sanofi SimplY AI: Digital Manufacturing & Supply Accelerator. Sanofi Official Magazine. https://www.sanofi.com/en/magazine/our-science/digital-and-ai-powered-manufacturing-and-supply-delivering-faster-and-safer-for-patients
  25. Shiseido (2024). Shiseido VOYAGER: AI-Powered Cosmetics Manufacturing Platform. Shiseido Press Release. https://www.shiseido.com/corporate/news/2024/voyager
  26. SoftWear (2025). SoftWear Automation Sewbot: Fully Automated Apparel Sewing. SoftWear Automation Press Release / BESTSELLER Partnership. https://softwearautomation.com/sewbot/
  27. Tyson (2025). Tyson Foods Factory Automation: $300M Danville Robotics Expansion. Tyson Foods Press Release / Reuters. https://www.tysonfoods.com/news/news-releases/2023/11/tyson-foods-invests-300-million-expand-danville-virginia-pork-facility
  28. Unilever (2024). Unilever Dubai Lighthouse: WEF-Recognized Manufacturing Excellence. WEF Global Lighthouse Network / Unilever Press. https://www.weforum.org/press/2024/09/world-economic-forum-announces-new-global-lighthouse-network-sites/
  29. WEF (2025). Physical AI in Food Manufacturing 2025: Industry Review. WEF White Paper / Industry Review. https://reports.weforum.org/docs/WEF_Physical_AI_Powering_the_New_Age_of_Industrial_Operations_2025.pdf