How Toshiba Tec and McKinsey are turning retail data into real-time decisions with NVIDIA

For decades, Toshiba Tec’s point-of-sale terminals have been at the front lines of global commerce, capturing billions of data points from every swipe and scan. Yet behind that data was untapped potential.

In 2024, Toshiba Tec saw a pivotal opportunity to reinvent itself—not just as a hardware maker, but as the intelligence engine powering the next generation of retail. Its systems handled millions of daily transactions, generating vast datasets that had long been processed through CPU-based analytics—slow, linear, and incapable of real-time insight.

Retailers have long been able to analyze promotions and sales trends retrospectively, yet real-time personalization and live profit optimization remained out of reach.

To address this challenge, Toshiba Tec established Gyainamics—a new software engineering company within its business-building factory—to drive its transformation beyond hardware. Through a collaboration with McKinsey and its AI arm, QuantumBlack, and NVIDIA, Toshiba Tec and Gyainamics turned this data challenge into a market-shifting opportunity, bringing accelerated computing to consumer data and analytics applications. Designed to incubate and scale software ventures, Gyainamics supports Toshiba Tec’s strategy to build new revenue streams by 2030. As part of this effort, McKinsey worked with Toshiba Tec to expand the team and build core capabilities, including the ability to run docker containers with advanced libraries such as Transformers4Rec and NVIDIA Merlin NVTabular.

This integration of NVIDIA AI infrastructure redefines how GPUs power the next generation of AI-driven consumer and retail data intelligence, enabling use cases such as more frequent promotion tests, stronger product recommendations across the full catalog, and real-time adjustment of offers at checkout.

The technical breakthrough of Toshiba Tec and Gyainamics came from deploying the NVIDIA Merlin framework, combining NVTabular for GPU-based feature engineering and Transformers4Rec for next-generation recommendation modeling.

NVTabular shifted feature engineering—normally CPU-bound and laborious—to NVIDIA AI infrastructure with NVIDIA Ampere GPUs, automating preprocessing for millions of transaction records in minutes. Retail data, previously taking hours to normalize and encode, could now be processed and tagged in near real time. The resulting dataset schema seamlessly fed into a transformer-based model that predicted customer preferences with precision.

Transformers4Rec, inspired by the same architecture that powers large language models (LLMs) like GPT, learns the sequential behavior of customers—such as what products they bought, in what order, how frequently, and in which basket or trip context—to predict the next likely purchase. The system dynamically adjusts coupon offers based on individual price sensitivity and profit structures unique to each retailer.

Compared to older methods like collaborative filtering, the GPU-enabled transformer achieved higher coverage of a retailer’s long tail catalog (from roughly 8 percent to nearly 100 percent) and improved personalization scores sevenfold. Training time for large datasets dropped by more than 80 percent, while inference speed for 100,000 customers fell from hours to under a minute.

“Most enterprises face challenges not in building AI once, but in continuously operating and scaling it,” says Hiroyuki Koyama, President and CEO of Gyainamics. “Our partnership model with McKinsey solves that gap. That means clients access cutting-edge technologies—such as GPU accelerators and transformer-based recommendation systems—while ensuring that these tools remain production-ready, secure, and continuously optimized.”

By integrating NVTabular with NVIDIA cuDF and Dask, the team preserved a familiar data-science workflow while gaining exponential speed. Metadata tagging of categorical features, such as product IDs or store codes, became automatic. These efficiencies compounded during model training, allowing for rapid iteration and experimentation.

This transformation signals a new chapter for Toshiba Tec.

“Speeding up retail analytics using NVIDIA Ampere-generation GPUs marks a true inflection point,” said Takuya Kudo, a McKinsey partner. “By combining Toshiba Tec’s deep retail-data foundation, NVIDIA AI infrastructure and software, and McKinsey’s applied-AI modeling, we’re demonstrating how advanced architectures such as transformers can reshape an entire industry.”

Indeed, transformer-based personalization has outperformed rule-based promotions in both accuracy and relevance. In controlled pilots, retailers recorded an average 5 percent lift in sales and profit per targeted segment, compared with historical baselines where coupon campaigns yielded less than 1 percent incremental margin.

Its AI-powered platform now helps retailers adjust promotions in real time—lifting transaction value by 5 percent, SKU coverage to 99.9 percent, and long-term customer value by up to 7 percent compared with previous manual segmentation campaigns.

How Toshiba Tec, NVIDIA, and McKinsey are turning retail data into real-time decisions

Each checkout terminal has become a source of insight, translating every purchase into data that fuels smarter decisions. “The impact has been transformative,” says Hironobu Nishikori, President and CEO of Toshiba Tec. “Retailers can deploy promotions with real-time ROI visibility, while consumer-goods companies gain transparency into how their trade spend is performing. This means no more spending trade budgets without understanding their true effectiveness.”

Powered by NVIDIA AI infrastructure and backed by McKinsey’s industry and applied-AI expertise, Toshiba Tec and Gyainamics have built a system that learns as fast as consumers shop.

In doing so, the company has not only expanded its business model but also helped define a new frontier for retail analytics—one where intelligence runs at the speed of experience.

“Ultimately, we are creating a system that benefits retailers, manufacturers, and customers together,” Nishikori says.