Nov 17, 2026, 08:42
by
Aaron Bennett
Gies Business professor Abhijeet Ghoshal and colleagues developed a quantum computing method to help retailers hide sensitive product pairings while preserving the usefulness of sales data shared with suppliers.
Every time a customer checks out at a grocery store, the retailer learns something valuable. Which products are often purchased together? What unexpected shopping habits might give the company an edge over its competitors?
That information could also be enormously valuable to suppliers, helping them forecast demand, manage inventory, and make smarter production decisions. But sharing it can be risky. The same data that helps suppliers operate more efficiently could reveal a retailer's closely guarded competitive advantages.
New research from Gies College of Business explores how quantum computing could help solve that problem, allowing businesses to share valuable transaction data while concealing sensitive purchasing patterns. The study, led by Gies Business professor Abhijeet Ghoshal, develops a quantum computing approach that identifies which transaction records should be modified to protect sensitive information while preserving as much of the original data as possible.
Researchers have been developing ways to hide sensitive patterns in shared data for more than two decades, largely using conventional computers. Ghoshal’s study takes that work in a new direction by turning to quantum computers.
“A retailer's purchase data is more than a list of sales. It reflects years of experiments with pricing, store layout, and promotions, and it shows what worked. Handing that over can feel like giving away the playbook,” said Ghoshal, an associate professor of business administration and expert on quantum computing and its potential business applications. “So, retailers face a trade-off. Sharing helps both sides plan production and inventory and cut waste. Not sharing protects their competitive edge. Our work aims to remove that trade-off: hide the few patterns that are truly sensitive and share everything else.”
The hidden value in shopping carts

Retailers have been sharing transaction data with suppliers for decades. Walmart pioneered the practice with its RetailLink system in the early 1990s, and Home Depot, Target, Amazon, and Lowe’s have since built similar platforms.
Better information about consumer demand helps manufacturers avoid overproduction and respond more quickly to changing shopping habits. But retailers have reason to worry about where their information might end up.
Suppliers typically work with multiple retailers, including direct competitors. Even unintentionally, they can become conduits for sensitive information through their sales representatives, product recommendations, or promotional strategies.
Imagine a grocery chain places a new store-brand sauce next to premium bread, offering a small discount when customers purchase both. The promotion works unexpectedly well, and the retailer earns a healthy margin on the sauce.
If competitors discover the relationship, they could quickly copy the promotion, undercut its pricing, or develop a competing strategy. The original retailer would lose an advantage it invested time and money to discover.
The researchers focus on concealing these kinds of patterns, known as frequent itemsets, which are combinations of products that regularly appear together in customer transactions. One way to hide a sensitive itemset is to remove certain products from selected transaction records before sharing the data. The challenge is determining which records to modify. Changing too many reduces the accuracy and usefulness of the dataset. Changing too few could leave sensitive purchasing patterns exposed.
Finding the smallest number of transactions that must be modified to conceal all designated sensitive patterns is a notoriously difficult mathematical problem. As the number of transactions grows, the possible combinations become increasingly difficult to evaluate.
Giving quantum computing a head start
Ghoshal and his coauthors, Yan Li of Pennsylvania State University and Syam Menon and Sumit Sarkar of the University of Texas at Dallas, turned to a quantum computing method called the Quantum Approximate Optimization Algorithm (QAOA). The algorithm combines quantum and conventional computing to search for high-quality solutions to complex optimization problems.
The researchers first translated the data privacy challenge into a mathematical form compatible with QAOA. They then introduced an improvement that draws on their understanding of how retail transactions overlap.
Some transactions contain several sensitive purchasing patterns, making them particularly valuable candidates for modification. By giving those transactions a higher initial probability of being selected, the researchers provided the quantum algorithm with a more promising starting point. They call this technique a domain-inspired warm start.
Think of it as a search party looking for a lost hiker in the woods. Searchers could spread evenly across the forest, or they could use what they know about hikers' behavior to concentrate their efforts near trails and water.
“The quantum computer is the search party. Our hint is that records involving more of the secret patterns are good places to look,” Ghoshal said. “The algorithm still considers other possibilities, so if the best answer lies elsewhere, it can still find it. We simply weight the odds toward where good answers usually are.”
The team also modified how the algorithm explores possible solutions, developing a matching component that helps it take advantage of the informed starting point. Their experiments showed that combining the two improvements produced better results than changing the starting point alone.
A more reliable way to find the best answer
The researchers tested their approach on 35 small-scale problems drawn from two real-world retail transaction datasets. Each algorithm was run 25 times per problem, using different starting settings.
The conventional quantum algorithm found the optimal solution in 83.43% of runs, on average. The researchers' modified algorithm increased that success rate to 96.91%, making it considerably more reliable at identifying the smallest number of transactions that needed to be changed.
The improvement was particularly striking on some of the most difficult problems. For several datasets, the conventional approach found the optimal solution in just two of 25 attempts. The warm-start method achieved a 100% success rate across all 25 runs for 31 of the 35 problems, compared with 20 using the conventional approach.
The researchers also tested 12 problems on actual quantum computers, finding optimal solutions in nine cases with the warm-start method and 10 with conventional QAOA. In the remaining cases, the solutions required modifying just one additional transaction.
“The biggest difference shows up on the hardest problems,” Ghoshal said. “In practical terms, a more reliable method needs fewer repeat attempts, and time on real quantum hardware is scarce and expensive.”
The experiments remain a proof of concept. The researchers tested relatively small problems, and the warm-start method required more processing time in simulations. Commercial deployment would require substantially larger, more reliable quantum computers.
Still, Ghoshal sees an important lesson in the findings.
“Domain knowledge still matters in the quantum era,” he said. “A general-purpose quantum algorithm gets noticeably better when you build in what experts already know about the problem.”
Beyond retail data privacy
The research, “Quantum Approaches to Hide Sensitive Itemsets: Domain-Inspired Warm Start,” was published in IEEE Transactions on Quantum Engineering. It builds on more than two decades of research into protecting sensitive information in shared datasets.
Ghoshal sees potential applications in other industries, including healthcare and finance, where organizations could benefit from sharing information while protecting confidential patterns. Similar optimization challenges also arise in operations management, including airline crew scheduling and facility placement.
The immediate challenge is technological. Today's quantum computers have limited capacity and are prone to errors. Larger machines, better error correction, and more effective ways of combining quantum and conventional computing will be necessary before the approach can handle commercial-scale datasets.
For now, Ghoshal sees value in developing algorithms that will be ready when the technology matures.
“This work is part of building the toolkit ahead of time,” he said. “When large, reliable quantum computers arrive, companies that already have methods like this ready will be able to use them quickly.”
For retailers, that future could mean sharing more detailed information with suppliers while protecting the purchasing patterns that give them a competitive edge. Suppliers could use the additional information to improve production planning, reduce waste, and keep products on store shelves.
And for quantum computing researchers, the study offers another promising direction for turning an emerging technology into a practical business tool.