The core idea behind the framework is that consumers constantly broadcast their own opinions about which brands belong together. When people post on social media mentioning two brands in the same breath, those co-mentions constitute a large-scale, continuously updated record of how the public perceives brand relationships. By combining this organic signal with traditional brand characteristics, the proposed model can hand brand managers a suite of potential collaboration opportunities tailored to their specific businesses, rather than a shortlist assembled through intuition or expensive market research.

The framework’s performance in validation tests suggests the approach can separate genuinely promising pairings from unlikely ones. In a survey of brand managers, brand pairs identified as favorable by the framework had nearly a 90% chance of receiving a higher selection rate than those the model eliminated as unlikely pairings, the researchers wrote. Beyond ranking candidates, the framework also categorizes its recommendations and provides practical guidance on the best way to pursue a collaboration, giving companies clearer insight into which brand alliances they should prioritize.

“Finding brand collaborations is a needle-in-the-haystack problem,” said co-author Keran Zhao, assistant professor of supply chain and information systems at Penn State’s Smeal College of Business. “Brand A might consider brands B, C and D as options, but they will never comprehend the whole brand network. The advantage of this framework is that it provides a systematic evaluation of hundreds of brands, as well as a detailed explanation about the how and why behind an opportunity.”

Technically, the researchers describe the framework as a four-step system built around two key concepts: signal fusion, the ability to synthesize disparate bits of information into a cohesive insight, and actionability, the capacity to translate that insight into guidance a manager can actually use. The first phase concentrates on signal fusion. It combines real-world brand information with co-mentions on social media to construct a comprehensive network of brand connections, a structure in which the strength and character of relationships between hundreds of brands can be evaluated simultaneously rather than one candidate pair at a time.

To build that network, the researchers assembled two complementary data sources. They collected brand personality survey information covering characteristics such as sophistication and familiarity for 450 top business-to-consumer brands. They then gathered 477,933 posts from the social media site Twitter, now X, in 2020 that mentioned these brands together. Using AI-based text classifiers, the team filtered out complaint-related posts and analyzed whether consumers viewed the co-mentioned brands as complementary or competing. These classified insights were combined with the brand characteristics and co-mention frequencies to produce a broader network of brand relationships capable of flagging promising alliance opportunities.

The scale advantage of automated analysis is central to the method. “The four of us could sit together for eight to 10 hours reading Twitter posts, and we would still only see a small fraction of the conversations happening around these brands,” Zhao said. “AI allows us to analyze this information at a much larger scale.” The researchers also tested whether the network could anticipate alliances that had not yet formed, examining its ability to predict future real-world partnerships beyond the period covered by the training data. The model accurately predicted 213 of the 251 alliances that occurred after the data sample’s cut-off date.

The second phase of the framework subjects these proposals to human judgment, and the researchers emphasize that this stage was conducted solely for validation purposes, meaning users would not be required to run such surveys during real-world applications of the model. A consumer survey with 350 participants found a strong preference for the model’s recommendations over pairings the model had eliminated. The team also surveyed 57 brand managers, asking each to evaluate 24 brand pairs, 12 predicted by the model and 12 eliminated, and to select up to eight of the most promising. On average, the managers selected the model’s proposed alliances roughly 34% of the time compared with only around 7% for the eliminated pairs, a nearly fivefold increase in interest.

Phases three and four were both designed to convert the raw list of alliance opportunities into actionable insight. In the third phase, the researchers used machine learning systems, a subset of AI that learns from patterns in data, to categorize the proposed collaborations according to how frequently the two brands appeared in posts together, the sentiment of those posts, and the number of likes, replies and reposts they attracted. The model then plotted the proposals on a matrix with four distinct quadrants, ranging from “premier opportunities,” representing the strongest and most high-visibility alliances, to “hidden niches,” which highlight more exploratory collaborations in emerging or specialized markets.

“One brand manager might run the model, and it might suggest 100 brands they should collaborate with,” Zhao said. “That’s practically impossible to do, so this stage gives a more systematic view for explaining to brands how important the opportunities are and which ones they should prioritize.” The matrix can also be tweaked to tailor its insights to different stakeholder roles, for example by offering a portfolio-level perspective that could help identify internal brand competition, or by organizing recommendations by industry to highlight cross-sector opportunities.

The fourth phase turns to natural language processing and generative AI to produce two pieces of practical guidance for each individual brand collaboration, drawing on context from the social media posts themselves. The first piece of generated guidance identifies which specific marketing element consumer discourse is dominated by, such as discussions about products or promotions. The second offers deeper insight into what the discourse is thematically about, like style or sustainability efforts. “AI is very powerful at summarizing information,” Zhao said. “If we can locate what customers are talking about and feed that information to AI, it can give us some direct interpretations about which direction to go when pursuing a brand alliance opportunity.”

The stakes behind these decisions are considerable. Finding a promising brand alliance, whether a jointly launched product or a collaborative ad campaign, is difficult on its own, the researchers explained, but committing to one also presents a significant risk for a business, especially given the time, money and brand reputation at stake once a collaboration is announced. A framework that both surfaces candidates and explains how to pursue them addresses both halves of that problem, pairing discovery with execution in a single pipeline.

“A scalable system that can identify promising opportunities and provide specific recommendations for how to proceed could be an invaluable tool for brand managers, and an intriguing well of crowd-sourced creativity,” Zhao said, noting that he and his collaborators would test the system with unexpected brand combinations for fun. “A lot of surprising things can happen.” The study’s co-authors include Pankhuri Malhotra of the University of Oklahoma, Daniel Ringel of Vienna University of Economics and Business, and Yaxin Cui, who earned a doctorate degree at Northwestern University and is now at Wells Fargo. The research was published in Information Systems Research on 28 July 2026.

Subject of Research: A data-driven AI framework that discovers and evaluates brand alliance opportunities from social media co-mentions

Article Title: Crowd-sourced collabs: Using AI, social media to find potential brand partners

Article References: Crowd-sourced collabs: Using AI, social media to find potential brand partners. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, brand collaborations, social media analytics, machine learning, natural language processing, marketing, consumer behavior, brand alliances, Information Systems Research, signal fusion, generative AI, Penn State