CFVG Embracing AI to Elevate Foodservice

unaware of the questions they could or should be asking. This familiarity can make it difficult to gain support for new technologies. The second challenge is that hourly staff may also be hesitant, either because they don’t understand how AI works or because they fear it could replace their jobs. Klinger emphasized that educating teams on the benefits of AI is critical, but adoption often takes longer than anticipated, particularly outside highly engaged groups. Zaring added that another challenge arises from the divide between business experts and data or analytics specialists. While business experts understand operational complexity, data teams excel at modeling and interpreting infor- mation. Misalignment can occur when AI tools fail to reflect real-world needs, leaving business users unsure how to apply outputs or feeling that the models miss essential details. Zaring stressed the importance of identifying critical data points and creating tools that meet non-negotiable requirements to produce actionable results. Poye concluded that each business faces its own silos, and building a frame- work that harmonizes AI adoption across teams, ensuring both sides are aligned at each stage, can help mitigate tension and maximize the value of AI implementation. A challenge that we experience is there are business experts who understand the complexity of the business they’re in and then there are data and analytic experts who are very well-versed in synthesizing and understanding data... And I find the two sometimes are not in sync, or we’re working more in silos. Bonnie Zaring , Executive Director, Food Programs, and Offers, RaceTrac

150 locations (100 with foodservice), and the availability of skilled personnel. Brask addressed the model question, stating that various AI models (like OpenAI, Grok, Claude) continue to leapfrog one another, with OpenAI currently “playing catch-up.” Regarding implementation time, Brask estimated that from receiving data, Upshop could go live with demand forecasting within approximately 14 days, with this period primarily used for evaluating model accuracy. However, he clarified that while the initial machine learning model can be built in “weeks,” refining all inputs such as promotions, pricing, and sales data to optimal per - formance typically takes “six to nine months.” On the staffing front, Brask noted the availability of “lots of data scientists who are more than capable of building a demand forecast model.” However, Weber offered an alternative, suggesting that individuals skilled in prompting the different AI tools can add significant value, potentially more cost-effectively than traditional data scientists, especially for tasks like user interface (UI) workflow changes. Poye connected this to Galentine’s point about experienced employees, sug- gesting that training them in prompting would foster better engagement and leverage their insights into customer behavior and events. He highlighted AI’s potential to help manage competitive intrusion by adjusting forecasts, pricing, or promotions. He concluded that prompting is a wise investment as it’s not complex and allows employees to engage with relevant business questions. He then asked participants about the challenges they face within their organi- zations regarding AI adoption, both personally and professionally. Klinger highlighted two main challenges. The first is overcoming resistance from both managers and frontline staff who are accustomed to long-standing processes. He noted that experienced personnel often operate on “auto-pilot,”