.png){kind=logo}
Restaurant profitability remains under pressure. In the U.S., roughly one in three restaurant operators said their businesses were not profitable in the first half of the year, according to the National Restaurant Association’s analysis, which expects elevated food and labor costs to keep pressuring margins through year-end.
Automation and data analytics, along with structural architectural innovations, are areas where technology can improve operational efficiency. Sakshit Teotia, a Principal AI Tech Architect at Boston Consulting Group with 15+ years of IT and artificial intelligence experience, proposes replacing legacy spreadsheets and systems with AI-driven labor-scheduling and demand-optimization engines. Here, he engineered a predictive forecasting system that integrates real-time external data to generate store-level labor shift models. Deployed and expanding across nearly 100 restaurant locations, affecting over 1500 hourly workers, the solution eliminates overtime overhead and makes staffing schedules more predictable.
Drawing on Sakshit’s experience and achievements, this article explores how external demand signals can translate into store-level schedules that reduce manual planning and make staffing more predictable.
Quick-service chains have historically managed staffing and inventory through legacy systems and historical sales averages. This reactive approach fails to account for sudden localized fluctuations in foot traffic, leading to either costly overstaffing or understaffed peak hours that degrade customer throughput.
"Imagine store managers spend four hours at the start of the week building schedules, only to spend the rest of the week making panicked mid-shift cuts or burning out lean teams during unexpected rushes," Sakshit notes.
He engineered an AI-driven labor scheduling and demand-optimization system that replaces manual estimates with machine-learning models.
The system integrates external data streams, such as localized weather forecasts, demographic patterns, supply chain conditions, and regional sporting event schedules. By analyzing these external inputs alongside store-level historical data, the engine predicts customer volume and generates optimized shift schedules for store managers.
The system has moved beyond a single pilot and is being used across two large U.S. restaurant chains, which have deployed it across more than 100 locations, covering over 1000+ hourly employees. The scheduling engine is designed to limit unnecessary overtime, improve peak-hour coverage, and give employees more predictable schedules. While the platform cuts overhead by eliminating unnecessary overtime, its real significance is providing predictable, reliable shift schedules for frontline wage earners.
"Technology creates value when it changes how people actually work, and they are willing to use it," Sakshit explains. "In quick-service environments, enterprise AI fails if it is engineered purely for corporate dashboards. The algorithm must solve immediate frontline friction for store managers and hourly staff. Involving end users in that process also helps us understand whether the system can be incorporated into daily workflows."
Yet even the most advanced algorithmic architecture risks failure if an organization lacks the internal capacity to execute and sustain it. To prevent algorithmic deployment failures, Sakshit focuses on executive alignment and enterprise-wide skill acquisition. He proposed and built a context-aware Executive Digital Twin with a team after the project received a $500,000 internal development grant. Within BCG North America, which had roughly 7,000–8,000 employees, an estimated 40 teams submitted proposals; 15 were shortlisted to present to senior leadership, and three ultimately received grants. The agent is designed to reproduce an executive’s knowledge and decision patterns for tasks such as attending meetings, making decisions and responding to operational events.
To solve adoption barriers at the workforce level, he co-developed an AI Fluency Platform which addresses the enterprise AI skill gap through structured programs called Agentic AI Hackathons, where non-technical staff build practical software tools. By incorporating gamification and standardized development frameworks, the platform enables employees across divisions to create functional production grade AI solutions within two to three months. The platform currently supports hundreds of internal users at BCG and is being deployed to major enterprise organizations.
"The difficult part of enterprise AI is often getting people to trust the system and use it in their work,” Teotia says. “We define the business result first, show value early, and make sure the internal team can run the system after the project ends.”
The operational evolution of quick-service restaurants illustrates a broader trend in enterprise management. Predictive, context-aware platforms are replacing fragmented data and manual scheduling by calculating real-world variables in real time. By linking external data inputs directly to frontline operations, organizations stabilize labor costs, support hourly employees, and secure operational margins in unpredictable economic conditions.