Campbell Brown is CEO & Co-Founder of PredictHQ – a real-world context platform powering enterprise AI decisions.
A 3,000-person fitness conference in Boston had a bigger impact on local hotel demand than a 30,000-person baseball game.
According to our data at PredictHQ, although the conference was smaller, less visible and less newsworthy, it drove a measurable 11.4% uplift above the hotel’s seasonal baseline. The baseball game 4.2 miles away and 10 times the size had minimal (1.3%) impact.
The event size didn’t predict the impact on demand. The proximity and audience type did. And neither variable existed in the hotel’s historical booking data.
I talk regularly to executives across retail, hospitality and transport, and the reaction is almost always the same: They recognize it immediately because they experience these misses systemically.
Why This Keeps Happening
There’s a consistent pattern in public earnings calls. A CEO or CFO explains a missed target. They attribute the miss to external events “beyond their control,” be it weather, events or unexpected demand shifts.
After a decade building in demand intelligence, I can say with confidence that in most of these cases, the volatility was predictable. The problem is that the AI models running their forecasting, pricing and operations never had access to the right data.
The root cause isn’t a shortage of data, models or compute capacity. It’s a context gap: the structural disconnect between what a model knows (historical patterns) and what it needs to know (what’s about to happen in the real world).
Enterprise AI models are built on internal data like transactions, inventory, customer records and historical patterns. They’re optimized to find signals within an organization’s own operations. What they can’t see is what’s happening outside those four walls.
Most forecasting and AI systems treat external signals as noise rather than as predictable patterns. Yet, concerts, festivals, sporting events, school calendars, public holidays, severe weather, conferences and community gatherings all drive foot traffic, purchase behavior and operational load.
It Compounds In Ways Models Can’t See
The Boston example isn’t an isolated anomaly. It’s how the real world works.
On March 20, 2025, three major events converged simultaneously at the Georgia World Congress Center: a 5,000-person Applied Power Electronics Conference & Exposition (APEC) summit, a 12,000-person dental expo and a 60,000-person auto show. According to our data, they resulted in a 25% lift in hotel demand that no single-event model would have predicted. Two of the three had changed location or timing from the prior year and wouldn’t appear in last year’s data for that date. The interaction between events, like their clustering, their audience overlap and their collective draw on accommodation and services, is what created the demand spike. Historical models can’t model interactions between events that didn’t co-occur in prior years.
Every new AI model you deploy that lacks this external context is making decisions in a partial world. And as enterprises deploy AI agents across pricing, inventory, staffing and marketing, the cost scales with the number of agents and decisions. An AI agent that adjusts staffing levels based on historical averages without knowing that a 20,000-person concert is happening two blocks away will produce an inaccurate decision that leaves revenue on the table. Agentic AI systems that lack real-world context will scale bad decisions faster than any human team ever could.
Four Questions To Ask Of Every AI System You Operate
For each AI system making operational decisions in your organization, what does it know about the world that exists outside of your own data?
1. Are you currently using any external context in your demand forecasting models?
2. Can your models explain why demand shifted on a given day, beyond seasonal patterns?
3. Are your forecasting models location-specific, or do they apply network or chain-wide averages?
4. Can your planning team see what’s about to happen, or only what’s already happened?
If the answer to most of these is no, your organization has significant context gap exposure. The highest-exposure categories to prioritize are demand forecasting, dynamic pricing, staffing allocation and inventory management. These are the domains where small information failures compound into significant revenue and cost consequences at scale.
Conclusion
External context should be treated as infrastructure, with the same seriousness that organizations apply to data quality and model governance.
Context improves models, and better models improve organizational decisions. This creates a virtuous loop where those better decisions then generate the data that improves models further. If your models are lacking real-world context, you’re flying blind when it comes to anticipating the future with any degree of accuracy.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?