Fleets urged to build AI strategy around data, business value

Hans Galland (right) of BeyondTrucks and Chas Wurster of PrePass provide insights during an "AI 101" session at The AI Summit. (American Trucking Associations)

PITTSBURGH — Fleets preparing to deploy artificial intelligence should focus first on clean, accessible data and clearly defined business problems, not the latest technology pitch, panelists said at The AI Summit, hosted by American Trucking Associations’ Technology & Maintenance Council.

During an “AI 101” session Sept. 22, Hans Galland, CEO and founder of BeyondTrucks, and Chas Wurster, chief technology officer at PrePass, explained how today’s AI differs from earlier rule-based systems, where it is already being used in trucking and what fleets should evaluate before investing.

Galland described AI broadly as systems designed to act or think rationally or in ways that resemble human reasoning. Earlier “symbolic” systems relied on hard-coded rules, such as medical decision trees that mapped symptoms to diagnoses. Those deterministic tools remain useful, he said, but much of modern machine learning and generative AI is probabilistic.

That distinction creates a practical risk, Galland noted, stressing that a model can produce an incorrect answer with a high degree of confidence.

“I always say, in 2025 AI had an adoption problem. In 2026, AI is having a trust problem,” he said.

Galland cited a first-quarter carrier survey showing about 75% of fleets lacked a formal approach or clear position on AI. More than half were using the technology in some form, although many rated their understanding as limited. Safety applications generated the clearest perceived value because commercial products are mature and widely available.

He identified load planning and dispatch as a larger but less-developed opportunity. AI can help assign drivers and equipment to loads and schedules, but relatively few fleets are using it for those decisions today.

Fragmented data remains a major obstacle. Many fleets rely on legacy systems of record that require manual entry and do not continuously capture time- and location-based operating information.

“You need to start with architecture,” Galland said. “You need to make sure that your systems are capable of feeding AI the data that it needs.”

To prioritize investments, Galland recommended weighing a task’s value against its frequency. Automating high-frequency, low-value activities such as repetitive check calls can generate meaningful savings, he said.

High-value but less-frequent decisions, including network design, may justify more advanced analytics. Low-probability events with severe consequences, such as fatigue-related crashes, also can be strong candidates for safety-focused AI.

Wurster said fleets are already encountering narrow AI through their technology providers. He distinguished those task-specific systems from broad generative models. Tools such as ChatGPT are trained on large datasets to predict likely responses, while safety-critical systems generally rely on narrower models, rules and engineered constraints designed for specific tasks.

Current trucking applications include route optimization, predictive maintenance based on sensor and telematics data, driver-assistance systems that process video and other inputs, and document-recognition tools that extract and classify information.

“You are already using it. A bunch of your vendors are already using it,” Wurster said. “It improves safety. It improves efficiency. It improves uptime. Better compliance.”

Data quality sets the ceiling

Wurster said even a well-designed model will fail if its data is incomplete, inaccurate or stripped of context. Fleets should convert paper records and informal knowledge into machine-readable information, including shop notes and details that otherwise remain in unrecorded phone calls.

“If you are not recording the information in a way that is computer readable, you are not going to be able to use stuff … today to go make your systems better tomorrow,” he said.

Galland also urged fleets to press technology providers for open data architectures and real-time access. Some systems were not designed to make every data point readily available, limiting a fleet’s ability to combine information across platforms.

Wurster recommended starting with tightly defined projects that can produce measurable results, such as document scanning, automating a specific report or deploying an established commercial application instead of building a model from scratch.

Both panelists also cautioned fleets to scrutinize how vendors store, share and use sensitive information. Wurster warned that an internal chatbot connected to multiple databases could expose payroll, human resources and other personally identifiable information to unauthorized employees. Galland said fleets should seek safeguards such as zero data retention and private model hosting when appropriate. He also advised fleets to require technology providers to study their actual workflows before proposing a solution.

“Do not be obsessed with the technicalities of what AI is,” he said. “Be really obsessed with value, be obsessed with risks, and be expert in your implementations.”