Fleets Should Build AI Strategy Around Data and Value

Estimated reading time: 5 minutes · Last updated: 24 September 2026

Fleet operators should base any fleet AI strategy on clean, machine-readable data, modern system architecture and narrowly defined business use cases, speakers at The AI Summit said. At the Technology & Maintenance Council session on Sept. 22, Hans Galland of BeyondTrucks and Chas Wurster of PrePass urged fleets to prioritise value over technology novelty and to fix fragmented data before buying models. Galland cited a first-quarter carrier survey finding that about 75% of fleets lacked a formal approach to AI even though more than half were already using some form of the technology. The immediate payoff, the panelists said, comes from projects that are frequent, measurable and low risk.

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

Hans Galland, BeyondTrucks founder and CEO

Start with data and architecture, not models

Hans Galland and Chas Wurster opened the session with the same prescription: resolve data fragmentation before buying or building models. Many fleets still rely on legacy systems that require manual entry and do not capture continuous time- and location-based operating data, which leaves models starved of the inputs they need to work reliably.

Cleaning paper records and converting informal notes into structured, machine-readable formats is an operational task, not an IT novelty. Galland said systems must be able to feed AI the data it needs; without that foundation, even well-designed models will produce unreliable outputs.

This section focuses on practical steps: identify the most valuable datasets (telematics, sensor logs, shop notes), stop any processes that strip context from records, and demand open data access from vendors so a fleet can combine information across platforms. The single phrase to keep in mind is clean, machine-readable data.

Match the tool to the task: three categories of fleet AI

Panelists grouped fleet AI into three categories to help buyers choose the right approach for each problem. They named automation (for repetitive tasks and document processing), decision support (for anomaly detection, predictive maintenance and safety alerts) and generative AI (for producing text, code or configurations).

Using these categories makes procurement simpler: automation projects often require only integration and an OCR or RPA layer; decision-support work leans on sensor and telematics streams plus engineered models; generative systems demand careful constraints and human oversight because they are probabilistic by design.

Wurster emphasised that many vendors already ship narrow, task-specific AI inside their products, so fleets may get gains by pressing existing providers for features rather than attempting to build broad models in-house. The section repeats this organising idea as Automation, decision support and generative AI.

Prioritise projects by value and frequency

Galland recommended a simple prioritisation rule: weigh a task’s value against its frequency to identify where AI will move the needle. High-frequency, low-value chores such as repetitive check calls are natural automation candidates because small per-instance savings compound quickly.

Conversely, high-value but infrequent choices—network design or strategic capacity planning—may justify more advanced analytics even if they run less often. The panel also flagged low-probability events with severe consequences, like fatigue-related crashes, as worthy of safety-focused AI despite their rarity.

The practical takeaway for fleet managers is to scope pilots narrowly so results are measurable: scan documents, automate a single report or deploy an established commercial application before attempting custom model builds. This approach reduces implementation risk and shortens the feedback loop.

Trust, privacy and vendor obligations

A recurring theme was trust: modern machine learning and generative models are probabilistic and can return confident but incorrect answers, a point Galland stressed when he said adoption has shifted to a trust problem. That characteristic makes vendor selection and contractual safeguards central to any deployment.

Wurster warned that internal chatbots or multi-database integrations can accidentally expose payroll, HR records and other personally identifiable information unless access controls and audit logs are in place. Both panelists advised fleets to insist on technical options such as zero data retention and private model hosting where appropriate.

Beyond contractual clauses, the panel urged technology vendors to study a fleet’s actual workflows before proposing solutions so implementations match operational reality. The phrase fleets should require zero data retention and private model hosting summarises the recommended protections.

How AI adoption could play out for fleets

The bottom line

Fleets that treat AI as an operational change rather than a plug-and-play product will see the best results. The immediate work is not model selection but fixing data flows, insisting on open architectures and running tightly scoped pilots that produce measurable outcomes. Vendors already deliver narrow AI inside many products, so fleets can realise early gains by demanding real-time access to their own data and by contracting for clear privacy guarantees. As Galland and Wurster argued at the AI Summit, success rests on matching the right tool to a defined task and managing the risks that probabilistic models introduce.

What to watch

• Watch whether fleets that begin pilots publish formal AI policies or governance; no date has been set.
• Watch vendor responses to requests for open data architectures and real-time access; no date has been set.
• Watch deployment outcomes from early predictive maintenance pilots that combine telematics and shop-note data; no date has been set.