Fleet AI Needs to Prove Its Margin Gains Without Missing Deliveries
A fleet plan can reduce operating costs on paper while making delivery commitments harder to meet. Fewer vehicles, tighter schedules or higher utilization may improve one measure and weaken another. For operators testing AI-generated plans, the question is whether the proposed savings survive contact with service requirements and conditions on the road.
Predictiv AI is using that comparison to evaluate Flockentis, its fleet-planning system. In written responses to The Supply Chainer about a proof of concept with a UK commercial fleet operator, Suman Pushparajah, the company’s CEO, outlined the measures being assessed and the role dispatchers retain in reviewing proposed plans.
Profitability and Service Need the Same Baseline
“We compare Flockentis-generated plans against the operator’s existing fleet operations, assessing operating margin, cost per delivery, vehicle utilisation and on-time delivery together. Our objective is to improve profitability and operational efficiency while maintaining service commitments,” Pushparajah said.
That approach puts the operator’s existing performance at the center of the evaluation. A lower cost per delivery would be an incomplete result if it came with more missed appointments. Higher vehicle utilization would likewise require scrutiny if tighter schedules left insufficient room for disruption. The response identifies the relevant measures but does not report their results. Predictiv AI did not provide a baseline margin, an improvement target, the pilot’s duration or the number of vehicles involved. Its stated objective should therefore be distinguished from a demonstrated operating gain.
Three Plans Still Require Local Judgment
Flockentis generates three alternative plans, according to Pushparajah. Dispatchers can compare costs, vehicle assignments and delivery schedules, review them against their knowledge of local conditions and request adjustments where those conditions differ from the planning data. Presenting alternatives makes the trade-offs available for review. It also leaves an important implementation question: how does a dispatcher’s correction become part of the plan that is ultimately executed?
The company’s answer confirms human oversight but does not explain the override mechanism, how adjustments are recorded or how conflicting constraints are resolved. Those details matter when the reason a plan will fail is known to the dispatcher but absent from the underlying data.
Disruption Tests the Data Behind the Plan
A related challenge emerged in a separate written response to The Supply Chainer from Hapag-Lloyd, supplied through its corporate communications team. The shipping company described the difficulty of turning individual AI applications into reliable tools across a business.
“AI readiness means having trusted data, connected systems, clear responsibilities and people who can turn insights into action. For shippers and logistics providers, that includes being able to assess changing conditions, compare scenarios and adjust plans quickly. AI cannot eliminate geopolitical uncertainty, and historical patterns may become less reliable during disruption. Its value lies in helping teams make better-informed decisions, with human expertise and accountability remaining essential.”
Hapag-Lloyd’s observations concern shipping rather than the Flockentis pilot. They nevertheless identify a common operating dependency: a recommendation is only as useful as the information and decision process supporting it. For fleet operators, the meaningful result is a plan that improves measured economics while preserving delivery performance. Establishing that requires comparable operating conditions, recorded dispatcher interventions and results across both cost and service. Three alternatives create a choice. The pilot still needs to establish which choice works.
