Supply Chain AI Has a Budget Problem: Why Gartner Says Leaders Are Running Two Timelines at Once

Supply chain AI is no longer a side experiment. It is in the budget, in the boardroom, and in the operating plan. That should be good news. But Gartner’s latest warning is uncomfortable: many supply chain leaders are being asked to fund tomorrow’s AI operating model while delivering today’s service, cost, and resilience targets.

Logistics Management reports that Lindsay Azim, Senior Director Analyst in Gartner’s Supply Chain and Procurement Management practice, framed the issue at the Gartner Supply Chain Symposium as “juggling two timelines at once.” Leaders must keep goods moving and costs under control now, while preparing for an AI-shaped future that is still forming.

The budget pressure is real. According to the same report, supply chain organizations spent an average of $24 million on AI in 2025, yet many projects have gone over budget and are not expected to produce results for at least a year. PwC’s 2026 operations survey makes the same governance point from the execution side: companies that integrate AI, data, and operating model transformation are positioned to redefine operational performance, which raises the stakes for governance now.

That is not an argument against AI. It is an argument against vagueness.

The AI budget is outrunning the operating model

The supply chain industry has seen this movie before. A new technology category arrives with enormous promise, executives demand speed, teams launch pilots, vendors show impressive demos, and then the hard work appears: master data, integration, process ownership, exception rules, user training, change management, and measurable ROI.

AI makes that pattern more dangerous because the language is bigger. “Transformation” can hide a weak business case. “Autonomy” can hide unclear accountability. “Copilot” can hide a workflow that still depends on three spreadsheets and a heroic planner.

A $24 million spend is enterprise-scale capital. If projects are over budget and still a year from results, leaders need a sharper governance model before the next funding cycle turns into AI fatigue.

The right question is not “Where can we use AI?” That question produces a graveyard of pilots. The better question is “Which operational decisions are expensive, frequent, data-rich, and measurable enough to improve with AI?”

Supply chains fail in specific loops: late supplier signals, inaccurate inventory positions, missed pickup risk, carrier rebooking delays, document exceptions, customer expedite requests, and margin-blind routing decisions.

Start there. Not with theater.

Governance starts with kill criteria

Most AI governance conversations jump immediately to ethics, risk, and security. Those matter. But logistics teams also need a more basic discipline: knowing when to stop.

Every AI pilot should begin with explicit kill criteria. If the use case does not improve a defined metric by a defined date, it should be paused, redesigned, or shut down. No zombie pilots. No “strategic learning journey” that keeps consuming budget because nobody wants to admit the workflow is not ready.

Good kill criteria are operational, not cosmetic: reduce manual exception triage time by 30% within 90 days, improve appointment scheduling throughput without increasing missed dock windows, cut preventable detention exposure, or reduce customer-status update latency for delayed shipments.

The metric should connect directly to cost, service, working capital, risk, or revenue protection. If the metric is “number of AI-generated recommendations,” stop immediately. That is dashboard vanity wearing a lab coat.

Data readiness is not optional plumbing

Inbound Logistics argues that AI readiness depends on five critical areas: data, technology, people, ethics, and security. Its first point is the one logistics teams keep relearning the hard way: no AI model can outperform the quality of the data it learns from. Supply chains may be data-rich, but that data is often scattered across incompatible systems, suppliers, and regions.

That is why AI governance needs data gates before production. A use case should not advance because a model demo looked clever. It should advance because the required data is available, timely, consistent, and connected to the workflow where decisions actually happen.

For freight forwarders, those gates are practical. Are shipment milestones normalized across carriers? Are rates and accessorial rules structured? Are documents tied to the shipment record? Are customer commitments visible to operations? Are exception codes consistent enough to evaluate recommendations?

If the answer is no, AI will not fix the process. It will accelerate the confusion.

Human-in-the-loop is a business design choice

AI adoption is moving toward decision support, but the market is not ready to hand over the steering wheel. SupplyChainBrain reports that 67% of retail and manufacturing leaders say their confidence in using AI for supply chain decision-making has increased compared with last year. But the same report found that 54% prefer AI to make recommendations while humans finalize decisions, and only 10% would trust AI to make fully independent supply chain decisions.

That split is healthy. Logistics decisions carry commercial and customer consequences. An AI system might identify a cheaper routing option, but a human may know the customer will not tolerate another transshipment. A model might recommend rebooking a carrier, but operations may know the lane is about to tighten. A tool might prioritize cost reduction, while the account team needs service recovery.

Human-in-the-loop is not a compliance slogan. It is how supply chains preserve judgment while improving speed.

The governance model should define decision rights by risk tier. Low-risk, high-volume actions can be automated inside approved thresholds. Medium-risk decisions should be recommended with explanations and one-click execution. High-risk decisions—margin changes, strategic customer commitments, customs exposure, hazmat handling, premium freight, or service-level exceptions—should require human approval and an audit trail.

Build two timelines without splitting the company

Gartner’s “two timelines” framing is useful because it explains the real leadership challenge. Teams cannot stop operating while they modernize. Freight still has to move. Orders still have to ship. Customers still want answers. Finance still wants cost control.

So the AI roadmap has to serve both timelines at once.

For the current timeline, prioritize workflows where cycle time, exception cost, and service impact are measurable in weeks, not years. Shipment exception triage, customer status automation, carrier selection support, document completeness checks, and appointment scheduling are strong candidates because they sit close to daily execution.

For the future timeline, invest in the foundations that make larger AI models useful: unified shipment data, clean customer and lane master data, integration with carrier and warehouse events, role-based workflows, governance logs, and consistent exception taxonomy.

The mistake is treating these as separate programs. Quick wins without architecture create another mess. Architecture without quick wins loses executive patience. The best supply chain AI programs use today’s operational use cases to fund and validate tomorrow’s data foundation.

What this means for freight forwarders

Forwarders do not need a moonshot AI strategy. They need governed decision support in the parts of the business where friction is already expensive.

That means starting with the workflows operators touch every day: delayed milestones, document gaps, carrier changes, customer notifications, spot quote decisions, delivery-risk alerts, and exception prioritization. Each workflow needs an owner, baseline metric, data-readiness checklist, approval thresholds, and post-deployment review.

AI budget discipline is not anti-innovation. It is what keeps innovation from becoming a very expensive fog machine.

CXTMS gives freight teams the connected execution layer needed for practical AI governance: shipment records, carrier coordination, customer communication, documentation, exception workflows, and analytics in one operating environment. If your team wants AI investment tied to measurable freight outcomes instead of vague transformation slides, schedule a CXTMS demo.