Your TMS vendor has just walked you through next year’s roadmap. There is a copilot that summarizes exceptions, smarter carrier and rate suggestions, automated invoice matching, and a promise of agents later in the year. It is a good presentation, and some of it will help. Then someone on your leadership team asks the question every leadership team is asking now: what is our AI strategy? A slide from that presentation quietly becomes the answer.
A roadmap tells you what a product will do. A strategy tells you what your company will do.
That is a category error, and an easy one to make. A roadmap tells you what a product will do. A strategy tells you what your company will do. This article is about the gap between the two, and about the AI use cases that live in the gap.
What your TMS roadmap really is
A TMS roadmap is the vendor’s plan for its median customer. It is built to be sold to every shipper, 3PL, and broker on the platform, which means its features reach your competitors on roughly the same schedule they reach you. That is fine for a product. It is not a source of advantage, and it is not a plan.
The market is also moving fast and unevenly. Supply Chain 24/7’s TMS trends piece, published in January 2026, quotes a Gartner director analyst saying that most major TMS vendors introduced generative AI features in 2025, and that knowing where AI actually creates value in an operation remains a challenge. The same piece quotes a systems project manager at St. Onge Co. saying that many TMS providers claim to be AI, but that the claim is “still a bit of a stretch in some areas.” We read that as healthy skepticism about feature claims, not as a knock on the vendors. Feature roadmaps are supposed to be incremental. The mistake is asking one to do the job of a strategy.
The right way to run a roadmap is as vendor management. Ask which features are generally available and which are planned. Ask what data each one needs and where that data has to come from. Ask what the feature does in your environment on the day your data is not ready. Ask how its decisions are logged and how you reverse one. Then give the roadmap an owner, probably your TMS product owner, and review it every quarter.
Why the economics point the same way
There is a second reason not to let the vendor’s roadmap stand in for strategy. Our 2026 AI Dividend Map puts logistics and supply chain at the Falling Floor edge, the most exposed sector on the map. Matching loads to capacity is specifiable, rate-selected, and network-driven, which is the recipe for a Falling Floor: efficiency gains that are available to everyone, and price pressure that does not let up. The same map places network design and complex orchestration on the Rising Bar.
Read that as a sorting rule. The AI that arrives inside everyone’s TMS mostly makes the matching and execution work cheaper for everyone. The room to differentiate sits in how well you orchestrate across the network of customers, carriers, providers, and systems around you. Features inside a product belong to the first group. Orchestration belongs to the second.
A simple test for telling them apart
Take any AI initiative on your list and ask whether your TMS vendor could deliver it by shipping a release. If it could, the initiative belongs on the roadmap, and managing it is a vendor and change-management job. If it needs data from your TMS plus your ERP, your warehouse system, your dock or yard scheduling, your carriers’ systems, or your customers’ systems, no release will deliver it, because no single vendor owns all of the data. That second group is your AI strategy.
Where the hard use cases live
Here are illustrative examples, not a measured list. Detention and dwell charges have causes spread across a carrier’s driver, your dock schedule, the warehouse labor plan, and the order release in your ERP, so reducing them takes data from all four. An honest promise date depends on inventory position, warehouse capacity, transportation capacity, and carrier performance, which sit in different systems under different owners. Freight audit and payment matches carrier invoices against contracts, accessorials, and proof of delivery across your TMS, your ERP, and the carrier’s billing system. A late load is a problem shared by shipper, carrier, and consignee, and the useful decision depends on what each one knows.
Then there are the cross-enterprise cases, where the other party has its own systems and its own incentives: sharing forecasts with carriers so they can commit capacity, or letting a customer’s software ask your systems for a quote or a shipment status. We wrote about that last pattern in Your Next Customer Won’t Be Human. None of these live inside a single product, and a roadmap cannot reach any of them.
Why the data decides it
The 31st annual Third-Party Logistics Study, developed with NTT DATA, Penske Logistics, the University of Tennessee’s Global Supply Chain Institute, and the Council of Supply Chain Management Professionals, was covered in October 2026 by Distribution Strategy Group. It names data quality and availability as the leading obstacle to AI, cited by 61 percent of respondents, followed by legacy systems and integration complexity at 54 percent. The coverage notes that AI is being applied across transportation, warehouse, and other applications that were not necessarily designed to exchange information, and that the problem gets harder when decisions depend on information shared among suppliers, carriers, and logistics providers. The coverage does not state the sample size or fieldwork dates, so we treat the percentages as directional.
Alpega, a transportation software vendor, published a Transport Leadership Benchmark in May 2026, a survey with Adelante SCM of nearly 300 supply chain and logistics executives and carriers in Europe. According to commentary from an Alpega executive, only 27 percent of shippers consider themselves well integrated with their logistics partners, against 65 percent of carriers and logistics providers, and half of shippers say the main barrier to faster decisions is a lack of complete, up-to-date data. This is vendor research about Europe, and the commentary does not say whether both groups answered the same question. We use it for the pattern more than the numbers: the two sides of a shipment do not see the connection between them the same way.
Gartner’s guidance points in the same direction. In May 2025 it predicted that by 2030, half of cross-functional supply chain management solutions will use intelligent agents to autonomously execute decisions across the ecosystem. Treat the 2030 number as a prediction with a date on it. The more useful part of the release is a caution that cross-functional solutions reflect the complexity of the supply chain and work against keeping the technology stack flexible, together with advice to start with use cases where the data is high quality and accessible, the behavior is verifiable, and the return is high. That is advice about strategy, not a feature list.
Your providers may be ahead of you
The same third-party logistics study reports that 53 percent of 3PLs have implemented autonomous or agent-based decision support, against 18 percent of shippers, and that 94 percent of 3PLs believe their customers are satisfied with their IT capabilities while 70 percent of shippers say they are. Both sets of figures are self-reported, and the sample is not stated, so read them as signals and not measurements.
A shipper with no AI strategy of its own tends to inherit whichever one its providers wrote.
If the signal is right, the AI making decisions about your freight increasingly lives in your providers’ systems and is built around their priorities. That is our reading, not the study’s. A shipper with no AI strategy of its own tends to inherit whichever one its providers wrote.
Keep two documents
Write the roadmap review and the AI strategy as separate documents with separate owners. The roadmap document is short: which vendor features you will adopt, in what order, with what testing. The strategy document picks two or three cross-system use cases and, for each one, names the decision being improved, the data it needs, who owns each source of that data, what agreement you need with partners to use it, what the AI may do on its own and what needs a person, and how every action will be logged. The logging matters more as agents act across systems, and we described the audit problem in Mitigating the AI Agent Audit Gap Under SOC-2. The idea behind the lanes is in Governance is a Freeway, Not a Roadblock: a few clearly marked boundaries let teams move faster inside them.
Do not wait for the strategy before using the roadmap. Features inside a single system, from a single vendor, with a single owner, are the lowest-risk place to start, and they build the habits you will need later. Also do not let that work stand in for the strategy. The data and partner work takes longer than any release cycle, so start it in parallel. If you want a read on where your data foundation stands, our 2026 logistics analytics maturity survey is a useful comparison point, and Logistics in 2040 lays out the larger scenarios the strategy has to survive.
What this will not do
Separating the two documents does not make the data problem easier. A strategy paper without data agreements is a slide. Some shippers, especially smaller ones, should lean heavily on their vendor’s roadmap and keep the strategy short. Cross-enterprise use cases also depend on partners who may not want to share, or may not be able to, and no framework changes that. Finally, most of the evidence above is survey research with unstated or limited methods, plus vendor commentary. It supports the direction of the argument. It does not prove the size of the gap in your company.
What it gives you is a clear line between the work you buy and the work you own.
What it gives you is a clear line between the work you buy and the work you own. If you want help drawing that line for your own operation, an independent assessment is a good place to start, and our logistics IT and AI consulting practice works on exactly this kind of cross-system planning. Either way, take your TMS roadmap and mark every item that needs data you do not yet govern. That list is the start of your AI strategy.
