The AI in Your TMS Isn't the Point

I use AI every day. That is why this needs saying.

I use AI as a working tool every day. Writing, thinking through structure, coding, building. It does real work, well, and it has changed how I operate.

That is exactly why I want to be careful about what comes next.

The AI in your TMS is not the point. AI matters — enormously — but most of what TMS vendors are calling AI is not the same thing I am using. And even when it is, it is rarely what determines whether your transportation operation actually improves.

That distinction is being lost in the noise. So let me try to make it.

There Are Different Kinds of AI. Some of It Is Just Workflow Automation in a Tuxedo.

There are roughly four tiers of intelligence built into modern TMS platforms. They are not equal. They are not interchangeable. And they are not all what marketers mean when they say AI.

•     Rule-based automation.If A, then B. If load is over 10,000 lbs, route to dedicated. If invoice exceeds quoted rate by more than 5%, flag for review. This is not AI. It is software doing what software has always done. Useful. Often labeled AI in vendor materials anyway.

•    Machine learning and predictive analytics.The system learns from historical data to predict outcomes — ETA accuracy, carrier performance, spot rate ranges, detention risk. This is real AI, and it has been in TMS platforms for ten years. It is also where most vendors actually live when they say AI.

•     Generative AI.Large language models that produce structured output from unstructured input. Document parsing, natural-language queries against shipment data, drafting carrier communications, summarizing exception reports. New to TMS in the last 24 months. Real, but inconsistently implemented and rarely the centerpiece of what gets demoed.

•     Agentic AI.Systems that take autonomous action against goals — negotiating with carriers, rebooking failed loads, managing tender acceptance without human intervention. Almost no production TMS does this today, despite what marketing suggests. It is coming. It is not here.

When a TMS vendor's home page says "AI-powered," it almost always means tier two — predictive analytics that have been in the platform for years, freshly relabeled. Sometimes it means tier one — rules-based automation in a fancy wrapper. Occasionally it means tier three. It almost never means tier four, even when the marketing strongly implies it.

76% of buyers say AI is a key TMS evaluation criterion. Fewer than 30% can articulate the specific AI capabilities they need or how those capabilities map to their operational pain points. The gap between what is being asked for and what is being understood is doing real damage.

Does the Distinction Matter?

Yes. And no.

It matters because predictive analytics, generative AI, and agentic AI solve different problems with very different reliability profiles. Predictive ETA models are accurate enough to drive customer commitments. Generative AI can draft a carrier email that needs human review before sending. Agentic AI making autonomous booking decisions on a $2M shipper account is a risk profile most operators are not prepared for.

It matters because the cost-to-value math is different at every tier. Rules-based automation is cheap and high-leverage. Predictive analytics requires good data. Generative AI requires governance. Agentic AI requires trust that no vendor has yet earned.

And yet — the distinction also does not matter, because neither the buyer nor the operator should be evaluating TMS by AI tier. They should be evaluating by something else entirely.

"AI does not power solutions. AI powers tools. Solutions are powered by problem definition, workflow design, and disciplined implementation. Without those three, the smartest AI in the world is bolted onto a broken foundation."

The Foundation Problem

You do not build a house on clay. No matter how impressive the framing, the wiring, the kitchen — if the foundation moves, everything else is at risk.

You do not bake bread without yeast. You can use the finest flour, the most expensive oven, the most beautiful artisan technique. Without yeast, you get a flat, dense, technically-correct-but-fundamentally-wrong product.

AI is the kitchen and the oven. The yeast is foundational alignment. The clay is whether anyone in the operation actually understands the problem the technology is supposed to solve.

Every TMS implementation that fails — and 75% of them fail to deliver the projected ROI — fails at the foundation. Not at the AI layer. The buyer did not define the problem clearly. The vendor did not rationalize the workflow. The implementation team did not meter the solution to the discrete operational reality. The change management was an afterthought. The data was not clean enough for any AI tier to produce reliable output.

Then everyone blames the AI.

Where Vendors Are Missing the Point

This is the part where I have to be careful. PreShiftIQ works with vendors. Vendors are not the enemy. The vast majority of TMS providers I have audited are run by competent, well-intentioned people building real products that deliver real value when implemented well.

But the marketing pattern is undeniable. Vendor home pages lead with AI. Demo flows showcase AI. RFP responses bold the AI capabilities. Conference keynotes describe AI roadmaps. Series B pitch decks center on AI moats.

The conversation that almost never gets the same airtime is how the vendor identifies the operational problem before they configure their software. How the implementation team rationalizes the workflow before the AI gets pointed at it. How the customer success organization meters the deployment to the customer's discrete operational reality. How change management is structured for the people who actually use the system.

That conversation is unsexy. It does not raise valuations. It does not win analyst quadrants. It does not get retweeted. So it gets quietly underweighted in the marketing — and, more dangerously, in the actual product roadmap.

The result is buyers who chase capability instead of outcomes, vendors who compete on model sophistication instead of implementation rigor, and operations teams who are left with tools that do not fit their actual problems.

What Buyers Should Actually Ask

If you are evaluating TMS in 2026, ask the AI questions. They matter. But ask them after you have asked the questions that actually drive outcomes:

•     How does your team identify the foundational operational problem before configuration begins?

•     What is your methodology for rationalizing workflows that already exist — including the workarounds the operations team has built around legacy limitations?

•     Show me three implementations where you discovered something during discovery that changed your configuration plan.

•     How is your change management structured? Who owns it, on your side and ours?

•     If our data is messier than we said it was — and it will be — what is the recovery path?

•     How are AI features metered to specific operational outcomes? What does failure look like, and how do we know?

Those questions will not show up on the vendor's website. They will tell you everything about whether the AI in their product is going to help your operation or just decorate it.

The Quiet Truth

AI is not magic. AI powers solutions, but it does not create them. Foundational truths still have to be discovered case by case. Solutions still have to be metered against an enormous number of data points specific to your operation. Every transportation operation is unique. The AI in your TMS, no matter how sophisticated, can only amplify what you have already understood about your own business.

This is not a critique of AI. It is a correction of where AI sits in the value chain. AI is the engine. The chassis, the drivetrain, the road conditions, and the driver still matter. A Ferrari engine in a car with no wheels does not move. A modest engine in a well-built car moves reliably for years.

Most TMS conversations in 2026 are about the engine. The conversations that should be happening are about everything else.

Which Leads to a Question Buyers Keep Asking

If the AI in TMS platforms is overhyped, the foundation work is what actually drives outcomes, and most vendors are not having the conversation buyers actually need — there is a logical follow-up.

Should I just build it myself?

LLMs have lowered the floor on what a non-engineer can build. The CEO of a $40M shipper can prototype something that resembles a working TMS in a long weekend. The temptation is real. So is the trap.

The next paper takes that decision apart. Should I Build my own TMS

Sources: PreShiftIQ vendor audit data (80+ vendors). The 75% TMS implementation failure-to-projected-ROI figure is documented in PreShiftIQ research articles and corroborates Gartner and Forrester reporting. AI tier framework adapted from PreShiftIQ Vendor Audit Scorecard.