As a logistics consultant, it is critical to stay close to what is happening across log-tech. Given the sheer number of vendors we monitor, I enlisted the help of ChatGPT Tasks to run a daily scan and notify me of new product announcements and release notes across many different providers.
In one recent announcement, a well-known log-tech provider introduced what it described as a “new AI feature.” As I was reading through the details, the functionality started to sound familiar. AI's summary of the feature was this: when a predefined condition occurred, the system initiated a prescribed response. Underneath the AI label, or what I refer to as “AI-washing,” it appeared to be rules-based automation.
Nobody can argue that automation is one of the primary ways technologies create value, and value creation should be the priority of any technology. The issue is what the AI label implies about the feature’s capabilities. It was not learning, adapting, or adjusting; a configured trigger was driving a configured response. This is rule-based automation.
I am particularly sensitive to that distinction because I have been on the other side of it, having served as a log-tech product leader. In technology, you know a hype cycle is beginning when specific questions start appearing in RFPs. At the beginning of the AI hype cycle approximately 4 years ago, I was asked by the sales team to answer which AI capabilities were native to our platform and what we had planned for the roadmap.
We had identified use cases where AI might be the right approach, but we were roadmapping business problems, not adding AI for the sake of adding AI. When I drafted our response, I referred to our optimization engine, which has been native to the platform for more than 20 years. After all, it was sophisticated and used heuristics to solve complex problems. I had our development lead review my response, and he pushed back immediately. The optimizer was not learning or adapting. It was mathematical optimization, not AI, and we could not present it otherwise.
I think about that response often because I now see the opposite happening across the market. Established automation and optimization are increasingly being repositioned as AI, even when the underlying capability has changed very little. Sometimes AI genuinely improves an existing capability. Other times, terminology has changed more than the technology itself.
That distinction becomes more important when AI capabilities influence how buyers assess whether a technology is innovative, future-ready, or worth the investment. Logistics leaders do not need to understand how every model is built, but they do need enough AI literacy to determine what capability exists, what role it will play, and whether its requirements align with what their organization is prepared to support.
Log-Tech Was Already Sophisticated
The reality is, log-tech has automated work and supported complex decisions for decades.
Rules-based automation executes logic that humans define. When a condition is met, the system follows the corresponding instruction. The workflow can be complex and may automate a substantial portion of a process, but the technology is still operating within predefined logic.
Optimization also has a long history in TMS and WMS platforms. A TMS optimizer might evaluate how shipments should be consolidated and routed while balancing cost, service, and operational constraints. It can consider far more possibilities than a planner could reasonably assess, which makes the output rightfully defined as intelligent.
Traditional optimization may use mathematical programming or heuristics to reach that answer. A heuristic is particularly useful when the number of possible solutions is so large that finding the mathematically optimal answer would take too long. It gives the system a practical way to produce a strong, feasible plan within a designated amount of time.
That capability can be extremely valuable without being AI. It does not need to learn from previous plans or interpret unstructured information. It is solving the problem it was designed and configured to solve.
The same applies to automation. A system can respond to conditions and complete work without human involvement while remaining entirely rules-based. Autonomy within a narrow, predefined workflow is not the same as learning or independent reasoning.
This distinction does not diminish the value of either capability. I would rather have proven automation or optimization that solves an operational problem than an AI feature that does not. Accurate language simply helps establish the right expectations.
The Different Roles AI Can Play
- Machine Learning (ML) identifies patterns in historical data and applies them to new situations. In logistics, it might improve an ETA or predict whether a carrier will accept a load. Maintaining that capability requires monitoring the underlying data, prediction accuracy, and whether changing conditions are causing the model’s performance to deteriorate.
- Generative AI (GenAI) interprets and generates content. It might explain why transportation costs increased or summarize the factors contributing to an exception. Its performance depends on the model, the instructions it receives, and the information it can access. Those elements must be maintained and tested as the underlying business context changes.
- Agentic AI describes a capability that works toward an objective across multiple steps and may act. An agent might investigate an exception, gather information, determine an appropriate response, and execute parts of the process. Maintaining it extends beyond the model itself to its system access, permissions, instructions, integrations, escalation points, and performance controls.
These capabilities can work together, alongside technology that is not AI. An agent could use an ML prediction to identify a potential delay, call an optimizer to compare possible responses, and rely on predefined workflow rules to determine whether the selected response requires approval.
Calling that entire process AI hides the role each component plays. That matters because the components do not have the same dependencies, failure points, or maintenance requirements. Understanding the technology therefore requires following the work across the process rather than treating AI as a single capability.
Follow the Work, Not the Label
Consider a shipment that is likely to arrive late. Technology might predict the delay, recommend how a planner should respond, or act on that recommendation by securing different capacity or updating the shipment plan.
Following the work means understanding not only what happens next, but also what produced the result. An ETA based only on historical shipment data is different from one that also considers the shipment’s current location and real-time operating conditions. The result might come from an ML model identifying patterns, an optimizer evaluating alternatives against defined objectives and constraints, or rules determining the appropriate response.
Each component affects the operation differently. A prediction gives the planner additional information but leaves the decision and response with that person. A recommendation reduces some of the decision effort, although someone may still need to account for customer context or an operational constraint the technology does not understand or cannot see.
Once the technology begins acting, the organization must determine what it is authorized to do, when it should stop, and when the work should return to a person.
Following the work also shows what the feature depends on and how it must be maintained. Predictions require reliable data and ongoing accuracy monitoring. Recommendations must continue to reflect the objectives and constraints shaping the decision. Features that act require appropriate system access, controls over how that access is used, and oversight of the resulting actions.
Without understanding the work behind a feature, logistics leaders cannot fully assess whether it solves the intended problem, what it will require from the organization, or whether the expected value can be sustained after implementation.
Why It Actually Matters
A fair evaluation requires understanding what the feature does, what produces the result, and what it will require once deployed. That includes its data and integration dependencies, security and access requirements, internal resource needs, usability, ongoing monitoring and maintenance, and the division of responsibility between the provider and the organization. These factors determine whether the capability demonstrated during the selection process will continue delivering value in practice.
AI can add meaningful value to logistics technology, just as automation and optimization have for decades. Buyers should be able to identify where that value is coming from and what will be required to sustain it before allowing the promise of AI to influence a technology decision.
If you are evaluating AI as part of a logistics technology investment and need help separating the claims from the underlying capability, reach out. This is exactly the type of evaluation we help clients work through within JBF Consulting’s Strategy Practice.
About the Author
Rachelle Butler is a Director, Strategy at JBF Consulting with more than 15 years of experience spanning logistics operations and technology. She partners with shippers to assess, design, and implement solutions that align operational needs with long-term business direction.
Rachelle’s background includes roles in product leadership, consulting, implementation, and post-deployment client success at e2open, BluJay Solutions, and LeanLogistics. She began her career in the United States Marine Corps, where she gained foundational experience in transportation coordination and logistics operations. Rachelle brings a practical, real-work approach to helping clients realize meaningful value from their operational investments.
FAQs
AI-washing refers to vendors labeling established capabilities — like rules-based automation or mathematical optimization — as "AI" without the underlying technology actually learning, adapting, or reasoning. The feature performs the same predefined logic it always has; only the marketing language has changed.
Rules-based automation executes predefined logic: when a specific condition is met, the system triggers a specific, configured response. It doesn't learn or adjust over time. True AI capabilities — like machine learning, generative AI, or agentic AI — identify patterns, generate content, or pursue objectives in ways that go beyond a fixed if-this-then-that structure.
Not necessarily. Traditional optimization uses mathematical programming or heuristics to solve complex problems, such as consolidating and routing shipments. It can be highly sophisticated and valuable without learning from data or interpreting unstructured information, which means it doesn't meet the definition of AI even though it produces intelligent-seeming results.
The three main types are: Machine Learning (ML), which identifies patterns in historical data to inform things like ETAs or carrier acceptance predictions; Generative AI (GenAI), which interprets and generates content such as cost explanations or exception summaries; and Agentic AI, which works toward an objective across multiple steps and can take action, such as investigating and resolving a shipment exception.
Buyers should "follow the work" rather than the AI label — understanding what actually produces the result (a prediction, a recommendation, or an autonomous action), what data and integrations it depends on, what security and access it requires, and who is responsible for ongoing monitoring and maintenance after deployment.
