Where are we with artificial intelligence in the supply chain? What is real?

As 2025 draws to a close, where do we stand with artificial intelligence in the supply chain? What is real? What is hype?

What AI is Really Working in Supply Chain Software Vendor Solutions?

Machine learning has been part of advanced demand forecasting for over 20 years. But these solutions have become more powerful, even creating daily forecasts at the SKU level. Demand management solutions that use machine learning perform better than solutions that don’t.

In supply planning, optimization is critical. Optimization, another solution used by supply chain software vendors for more than two decades, is now considered a form of artificial intelligence. Optimization works, delivers real ROI, and has improved over the years.

So artificial intelligence in supply chain management is real and has been for a long time. However, the places where AI is used, and the types of AI used, continue to expand.

Over the years, for example, optimization has been used in new ways. Warehouse management is a prime example. WMS was once considered an execution solution. More advanced WMS solutions now use optimization to improve how orders are dropped into the fulfillment queue, which crew members will then work on in other areas as well.

A prediction is a prediction. ML predictions are now combined with optimization to improve programming. Optimal Dynamics offers an innovative transport solution for transport companies. While most routing solutions optimize only after a set of moves has been committed, Optimal Dynamics takes a fundamentally different approach. Their platform enables carriers to assess whether to accept a load prior to commitment — based on a prediction of what is likely to happen across their entire network. I spoke with one of their customers who praised this solution.

Working standards in a warehouse have a good return on investment (ROI). Historically, establishing and maintaining these standards required significant effort. AI-based work/warehouse management solutions can do this with much less effort. While AI-based standards are easier to define and offer a positive ROI, standards defined in the more traditional, labor-intensive way are more expensive and have an even better ROI.

Real-time risk management solutions powered by Big Data and AI are nothing short of amazing. However, it is not enough to simply receive a notification. a company must develop the capabilities to respond to them in a timely and effective manner. Companies that receive critical alerts and respond promptly have a competitive advantage. These solutions are imperative for running a company direct supply chain.

Some risk management vendors can use artificial intelligence to help map a company’s extended supply chain. The mapping is not 100% accurate, but it greatly speeds up the mapping process.

AI-based invoice management solutions can classify goods more accurately than humans.

Supply chain software companies have used Generative AI to improve the documentation and ease of use of their solutions.

AI-based solutions can improve training and recruitment in the supply chain between partners. AI can play a role in how companies hire, how people experience work after being hired, and in education. Smart tools can help companies personalize new partner onboarding. AI recommends learning paths that accelerate skill development.

Parenthetically, when it comes to hiring new managers and designers, AI makes it harder for the hiring manager. It used to be that a hiring manager could look at a resume, see typos and grammatical errors, and infer something about the prospect’s ability. Additionally, the content of resumes could be a gauge of a prospect’s depth of knowledge about key supply chain concepts. With ChatGPT being used to build resumes, those days are over.

What about autonomous supply chain? This involves taking humans out of the loop and letting the machine handle the programming. This happens in a very limited way. In the center-to-store retail shelf supply chain, there are a few examples of this. But it does not happen in larger supply chains that include factories as supply chain hubs.

What I can’t verify

The best way to verify a seller’s claims is to talk to their customers. For a few years now I have been asking for customer reports on certain claims.

The saying goes, garbage in, garbage out. Newer solutions can use artificial intelligence to clean their data and correct key parameters. However, this is talked about more as a feature than something I’ve heard the vendors’ customers talk about as a key benefit of their chosen solution or a place where I’ve seen a convincing demonstration.

There’s also the “black box” issue – solutions that spit out answers that people can’t make sense of. This problem has been discussed for years. For years, vendors have claimed to have solved it. Generative AI is also being touted as a solution to this problem. I have yet to verify that this feature exists in any meaningful way. Certainly, no customer has ever discussed this with me.

What about eliminating the barrier between planning and execution? For years, businesses have struggled with a fundamental disconnect between planning and execution. Demand forecasts, replenishment strategies, and inventory allocations often fail to align with actual warehouse and transportation network constraints. The result? Unrealistic plans, operational bottlenecks and costly inefficiencies. In theory, newer solutions remove these operational silos by enabling two-way collaboration between planning and execution systems—ensuring that supply chain decisions are not only optimized but also realistic, achievable and responsive to real-time conditions.

I have not been able to validate these capabilities by speaking with the vendor’s customers. But this is a new solution based on agentic AI. Manhattan Associates says it has clients in beta applications. I believe Blue Yonder has made a similar claim. I hope to speak with a reference client of Manhattan or Blue Yonder next year.

The Hardware/Software AI Nexus

Some machines, such as robotics, combine hardware with artificial intelligence-based software. But the software is the key to the advanced capabilities of the equipment. Warehouse robotics is in this field. The ability to navigate a warehouse is based on artificial intelligence. This is a mature technology that provides a strong ROI.

Artificial intelligence can be used for preventive maintenance and to predict that a piece of equipment is likely to fail within the next days or weeks. For manufacturing plants with critical pieces of equipment, this can help avoid production bottlenecks and disruptions. I’m disappointed, however, that I haven’t seen these equipment alerts integrated seamlessly into programming.

Artificial intelligence is used in conjunction with telematics and cameras to improve truck safety. A trucking company I spoke with recently reduced avoidable accidents by 30%, saw an 83% drop in workers’ compensation claims, achieved a 40% reduction in manual paperwork, and saved $730,000 in fuel costs each year. It is worth noting that this advanced AI solution will become obsolete when autonomous trucking becomes ubiquitous.

Nothing will transform logistics like autonomous trucking. But when will autonomous trucking become ubiquitous? In May, Aurora Innovation, Inc. (NASDAQ: AUR) announced that it has successfully launched its fully autonomous self-driving truck service on the Dallas/Houston corridor. TORC hints that it may have these capabilities next year.

But the mood was frustratingly slow, especially compared to autonomous taxis. For the next several years, disposition will occur only in the southwest, where snow and rain are rare. I understand that. But even so, this solution has been scaled up much more slowly than expected. I’ve bugged Aurora and Torc about what’s slowing things down. It is not the artificial intelligence that is used to map and navigate a new lane. Apparently, this can be done within about 6 months. So what is this? Finances not good? Is the onboarding process for new customers more excruciating than it seems? Are customers worried about litigation issues? I can’t get a good answer or talk to a reference customer.

What about Roll Your Own Solutions?

What if, instead of going to a supply chain vendor, a company goes to an AI platform provider and develops its own solutions? Numerous references show that a significant number of companies are seeing poor returns on investment from their AI initiatives so far. A widely cited MIT 2025 report, for example, found that 95% of business AI pilots failed to deliver a measurable return on investment (ROI), despite billions of dollars in investment. These are platform investments.

But finally, finally, we have a company that has succeeded in this area. They are also the first company I have heard of success with their investments in Agentic AI. CH Robinson Worldwide (NASDAQ: CHRW ), a global logistics provider, has built a solution that dramatically improves its ability to provide freight quotes to customers. Agentic AI helps them deliver both more offers and higher quality offers. Quantifying the gains from AI has been difficult for them because it is difficult to separate the contribution of this technology from its lean operating model. However, their best estimate is that lean travel has delivered single-digit productivity improvements, while the addition of agentic AI has allowed them to target double-digit gains in 2026.