The AI Strategy That Dies in a Spreadsheet

Some time ago I sat in a meeting where a freight forwarding company presented its digital transformation plan, and I remember it well because the plan itself was decent: AI-supported quoting, predicted delivery times, automated documentation, eighteen months of work and a serious budget. The problem appeared with one simple question from the room: where exactly will we get the historical shipment data that all of this is supposed to run on?

Nobody had a good answer. The data sat in three different TMS systems left over from two acquisitions, part of it in a customs broker’s portal the company didn't fully control, and the knowledge of the most profitable trade lane lived mainly in the head of an experienced colleague who was retiring in a few months. So the company had a strategy for artificial intelligence, but no strategy for its own information, and in my experience this is not a rare case. I would rather say it's the norm in our industry; most companies just haven't been forced to admit it.

What the numbers say and what I see behind them

I read a fair amount of industry research, and this year the picture is quite consistent: depending on the study, only 10 to 30 percent of logistics processes use AI in any real way, fewer than one company in six has rolled it out across the whole organization, and when managers are asked what blocks them, the most common answer concerns neither the cost of the technology nor the quality of the models, but old systems that cannot exchange data with each other.

I find this interesting because, over the same period, the tools have become genuinely good. DHL now predicts vessel arrivals in near real time by combining machine learning with GPS and AIS signals and carrier history, and an operations manager can ask a chat for a shipment comparison or a report, something that five years ago required an analyst and half a day. When I put these two facts together, very capable technology and very low actual adoption, I come to the conclusion that our industry no longer has a technology problem; it has a homework problem. The companies that spent the last two years quietly putting their data in order are now moving from pilots to production, while the companies that assumed AI is something you simply buy are learning, often at a high price, that a model working on incomplete and outdated data gives incomplete and outdated answers, only faster and with more confidence.

Forwarding has always lived off scattered information

Anyone who has worked a forwarding desk knows why this hits us harder than most industries. One FCL shipment involves the shipper, the trucker, the terminal, the carrier, customs and the consignee, plus agents on both sides, and each of them keeps its piece of the truth in its own way: EDI here, a carrier portal there, a PDF in an email somewhere else, and always an Excel file somewhere. The whole forwarding profession grew out of the ability to collect these scattered pieces and turn them into a promise the customer can rely on, and for decades people who knew where to look and whom to call did that.

Now we expect AI to take over the collecting, and I believe it eventually will, but most sales presentations stay silent about one condition: the system can only work with what it can see. If your rates live in email attachments and your shipment statuses arrive with a two-day delay, then no algorithm will quote faster than your competitor or warn your customer about a delay before their production line stops. I stared at the same wall for years in the energy sector, by the way, where data about wind and solar farms, grid information, and sustainability reporting that regulators have actually started checking collide with records created to be filed away in a binder, not to be questioned.

What I would advise

The companies that succeeded did an unfashionable thing first: before spending money on intelligence, they spent it on plumbing. I do not mean a five-year data warehouse project, because such projects have deservedly ruined the reputation of this work, but a practical platform that brings the TMS, the carrier data, the finance system and all those Excel files into one place where the information is current and where people can ask questions of it. They also gave the whole effort an owner with a real reason to care, because in my experience an AI initiative that belongs solely to the IT department stays a pilot forever.

So if you are starting to think about your 2027 budget, my advice is this: before you approve another AI pilot, pay for an audit of where your shipment, rate and customer data really lives, how many hours or days old it is when it reaches you, and who in the company can actually make use of it. If the answer includes a spreadsheet on someone’s desktop or a colleague whose knowledge will retire together with him, then you already know where the real work is. It looks poor on slides, but I am convinced that this is exactly what will separate the companies where AI earns its keep from the companies that will still be running pilots in 2030.

Grzegorz Hinc spent his career in freight forwarding and logistics and in the green energy sector, and today he runs Cadvel, a company that designs, builds and maintains data platforms for logistics and energy. If this sounds familiar, we will calculate what this problem costs your company.