Why AI logistics software fails: bad container data, not bad models

What makes AI logistics software fail in production?

AI does not create accuracy, it amplifies whatever you feed it. If the underlying container data is incomplete, delayed, or conflicting, the AI confidently makes the wrong call, automates the wrong action, and scales the mistake across your entire operation. That diagnosis comes from Michael Rentz, Chief Revenue Officer of Gnosis Freight, a container tracking platform founded in 2017.

The freight technology industry has made AI the dominant selling point across virtually every logistics software category. Carriers, forwarders, and shippers are being pitched AI-powered dashboards, AI-driven ETAs, and AI-enabled workflow automation at a pace that has outrun the industry's ability to evaluate what any of it actually means in practice. Most companies are skipping the foundation and going straight to the model, Rentz said, and that is why so many AI pilots in supply chain look great in a demo and fall apart in production.

The gap between AI promise and AI reality in freight operations is, at its core, a data problem. True data readiness is rare. What Gnosis sees most often is organizations that have data, but it is fragmented across carrier portals, spreadsheets, freight forwarder emails, and legacy TMS systems with no common structure or timestamp logic. A lot of shippers are just stitching together three sources and hoping they agree, Rentz said.

What does operational-grade container data actually require?

Data readiness, as Rentz defines it, means a single, validated, real-time record of every container milestone that every team and every system works from simultaneously. By that standard, most shippers are still early in what he describes as a data sovereignty journey. The ones who have done the work to get there are the ones seeing real ROI from automation.

Gnosis Freight's answer to the data problem is the container tracking engine at the core of the platform. Rather than routing data through third-party aggregators, Gnosis establishes direct, first-party relationships with ocean carriers, ports, terminals, Class I railroads, AIS satellite feeds, and U.S. Customs. That first-party access is foundational to what separates Gnosis's approach from much of the rest of the market, Rentz said. Unlike a lot of providers that lean on third-party aggregators, we go directly to the source.

Raw ingestion is only half of it. The validation layer is where the real work happens. When sources conflict, Gnosis uses a smart hierarchy and contextual logic to resolve those conflicts rather than just displaying whatever came in last. A carrier API telling you one thing and a terminal feed telling you another does not surface as noise to the user. It gets resolved before it ever hits the platform.

Gnosis Freight's forward-deployed engineering model comes into play with that validation layer. A lot of people think we are just extracting milestone data, Rentz said. What we are actually extracting is operational knowledge. The most valuable logistics data often lives in the heads of the people managing exceptions every day. By embedding with customers and ecosystem partners, our teams capture that tribal knowledge. There is a lot of nuance in how a specific business interprets a milestone, handles an exception, or structures a workflow. No integration alone gets you that. It has to be built alongside the customer.

What happens when the data foundation is missing?

When the underlying data is not ready, the consequences rarely show up as a dramatic system failure. More often, the team slowly loses trust in the imperfect technology. It looks like a demurrage bill nobody saw coming, Rentz said. It looks like an ETA prediction that was off by four days and nobody caught it because the system said everything was fine. It looks like an automated workflow that triggered the wrong drayage pickup because a terminal update never made it into the system cleanly.

The failure mode is not dramatic. It is death by a thousand small errors that erode trust in the technology until the team stops using it and goes back to manual. That is the graveyard most AI logistics pilots end up in, and bad data is almost always the cause.

How accurate is Gnosis Freight's container tracking?

If anybody tells you they are 98.7% accurate without telling you accurate compared to what, they are not giving you the full picture, Rentz said. Accuracy is not a static number. It is a continuous journey. You should be improving completeness, latency, reliability, and operational context over time. That requires a strong data foundation, alignment across ecosystem partners, deep integrations, first-party access, embedded tribal knowledge, and continuous feedback loops on exceptions. That is the infrastructure question. The percentage is the easy part to talk about and the hardest part to actually earn.

That infrastructure is also what enables the platform's predictive capabilities. When you have that volume of clean, structured, real-time data flowing from that many sources, combined with the operational context our teams bring in, you can start generating predictive ETA milestones that are not based on what the carrier told you, but on what the data actually shows is happening across every touchpoint in that container's journey, Rentz said. That is not something you can buy off the shelf, and it is not something you can fake with a single feed.

What ROI are shippers seeing from validated container data?

One top-50 U.S. importer reported more than $12 million in demurrage and detention savings in under 12 months using the Gnosis platform. Customers report an average of 81% reduction in demurrage charges and 64% reduction in detention charges in their first year. Rentz traces those results back to the same infrastructure question. Once the data foundation is there, the opportunities compound fast.

The use cases range from EIR email capture that automatically reads inbound terminal emails, extracts gate-in and gate-out details, and files them against the right container without anyone touching it, to real-time demurrage and detention recalculation that updates your free-time risk exposure every time a milestone changes so surprise charges stop happening. The list extends to automated delivery order creation, drayage scheduling triggers, arrival notice processing, and freight invoice auditing against actual operational events. All of this is tied back to margin protection and not treated as automation for its own sake.

None of that is possible without the foundation that makes the underlying data trustworthy, Rentz said. AI is not the hard part. The hard part is building the infrastructure that makes AI trustworthy enough to act on. Once you have that, and once everything lives in one place, the possibilities for protecting margin compound quickly.

What questions should shippers ask AI logistics vendors?

At this point, the question is not whether or not your technology vendor offers an AI solution. The question is whether the data infrastructure behind it is capable of making that AI reliable enough to act on.

That means asking vendors where, specifically, their data actually comes from: which carriers, which terminals, which rail partners, and through what mechanism. It means asking how conflicting data from different sources gets resolved, and what happens when two feeds disagree. It means asking what accuracy is being measured against, and whether completeness and latency are held to the same standard. It also means asking what happens after implementation, like whether the vendor embeds with your operation to learn how your business interprets a milestone or handles an exception, or whether they hand over a data feed and wish you luck.

Every business operates differently, Rentz said. Not every organization structures exceptions or measures outcomes the same way. If a vendor cannot answer those questions with specificity, they are selling the AI layer without the infrastructure to support it. The vendors that cannot answer those questions are selling the AI layer without the infrastructure to support it, and Rentz has seen enough of those implementations to know how they end.

What this means for fleets running intermodal

For carriers running intermodal or drayage, the takeaway is straightforward: the shipper TMS platforms and container-tracking tools your customers use determine how much friction you face at pickup and delivery. If the shipper is working from fragmented data, you inherit the chaos in the form of late delivery orders, incorrect terminal appointments, and demurrage bills that should have been avoided. If the shipper has invested in validated, real-time container tracking with direct carrier and terminal feeds, you get accurate pickup windows, fewer detention charges, and fewer calls to dispatch asking where the container actually is.

The difference shows up in your cost per move. When a shipper's system triggers drayage scheduling based on a terminal gate-in event that actually happened, rather than a stale carrier API update from two days prior, you avoid the empty chassis trip. When the delivery order hits your inbox automatically because the system knows the container cleared customs and is available for pickup, you avoid the 24-hour delay waiting for someone to manually generate the paperwork. Those are the operational details that determine whether an intermodal move pays or loses money, and they trace directly back to whether the shipper built the data foundation before they bought the AI dashboard.