How Logistics Leaders Can Get Ready To Work Alongside AI Agents
Philip Ellis is a founding frontend engineer at chrt, where he builds software for time-critical logistics operations.
gettyMost logistics teams have more data than they know what to do with. Every hour, they see telematics data, scan data, photos of proof of delivery, fuel data, invoices and exception emails. Often, they don’t have time to go through all of this, decipher what’s important and act before a minor issue could escalate into a claim, chargeback or lost customer.
That’s where AI agents can add value. A dashboard may sit unused until someone opens it, but an agent can sift through a flow of events, compare it to what should have happened and then take a specific action.
Many logistics leaders are already exploring using agents in their operations, with 76% of surveyed chief supply chain officers saying that agentic AI will improve process efficiency in their organizations, according to an IBM report.
However, as the founding front-end engineer of a company that builds agentic technology for the logistics industry, I’ve found that many leaders are focused on the technology itself when they first have to understand where agentic AI can fit into their process and what’s required for successful implementation.
Good early agent uses involve clear rules, high volume and a high cost when something gets missed. In logistics, here are four possible areas that meet those criteria:
An agent can check a route against the plan, pick up a long stop or a deviation and then decide whether the delay would affect an appointment time. It can then pass relevant information on to the right person.
Rather than looking at a map full of dots, a dispatcher could then pick up a short list of problems from that route.
Often, logistics fraud is subtle. Maybe an invoice is for more than it should be, or a pickup is made up. Maybe cargo is diverted or someone tries to trick the system when booking.
Taken one at a time, each of these may look perfectly normal. But patterns start to emerge after hundreds of such records. An agent can, for example, look at documents, time stamps, location data and billing. It can also highlight inconsistencies and pass them on to someone to investigate.
Most audits are of a sample of transactions and take place after the fact. But an agent can monitor each transaction as it happens, check rates against a contract, confirm that there were relevant documents and identify discrepancies while they can still be corrected.
In this way, an audit can become a regular daily activity rather than something done once a quarter.
Every operation has standard operating procedures (SOPs), but there is often a gap between those procedures and what actually happens on the ground.
An agent can show you gaps in your procedures. And more importantly, it can show when your records don’t match reality. For instance, an agent can check each element of a procedure (e.g., temperature check, photo, seal check). In cases where it wasn’t done, the agent can trigger the necessary procedure.
An agent’s trustworthiness depends on the quality of the evidence it’s provided. If in your records you write what the plan is, but nothing about what happened, the agent will quickly draw incorrect conclusions.
And if connectivity fails in a trailer, in a yard or along a rural corridor, the agent could misread silence as compliance or silence as failure.
Before relying on agents, leaders should ask three questions.
• Can you trust the signal? Do you know where your records are incomplete, missing or estimated? And do you make that clear to the agent, rather than obscure it?
• Do you record events or intentions? Are planned times, scans and confirmations the same as actual times, scans and confirmations? (They shouldn’t be.)
• Does every process have an owner? When an agent sees a problem, it needs either a person or a rule to act on it. You don’t want a lights-out process. Instead, you want better use of people and of agents.
Agents can give relentless and repetitive attention to something, but people still bring judgment and the ability to think through the exceptions to rules.
This leads to important considerations about decision rights. What should an agent be allowed to do on its own, such as raise a ticket or send a reminder? What needs approval, like payment of an invoice or action on a driver problem?
Perhaps most importantly, does every action by the agent record what it saw, what it did and why? If you can’t explain an agent’s actions to a driver, regulator or customer, then it is inherently overstepping.
Drivers, in particular, will rightly resist surveillance of their work. But used well, an agent can identify safety risks earlier, confirm the driver followed an SOP or confirm delays beyond the driver’s control. To get buy-in, explain what the agent is doing and why. Then, drivers may well see value in it as their records and jobs improve.
Once you have the process and oversight in place, take a specific, small problem in your business. Is an SOP step being followed? Are you coming across invoices that have already been sent? Start with the agent in “recommendation” mode. Let people compare what it suggests with their own view.
Equally important: Watch for false alarms from the agent. Over time, too many could undermine value in a system. When trust develops and accuracy improves, gradually expand the authority and scope of agents.
Most of the value from agents will flow to companies that provide them with reliable data, set limits for them and think of them as teammates in need of oversight. Historically, value in logistics has accrued to companies that spotted problems before others. Agents give more people the chance to spot problems sooner.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?