Could LLMs provide the foundation for the future of customer service in the logistics sector?

Large language Models offer the potential to bring great change, but there is a lot fine tuning to be done, says David Ray Jr., who is hoping to harness the tech to address customer service challenges in the logistics industry.

You can read this article in 10 minutes

In the world of logistics and supply chain, where every minute counts and every shipment matters, customer service can take up a significant amount of resources. It thus goes without saying that the use of AI could bring great efficiency and cost benefits to this particular aspect of logistics, just as it has done in other areas.

Someone looking to bring about this change is David Ray Jr., who has spent the last 5 years working on various tools for shippers, 3PLs and carriers – including bespoke order management, transportation management systems and driver productivity tools.

Ray Jr. believes there is real potential to transform customer service in logistics through the power of AI in the shape of Language Language Models (LLMs). He is nonetheless wary of the limitations of AI technology as it stands, and stresses the importance of restricting use cases and fine tuning models.

To get the perspective of someone in thick of the development process of such solutions, we took the chance to quiz Ray Jr. about the potential of LLMs to be harnessed for customer service improvements in the logistics industry, the limitations of some existing GPS-based ETA tech, scepticism towards new digital logistics solutions, and the importance of AI model tuning.

First off, we asked Ray Jr. why he believes LLMs could be the future of customer service in the logistics sector.

LLMs and customer service in the logistics sector

Ray Jr. told trans.iNFO that he had been working on a lot of transport management software that focused on route planning and optimisation. However, he then encountered something that made the idea of using LLMs for customer service stick:

“I’d learned of a road transport company that had lost a large customer primarily due to quotes, ETA requests, and other customer service issues falling through the cracks. The enquiries were not being responded to in a timely manner; Information was being requested and the customer was not getting replies. I understand that their customers went somewhere else because they believed that they would be better serviced by a different provider that offered more robust technology.”

This experience has in part spurred Ray Jr. to begin developing a customer service solution for the logistics industry that utilises LLMs to streamline communication processes and enhance response times.

By automating responses and facilitating quicker resolutions to inquiries, Ray Jr. says LLMs have the potential to bridge the gap between transportation management and customer relationship management, an area he says can be overlooked by existing software providers.

“I see generative AI playing a huge role in helping companies provide better response times and quicker resolution times than humans. For instance, searching between disparate systems, or sending messages to locate items that should be known by another system,” Ray Jr. told trans.iNFO.

AI limitations and the importance of tuning

This doesn’t mean, however, that Ray Jr. isn’t aware of some of the shortcomings of AI at the present moment, as well as the importance of ensuring models are tuned appropriately:

“Some models don’t understand certain terms, like “drop off” for example. So there’s some tweaking that’s going to need to be done for sure. We really need to train these models to become more efficient. If something goes into a miscellaneous bucket as opposed to an ETA enquiry, that would not achieve the goal of creating a quicker response time for a customer.”

Ray Jr. added:

“It’s a little exciting and terrifying at the same time. You know it’s a new technology and there’s a huge opportunity there, but as you dig more and more, you start to see the limitations and the sheer amount of work and data to process to get these models to perform the way we want.”

A number of business in the logistics sector are of course already using AI to help with customer service. Nevertheless, things can go wrong, as was evidenced by the case of parcel delivery company DPD, whose AI-powered customer service chat bot recently swore at a customer and even wrote a poem criticising the company.

Therefore, according to Ray Jr., it will also be important to restrict use cases to avoid such problems.

“I’m a little more risk-averse, so I’d keep the use cases very limited. I recall hearing that a company had used a chatbot that created a promise to solve a customer problem that didn’t exist. They were then liable for that promise that they’d made. Those are the kind of issues to think about when fine tuning models, and it’s where I see real danger at this point. There’s so many ways things can go wrong if you’re not writing clear and precise prompts. It’s trial and error. I just wouldn’t be comfortable with the risk of throwing something into the wild having seen those issues.”

Issues with some GPS solutions

Ray Jr. is also of the opinion that there are limitations when it comes to existing GPS-based solutions.

“When providers offer a GPS location or the ability to open an app and see where an order is located, once you peel back the layers of the onion, you start to see the technology is there but that the driver may not pick up their phone and turn on the app. That means you have their location 8 hours ago as opposed to now. On top of that, even if you have their location, in the less-than-truckload world, you may not really see when they’re going to arrive because they’re making multiple stops with inventory that is not yours.”

Ray Jr. added: “If a driver has 10 pallets from 3 different customers, and they’re going across the US and have their GPS location and tracking on, but they look like they’re veering off the route the customer’s expecting, that’s not really helping solve the customer service problem, because the customer’s wondering why the truck’s going in the wrong direction. Then you have to pick up the phone, or send an email or text to figure out why the shipment appears to be off-route.”

Scrutiny

Interestingly, Ray Jr. was also conscious of the social media criticism that has come the way of some digital logistics entrepreneurs and startups, particularly after the collapse of Convoy.

He added that the situation has seen him think twice about the messaging associated with the solutions he’s working on:

“I have seen this story play out so many times; I’m somewhat hesitant to even bring up what I’m working on. I’ve launched products before that could fall into that category of “the next shiny rabbit” kind of thing. If I start off by saying “here I’m David, I’m building AI to solve logistics problems,” I think that just rubs a lot of people up the wrong way. I have essentially felt like I was blowing a little bit of smoke at times. So I’m focused on educating the market of the importance of great customer service. I think I’m more married to the problem and solution than promoting AI buzzwords and all the hype,”

Brokers and driver pressure

Finally, one area we also honed in on was the specific needs of the US market.

One observation Ray Jr. made here was that truck drivers based in the US are struggling with the fact they often have to switch between numerous different applications.

“There’s a couple of things going on there. In the USA, I think at this point there’s some 26-27,000 registered brokers. The majority of them are going to be smaller players,” Ray Jr. elaborated.

He continued, “However, there are going to be brokers who want to attract more shippers via technology. So imagine that of the 26-27,000 brokers, a couple hundred of them create their own proprietary app or technology to attract more shippers.”

Ray Jr. then expressed concern about the pressure this puts on drivers:

“So the shippers may think, great, you have the new shiny rabbit tool, so we’re going to go with you. That then forces downward pressure on the drivers because if they want to work with this broker, they’ll have to download the app and be registered on it. This is more of a problem in the LTL market, but that’s still a significant amount of the market here in the US.”

Reflecting on his past experiences, Ray Jr. also shared insights gained from interviews he did with drivers: “After interviewing 40-50 drivers back in 2022, when I was working on a product called LoadWise, I realised that these drivers were having a really hard time and they were very frustrated.”

Moreover, he emphasised the challenges faced by drivers due to the proliferation of apps, saying, “Each additional app creates an additional node and a network that has to at some point be consolidated across all of these different teams.”

Ray Jr. then finished off by adding:

“I don’t see this future Nirvana where everyone has one application and that’s just the way things are done. That almost seems too good to be true. I’d rather build products that allow them to rapidly build their app or to leverage a language model within their app.”