The traveling salesman problem is a staple of computational complexity, and a benchmark for optimization methods.

Mathematically formulated in the 19th century, the problem tackles a seemingly simple scenario. There is a salesman who has to go to numerous locations that day to promote his wares; what is the fastest and most efficient route he can take to visit all those places?

While computationally difficult, modern technology can help with the traveling salesman problem, and two centuries after it was formulated, calculations to solve it are routinely carried out across a variety of sectors.

One such example is found in postal delivery services. In the UK, Evri is a household name. The company was founded in 2022, but finds its origins in Hermes – aptly named after Greek mythology’s ‘messenger of the gods’ – which has been around since 1972.

Today, Evri’s technology department is headed up by Marcus Hunter.

Hunter started on the path to CTO relatively early in life. He tells DCD he was initially set to leave school with a single GCSE (the UK’s national examination for 15- and 16-year-olds) in math, but ended up retaking the rest of his exams because he “got told off by [his] Dad.”

The second attempt did the trick, and he ended up moving on to do a civil engineering apprenticeship at Bradford City Council.

“My claim to fame is the first double mini roundabout in Bradford,” Hunter tells DCD with a chuckle. “It was underpinned by CAD (computer-aided design), and we didn’t have local IT support because this was back in the days when the dinosaurs were running around, so I did all the local IT support as well, and I thought, ‘this isn’t bad, this. I’m actually pretty good at it.’ So I paid for a night course for a bridge degree, and then I landed a job in the NHS as a senior IT technician.” The rest, as they say, is history.

After a series of interim roles, Hunter ended up at what was then Hermes’ UK branch, now known as Evri.

With Hermes having a 50+ year history at this point, naturally, one would assume a hefty portfolio of legacy IT would be trailing behind. But according to Hunter, much of this has actually been dealt with already.

“The real change in direction was in 2018 when we were under Hermes Europe, and the company was changing from one static German data center to another, and we said ‘enough is enough’,” he recalls. “The UK arm of the business was actually profitable, and it was more like an arm’s-length organization than actually part of the family. So we did a ‘lift and shift’ straight into AWS (Amazon Web Services), and that’s everything – we have been around 95 percent in AWS since 2018.”

The migration to Amazon’s cloud service took some time, and saw Evri adopt a number of cloud-native solutions.

According to the delivery firm’s latest annual report, it handled more than 2 billion letters and parcels in 2025, and each of those deliveries had to be plotted, scheduled, tracked, and physically handed off by Evri couriers.

For context, that is an average of 5.4 million deliveries each day, with each courier handling between 70 and 150 parcels. By anyone’s standards, that is a very large traveling salesman problem, and Hunter says Evri’s move to the cloud was partly driven by a desire to achieve the scalability required to meet this demand for its services.

“We build for two years’ scalability, but to be able to flex for five years just in case another Covid-19 event or something happens,” Hunter says, noting that even beyond the compute side of things, the company has an eight-week peak period each year where it has to hire 6,000 additional couriers to account for the influx.

“On top of that is the routing,” he continues. “We get the information from the client about the size and weight, and then we understand the typical vehicle size, and we suggest to the courier what the pattern of drops is for the parcels, and that suggestion is tailored to the experience of the couriers.”

Mapping the delivery routes is relatively simple, but impossibly difficult for a human being to plot out by themselves.

“The whole UK is cut up into different zones; we manage the zones, we manage the parcel volume within the zones, the rough size of a typical car, and we allocate throughout the entire UK the suggested parcel drop,” Hunter explains. “The courier ultimately is self-employed, so our route is ‘recommended’; they can change the route of the parcels and the drop order.”

This is where experience comes in, because as Hunter puts it, “they may know a cheeky little shortcut.”

“Modeling is used traditionally in the working day, but then we have models outside that which do forecasting, predictability, and then they change the traditional way of doing it. We take into account weather, road works, festivals and big events in the UK that could impact experience.”

Deep in the depot

The routing side of things is no surprise – as previously mentioned, the concept of such mathematical problems has been discussed for hundreds of years – but increasingly Evri is seeking to improve its customer experience with data, and lots of it.

When asked to estimate, Hunter says the company probably has around 2PB of data stored in various forms.

Evri has long been consolidating its data stores – Hunter suggests this has been done around six times in the last 15 years – and today there is one centralized data environment that is built in Databricks on AWS.

The data is collected at every stage. On the customer side, Evri is looking at GPS location, parcel size, whether a signature was provided (something they no longer use), if it was delivered to a neighbor, and where exactly the package has been left.

“The client gives us some data too: postcode, address, size, and we check that to make sure it’s compliant with where the customer is if we have a record – which we don’t always,” Hunter says. “Clients sometimes obfuscate that information because it is their customer, not ours. Ultimately, we want to please both stakeholders as well as the couriers in terms of getting that first-time delivery right. With all the data in one single place, we can have good knowledge around the parcel, and then when it moves through our environments, from the hub to the depot to the sub depot and then out to the courier, we take data points all throughout that journey.”

Some of that data is stored in the AWS S3 storage service, but then the ‘stuff’ that Evri doesn’t need to access as often gets pushed to Glacier, Amazon’s long term storage option.

Naturally, AI is increasingly getting a foothold in Evri’s operations. Earlier this year, the company partnered with AWS to develop “Veri Snap AI” using the Amazon Nova Lite AI model.

Veri Snap is an evolution of the company’s existing parcel vision technology and automatically processes delivery photos taken by couriers through generative AI.

“Once we take a photo of the delivery, we do a lot of matching with the data we have in the data warehouse,” Hunter explains. “We get a geolocation if you set your own safe place, which we are keen for people to do because it helps us. We match the GPS location for the safe place, and for the address as well, so the courier can’t deliver outside of that geolocation, and we can change that to get smaller and smaller over time as the couriers get more experienced.

“AWS and Nova pull the relevant information from the photo, and because of the trained model, we can go to very specific items – so, whether the door is open, whether there is a parcel in the photo, whether there is a person visible, etc. Obviously, we black out identities and typically try to get photos of, say, their feet or some other non-descript part, but we also have a blanking and anonymizing system that underpins that.”

Doing this required training an AI model, for which Hunter says the company took the strategy of starting with “an expensive model at first to do all the learning and coaching,” before dialing it down to a smaller and more affordable model with which they are still able to improve the matching capability (currently operating at around 98 percent compliant on photo matching).

“We chose AWS because in just 12 weeks, we were 80 percent enterprise-ready, with the proof of concept being able to handle four or five million photos,” he adds.

With this being said, AI is increasingly a cost concern for CTOs and CIOs at enterprises. With companies keen to rapidly employ AI technologies, budgeting “token usage” has never been more important.

Hunter tells DCD that he is “100 percent worried about token usage.”

“Somebody’s got to pay for the investment in all of the data centers, so I’m worried that it’s [token prices] only going to increase until they find an optimum financial model, and then it will go down,” he says.

Mergers and acquiring IT complexity

Currently, Hunter and the tech team at Evri are dealing with the aftermath of Evri’s merger with DHL eCommerce UK, the e-commerce logistics specialist of DHL Group.

After receiving the go-ahead from the UK’s Competition and Markets Authority in September 2025, DHL eCommerce UK has joined Evri and is now the company’s “Evri Premium” service. But with this, Hunter and his team have also acquired the IT infrastructure used by DHL.

“Nothing in life is easy. It’s a challenge because it’s not only a different tech, but it has different business features – how the data is consumed, the reporting of that business as well is slightly different, so we are tailoring it to be what is the best of both worlds,” he says.

The migration effort is underway, and will see Evri moving all of DHL’s services into a single environment with Evri. “We are doing it at the moment, and expect around 50 percent will be migrated by the end of September.”

Mergers are always complicated, but with Evri using cloud infrastructure, that “scalability” that Hunter is so reliant on is easy to come by. With this, the company can hope to reach its goal of delivering 1.4 billion parcels a year by 2030, and the ‘salesman’ can keep on traveling the most efficient route.