Transcription
Transcript
This customer story has been adapted from a presentation given by Des Travers, CEO of DPD Ireland, and Colin Kennedy, Managing Director of DPD Pickup (DPD Ireland's out-of-home delivery network), at CARTO's Spatial Data Science conference.
Introduction: Where Route Optimization Meets the Real World
DPD Ireland delivers and collects parcels, and tries to do it in the most efficient way possible. The company has been a CARTO customer for around six or seven years, going through several iterations of its routing technology along the way. As Colin Kennedy puts it, this is a story about “how the rubber hit the road when it actually landed on the ground — that's where it meets the humans.” It's the story of what happened when Des Travers arrived at the business seven or eight years ago with a clear vision and ambition, and what the organization had to build, and unlearn, to deliver on it.
The Goal: Turn-by-Turn Navigation and 1-Hour Delivery Windows
Route optimization, for DPD, meant finding a better, easier way to get parcels to the doorstep — with proper turn-by-turn navigation, to the right door, every time, whether that's a house, a shop, or a locker. It also underpins one of DPD's signature commitments to customers: telling them within a 1-hour window when their parcel will arrive. Once the company knew where a route started and could calculate the journey through it accurately, it could offer that guarantee with confidence.
Getting there also meant changing who could do the job. Historically, the driver was the expert: he knew his route because he'd been driving it for three, four, five years. But that created a dependency — drivers need holidays, need Christmas off, and someone without that accumulated route knowledge had to be able to step into the van. The goal was to build technology that let any driver, not just the veteran on that specific route, deliver with the same efficiency.
The Technical Challenge: Navigating Ireland's Addressing System
One of the first obstacles was that Ireland had no postcode system for most of the company's history. Eircodes, a code-per-address system, launched in 2015 and give access to precise GPS coordinates and enriched location data — but by the time Des Travers joined in 2019, four years after launch, adoption was still stuck at around 4%. Almost nobody knew or used their own Eircode.
That left DPD with a deeply inconsistent addressing system to work with: 2.2 million addresses in Ireland, and roughly 200 million variations of how those addresses get written down. Some rural roads stretch 5–6 km with a hundred houses sharing the exact same address, and on one such road near Colin Kennedy's own home, four different households share the surname Kennedy. Simply figuring out the correct endpoint for a delivery was the number one challenge before route optimization could even begin.
Achieving Routing Accuracy With Probability Engines and Pinning
DPD had already built some in-house technology based on probability modeling, but it wasn't accurate enough on its own. Working with CARTO on top of that groundwork, the company pushed accuracy to around 90–92% — a strong result given the near-total absence of postcodes, but still not good enough to run a production route optimization system on.
To close the gap, DPD introduced “pinning”: a self-serve option that let consumers drop a pin on a map to confirm exactly where they wanted a parcel delivered, whenever the address didn't meet the confidence threshold. Eircode adoption itself later jumped from 4% to 94% after COVID, simply because people were home to notice and use their codes — but pinning was what carried DPD through the years before that shift. Between an Eircode when available, a probability-engine estimate, and a consumer-dropped pin as a fallback, DPD reached 100% address resolution every day, which allowed it to run a full optimization every night.
The Human Challenge: When Expert Drivers Reject the “Perfect” Route
That nightly optimization generated what the team had explicitly asked CARTO for: the single most efficient possible delivery route for the country. And that's exactly where the project ran into trouble. The drivers who were new to the business, freshly trained on the optimized routes, adopted them without friction. But the veteran drivers — the previous experts — felt threatened and pushed back hard, insisting they could route themselves faster than the algorithm by relying on years of accumulated local knowledge: which lane to cut through, which corner to come around, which stop to skip for now and come back to later.
DPD's response was to ask drivers to run the optimized routes for three or four weeks, on the belief that the benefits would speak for themselves — and they did, once drivers gave it a chance. One concrete win: drivers could now load their van stop-by-stop without hunting for addresses, cutting roughly 40 minutes off their time at the depot. But getting people to that point took real effort. As Des Travers put it, winning over drivers who had “done it that way for the last five years” and wanted to keep doing it that way tomorrow meant facing significant resistance before the value became obvious to everyone.
The Compromise: Building “Driver Preferential Optimization”
DPD had started with what it calls a dynamic optimization engine: the purest, most efficient route possible, full stop. It had to move to something different — driver preferential optimization — which gave drivers a say without handing full control back to instinct. Drivers could ask to work broadly within an area aligned to what they already knew, rather than being dictated to.
Building that model meant encoding a long list of real-world constraints into the optimization engine: vehicle capacity, parcel size and fill rate, time windows around collections happening alongside deliveries on the same route, driver working hours and shifts, and road and terrain knowledge specific to each area. Driver expertise mattered less for point-to-point travel time and more for what happened between points — handbrake up to handbrake down — so the system had to account for variable stop times, which matter enormously across 130 to 200 stops in a single day. DPD also had to build in a constraint nobody had asked for in earlier iterations: drivers wanted to finish their route close to home, which matters in a country where some routes run 300 km long.
The underlying lesson, as Des Travers frames it, is that “the best technology will guaranteed, always, without doubt, forever fail, no matter how good it is, if the users won't adopt it.” Moving to driver preferential areas was, in his words, “a huge pill to swallow” — but it's what turned a technically brilliant product that people resisted into a slightly-less-than-perfect one that everybody actually uses.
Streamlining Depot Operations
Route optimization on the road needed a matching process in the warehouse. Previously, warehouse staff had no way of knowing where to stage a parcel before a driver went out to load the van. Working with CARTO, DPD built a simple label-based system: when a trailer arrived and a parcel was scanned in at the depot, it printed a small label showing the route and stop number — route 14, stop six; route 21, stop seven; and so on. Warehouse staff placed each parcel in the corresponding spot, so that by the time a driver arrived in the morning, everything was already laid out in delivery order.
That change alone saved drivers about 40 minutes a day that would otherwise go to sifting through parcels to find the right one first. It also removed a point of fragility: the process no longer depended on a specific warehouse person showing up, since the labeling system worked the same way regardless of who was staffing the depot that day.
The ROI: Taking Vans Off the Road and Making Training Easier
DPD now runs the CARTO optimization across 40 depots and roughly 2,000 vehicles, with a phased rollout that will put more than 80% of depots on the system by the end of the year. What started as active resistance from experienced drivers has flipped entirely — depots are now asking to be brought onto the new system because they can see the value directly.
The financial impact is concrete: DPD's more efficient routing has typically taken three to four vans off the road per depot. At roughly €1,000 per van per week, that's about €3,000 a week in savings per depot from routing efficiency alone. It has also fundamentally changed training: with driver hiring growing harder across the industry, a system that gets new drivers productive faster, without years of accumulated local knowledge, has become a meaningful advantage.
Future Plans and Final Takeaways on Change Management
Looking ahead, DPD wants to enrich what it pulls out of the optimization system further — using historical comparisons to track how routes and performance evolve over time, and building on delivery clusters and least-cost routing to extract more efficiency from a system that already works well.
The core takeaway, in Colin Kennedy's words, is that the technology itself was never the end goal — what it delivers to the business is what matters. “You cannot ignore the human factor,” he says, reflecting on lessons learned the hard way. Change management is a defining factor in projects at this scale, and DPD's journey had plenty of painful moments along the way. But taking the needs of every stakeholder seriously, including the drivers who felt most threatened by the change, is what ultimately made the result work.
