AI drives low-cost low-carbon delivery

It’s a lot to expect the human brain to figure out how to meet customers’ heightened demand for delivery while simultaneously reducing the cost of sales for the retailer and being mindful of the carbon impact of operations, says Paul Hart of WPP’s Satalia

Online purchasing across a huge range of goods from multiple retailers to almost all locations is what we, as consumers, now expect. And that, largely, is what we get.

But while customer experience is on the up, delivery costs for the retailer can easily spiral, and all those vans that increasingly dominate our neighbourhoods can easily produce an adverse impact on the environment. This is what concerns Hart, and this is why he is so emphatic that leaving the calculations to AI is how we will mitigate the impact of distribution on both retailers’ bottom lines and on the planet. That is the beauty of AI – yes, it can optimise delivery and cost, but it also takes care of the consequences.

Making the last mile affordable

Logistics has a large and varied language of its own. Generally, a retailer, say, will oversee primary incoming deliveries to large warehouses, then it will coordinate trunking within a network or ‘the middle mile’, and then there is ‘the last mile’ to the consumer. Of course, there could be a plethora of options within this basic scheme depending on how complex the system and how reactive the operation, but it is in large part how most retailers operate.

“Our focus has typically been the last mile,” says Hart, “as this is the most expensive piece so it gets a lot of focus. It’s really an optimisation problem.” It is actually an extension of the ‘travelling salesman problem’, in other words: which, of all their options, is the optimum travel solution given all the permutations? For a supermarket, there could be trillions of options as the number of delivery routes increases exponentially with the number of customers and orders.

“So, if you think about a grocery retailer that has, say, 3,000 deliveries to make out of one depot in one day, and you take into account the fact that each of these deliveries will have a time constraint (i.e. a ‘slot’), but there will also be things like driving regulations to consider, that is why ‘last mile’ deliveries become a complex problem with far more options than you can crunch manually,” he says. “Even the best computers would not be able to calculate all the routes to find the optimal outcomes. This is why you need algorithms, and that is where Satalia comes in.”

But the solution is not the same for each client – they will all have their own objectives and priorities. “We ask what the client is trying to maximise or minimise. Is it the distance travelled? Is it amount of carbon emitted? Is the number of driver hours? Is it a combination of these and other factors? Then we'll work with them to come up with a solution,” says Hart.

For some clients, an AI solution is literally born out of paper operations – in these cases there will be no huge databases of historical data. For others – larger clients – the historical data will be held in massive datasets, then it will be a question of running a comparison between the existing operation and an AI-enhanced operation to see what advantages the algorithms can deliver. Be a client large or small, paper-oriented or digital, what clients have in common is a desire to drive inefficiencies out of their operations.

Although Satalia specialises in ‘the last mile’, there are many iterations of optimised delivery – including supply-chain solutions – that can be made more efficient in this way.

For one manufacturing client, optimising manufacturing and storage did not necessarily produce the best customer delivery outcomes. In this case, customer satisfaction and carbon reduction could be improved by rethinking the entire process – end to end. That is why ‘last mile’ delivery solutions – to work well – should take into account the entire supply chain, from manufacture to final delivery, and not just what is going on at the customer end.

Even when the system is perfected, there will always be shocks to feed into the algorithms. One such shock has been Brexit and the impact that has had on the availability of drivers. “Since Brexit, there has been a real emphasis on retaining drivers,” says Hart. “After all, you need predictability and accuracy in a system so drivers are scheduled with the correct amount of time for each route.”

Sustainability is also key

And the planet has to be a key stakeholder in any predictable and accurate system. Introducing electric vehicles into a delivery fleet is a game-changer. It not only delivers on environmental credentials, but it also requires fresh thinking.

“If you've got a fleet of, say, 3,000 vehicles, you're not just going to switch overnight to a full fleet of electric vehicles. Understanding how to operate a mixed fleet in an optimal way, and how to divide up your orders, is the current challenge,” says Hart. “Range, hilly terrain, weather are all considerations for electric vehicles. But then so is how to use your electric vehicle fleet to reduce the most carbon.”

So, for example, a grocery retailer might instruct its delivery system to restrict its electric fleet to short trips on flat roads in good weather. But, if it optimised its system to deliver the best carbon outcomes, it would do the opposite and use fossil-fuel powered vehicles for the short trips close to the depot while sweating the electric vehicle assets to deliver a better return on carbon reduction. “This is where modelling comes in. These types of calculation take a lot of understanding, but you always need to know your client’s primary objectives,” he says.

What’s next?

Smart, green, cost-effective delivery has come a long way over the last decade but there is so much more to achieve. Manual picking in fulfilment centres is increasingly being replaced by warehouse automation which is making the entire delivery chain so much more efficient and able to support very short delivery times.

Coupled with this there is a trend towards multiple micro-fulfilment centres in response to short orders and fast delivery requirements. All those considerations can be plugged into an AI-enabled system.

But it all takes brains. At Satalia, data science teams handle predictions. A geospatial team creates geospatial graphs – so much more than maps. Then there is the optimisation team – they create the algorithms and model the problem at hand.

So how do you know if the experts are winning? Metrics of course. “Frequently, these AI systems are introduced as part of a wider programme of change and you’re not comparing like with like,” concedes Hart. “That said, we always come up with a way of modelling change – that is what we do.”