With a varied and fast moving supply chain, loads of stock moving quickly through warehouses, delivery trucks, stores, and an increasingly digital mandate, the food retail sector is unlike any other retail segment. Paul Clarke, director of technology at Ocado, a leading online food retailer, explains how the cloud, robotics, and the Internet of Things is increasingly at the heart of everything the company does.
Ocado started 13 years ago as an online supermarket where consumers could quickly and easily order food goods. It does not own or operate any brick-and-mortar stores, though it effectively competes with all other food retailers, in some ways now more than ever because of how supermarkets have evolved in the UK. Most of them offer online ordering and food delivery services.
But in 2013 the company struck a £216m deal with Morrisons that would see Ocado effectively operate as the Morrisons online food store, a shift from its previous strategy of offering a standalone end-to-end grocery service with its own brand on the front-end – and a move that would become central to its growth strategy moving forward. The day the Morrisons platform went live in early 2014 the company set to work on re-platforming the Ocado service and turning it into the Ocado Smart Platform (OSP), a white-label end-to-end grocery service that can be deployed by food retailers globally. Clarke was fairly tight-lipped about some of the details for commercial reasons, but suggested “there isn’t a continent where the company is not currently in discussions” with a major food retailers to deliver OSP.
The central idea behind this is that standing up a grocery delivery service – the technical infrastructure as well as support services – is hugely expensive for food retailers and involves lots of technical integration, so why not simply deploy a white label end-to-end service that will still retain the branding of said retailer but offer all the benefits?
“In new territories you don’t need the size of facilities that we have here in the Midlands. For instance, our site in the Midlands costs over £230m, and that is fine for the UK which has an established online grocery business and our customer base, but it wouldn’t fit well in a new territory where you’re starting from scratch, nor is there the willingness to spend such sums,” he explains.
The food delivery service operates in a hub-and-spoke model. The cloud service being developed by Ocado connects the ‘spokes’, smaller food depots (which could even be large food delivery trucks) to customer fulfilment centres, which are larger warehouses that house the majority of the stock (the ‘hub’).
The company is developing and hosting the service on a combination of AWS and Google’s cloud platforms – for the compute and data side, respectively.
“The breadth and depth of our estate is huge. You have robotics systems, vision systems, simulation systems, data science applications, and the number of different kinds of use cases we’re putting in the cloud is significant. It’s a microservices architecture that we’re building with hundreds of different microservices. A lot of emphasis is being put on security through design, and robust APIs so it can be integrated with third party products – it’s an end-to-end solution but many of those incumbents will have other supply chain or ERP solutions and will want to integrate it with those.”
AWS and Google complement eachother well, he says. “We’re using most things that both of those companies have in their toolbox; there’s probably not much that we’re not using there.”
The warehousing element including the data systems will run on a private cloud in the actual product warehouses, so low latency real-time control systems will run in the private cloud, but pretty much everything else will run in the public cloud.
The company is also looking at technologies like OpenStack, Apache Mesos and CoreOS because it wants to run as many workloads as possible in Linux containers; they’re more portable than VMs and because of the variation between the regions (legislation and performance) where it will operate the company may have to change whether it deploys certain workloads in a public cloud or private cloud quite quickly.
The Internet of Things and the Great Data Lakes
IoT is very important for the company in several areas. Its warehouses are like little IoT worlds all on their own, Clarke says, with lots of M2M, hundreds of kilometres of conveyor, and thousands of things on the move at any one time including automated cranes and robotics.
Then there’s all of the data the company collects from drivers for things like route optimisation and operational improvement – things like wheel speed, tire pressure, road speed, engine revs, fuel consumption, cornering performance, which are all fed back to the company in real-time and used to track driver performance.
There’s also a big role for wearables in those warehouses. Clarke says down the line wearables have the potential to help it improve safety and productivity (“we’re not there yet but there is so much potential.”)
But where IoT can have the biggest impact in food retail, and where it’s most underestimated, Clarke explains, is the customer element: “This is where many companies underestimate the scale of transformation IoT is going to bring, the intersection of IoT and smart machines. In our space we see that in terms of the smart home, smart appliances, smart packaging, it’s all very relevant. The customers living in this world are going to demand this kind of smartness from all the systems they use, so it’s going to raise the bar for all the mobile apps and service we build.”
“Predictive analytics are going to play a big part there, as will machine learning, to help them do their shop up in our case, or knowing what they want before they even have a clue themselves. IoT has a very important part to play in that in terms of delivering that kind of information to the customer to the extent that they wish to share it,” he says.
But challenges, ones that straddle the legal, technical and cultural, persist in this nascent space. One of them, largely technical, is data management, which isn’t insurmountable. The company has implemented a data lake built on Google BigQuery, where it publishes a log of pretty much every business event onto a backbone that it persists through that service as well as data exhaust from its warehouse logs, alerts, driver monitoring information, clickstream data and front-end supply chain information (at the point of order), and it uses technologies like Dataflow and Hadoop for number crunching.
Generally speaking, Clarke says, grocery is just fundamentally different to non-grocery or food in ways that have data-specific implications. “When you go buy stationary or a printer cartridge you usually buy one or two items. With grocery there can often be upwards of 50 items, there are multiple suppliers and multiple people involved, sometimes at different places, often on different devices and different checkouts. So that journey of stitching that order, that journey together, is a challenge from a data perspective in itself.”
Bigger challenges in the IoT arena, where more unanswered questions lie, include security and identity management, discoverability, data privacy and standards – or the lack of. These are the problems that aren’t so straightforward.
“A machine is going to have to have an identity. That whole identity management question for these devices is key and so far goes unanswered. It’s also linked to discoverability. How do you find out what the device functions are? Discovery is going to get far too complex for humans. You get into a train station these days and there are already 40 different Wi-Fi networks, and hundreds of Bluetooth devices visible. So the big question is: How do you curate this, on a much larger scale, for the IoT world?”
“The type of service that creates parameters around who you’re willing to talk to as a device, how much you’re willing to pay for communications, who you want to be masked from, and so forth – that’s going to be really key, as well as how you implement this so that you don’t make a mistake and share the wrong kinds of information with the wrong device. It’s core to the privacy issue.”
“The last piece is standardisation. How these devices talk to one another – or don’t – is going to be key. What is very exciting is the role that all the platforms like Intel Edison, Arduino, BeagleBone have played in lowering the barrier by providing amazing Lego with which to prototype, and in some cases build these systems; it has allowed so many people to get involved,” he concluded.
Food retail doesn’t have a large industry-specific app ecosystem, which in some ways has benefited a company like Ocado. And as it makes the transition away from being the sole vendor of its product towards being a platform business, Clarke said the company will inevitably have to develop some new capabilities, from sales to support and consultancy, which it didn’t previously depend so strongly upon. But its core development efforts will only accelerate as it ramps up to launch the platform. It has 610 developers and is looking to expand to 750 by January next year across its main development centre in Hatfield and two others in Poland, one of which is being set up at the moment.
“I see no reason why it has to stop there,” he concludes.