Retail has always been a particularly challenging activity, as it endeavours to find new ways to deliver quality groceries at the best price. Carrefour is well known for its thought leadership in the field, for instance inventing the hypermarket format in 1963. Since then, Carrefour keeps pushing to innovate, striving to deliver the best value to our customers.
More recently, in 2018, Alexandre Bompard launched “Act for food” to succeed the food transition within all our geographies. A few months later, he also signed a partnership with Google, and, at the end of 2019, validated the new “move2cloud” strategy. He perfectly understood that cloud could be a major enabler for new mindsets and behaviors, including putting technology at the service at the business.
Retail is one of Google Cloud’s top priority verticals, as one of the largest and fastest-growing sectors, and with industry dynamics highly impacted by digital. Google supports major retailers throughout the world such as Ocado, Target, Metro or The Home Depot. It’s only natural for Google to help Carrefour in its transformation journey, not only sharing its expertise on data analytics, artificial intelligence and machine learning, supporting in marketing and media transformation or powering Carrefour’s new digital workplace, but also bringing transformative change to infrastructure with cloud.
This four-handed article has been written combining the views of Yann Barraud, Global Head of Cloud and Data platforms at Carrefour and Samuel Bonamigo, Vice-President of Southern Europe Google Cloud at Google.
Carrefour cloud journey
Back in 2018, Carrefour was using a mix of on-premise data centers, private cloud, and opportunistically, some public cloud, mainly to support e-commerce activities. Choices were made fully locally -individually by every Carrefour country- resulting in a scattered landscape of infrastructure providers.
At that time, Carrefour had less than 5% of apps across the Group running in some cloud. By the end of 2020, more than 25% of applications (approximately 800!) are cloud-based. And it is not only a matter of volume: Carrefour “cloudy” software powers critical capabilities like e-commerce, store tools, data or finance.
How did Carrefour accomplish this?
First, we drafted the cloud strategy, being as bold as possible: we made it public that Carrefour is now globally moving to the (public) cloud, and identified GCP (Google Cloud Platform) as partner of choice. The company did not have the luxury of time to develop its own private cloud (the central idea of Carrefour’s tech transformation being to focus all IT efforts in better serving customers).
Second, we defined clear criteria around the different cloud service models: When to use IaaS (infra / hosting), PaaS (technical components like BigQuery) or SaaS (pre-built cloud software) may seem something trivial, but the fact was that it was not that clear for everyone in Carrefour. We are pushing for a fair share of IaaS, but we prefer cloud-native applications for our core business (think about new e-commerce services or the store point of sale), and we are extremely open to using standard SaaS solutions for support functions (think about HR or finance).
Third, we also had to adapt the organization. As you can guess, all this requires new capabilities and new roles. We created a Cloud Center of Excellence (CCoE) in each of our geographies to support this move. This includes developing expertise in cloud operations to understand how to manage the new stack, recruiting cloud architects to identify the right solutions over the hundreds existing and infusing FinOps culture to constantly monitor and optimize spendings (several applications saw their costs cut by half with the same level of performance and same scope!).
Lastly, we explained very clearly to Carrefour top management how important and impactful it was to embrace this shift. We took the necessary time to explain what was the strategy and what kind of gains the public cloud was bringing to the company. Gaining the support of senior executives drastically accelerated the speed of change since they helped us to spread the word across countries, organizations and teams.
Why is Carrefour pushing so hard on the cloud?
Taking a step back, we can say without doubt that the most important step in convincing Carrefour’s executive leaders was to clearly outline the benefits -sort of the “value proposition”- coming with cloud. Here are the seven reasons we put on paper at the end of 2019.
Designed in the right way, cloud applications help deal with ups and downs in customer demand, adjusting workload automatically as needed. Days like Black Friday are less of an issue.
One concrete example: Not so long ago, the Spanish e-commerce platform was a monolith, heavy to evolve and hard to maintain. The team decided moving to a microservice architecture fully grounded on the cloud. In addition to the gain in agility (e.g., more than 400 deployments last year vs. struggling for 1/week before) the new platform is split into scalable and independant chunks of code that swallowed COVID workloads like a charm with limited outages (you might want to check Quarkus’ post on this specific point). We use Google Kubernetes Engine (GKE) to scale up and down very fast – for instance we start a new e-comm services in less than 1 second.
Scalability is also useful for the back-end. Our main data lake in France was hosted in a traditional data center, and it was hitting the walls. The platform was at the time managing 700Tb of data and we could not scale out anymore. The whole data stack was then “lifted and shifted” from on-premise infra to GCP hosting -in only a few months and without any service interruption- and it is now scaling again (+2Tb per day). On top of that, the total cost of ownership of this platform is lower than before, serving more than 80 applications and executing 100 million API calls every month. Thanks to this powerful data beast, we have access to real-time KPIs that help us drive our business. You need to know what’s the most sold product in the Carré-Senart hypermarket? No problem, your applications will find the information there.
2. Time to market
Having resources available in a few seconds really changes the way we design and run our projects. Launching a new product in the cloud is way simpler and faster.
The term “agility” refers in many cases to new team’s ways of working to deliver new software solutions fast, grounded on proximity to customers and cutting through complexity with rapid iterations. But following the pace also requires an infrastructure as agile as possible. We all have sad memories of projects failing because hardware was not deployed fast enough. Remember the good (or bad…) old days when it took up to three months to wait for the supplier to provide the machines, and then three more to rack them and install a hardened operating system?
In the cloud, mind puzzles like determining the right “size” for computing or storage ressources get solved with beautiful nimbleness. In some of the more mature Carrefour countries, our teams are able to provision the right cloud services with the right set of parameters and the right security in a single click.
This new speed proved to be very important, once again, in the recent sanitary crisis. During the first COVID lockdown we were able to launch, in a matter of days and in various countries, a set of websites designed to provide essential goods. Our e-commerce and our cloud teams worked together to make sure we would be able to deliver a brand new e-commerce service within the allotted time. When the demand came back to manageable levels, we simply shut them off. Hard to do this when dealing with physical servers, right?
Nowadays, most of what is new and cool is already happening on the cloud.
In the cloud, you can take advantage of pre trained AI models, easy to access via APIs, running on a highly optimized infrastructure for those specific workloads. Consider capabilities like image recognition and have a look to what Vision AI is providing out of the box.
Moreover, Google believes and invests massively in open source. They released Beam, Tensorflow, Kubernetes, Istio and more than 20 mn additional lines of code to fuel the OSS community. Google discovered Spectre and Meltdown and patched them with zero downtime, just like Heartbleed, Rowhammer and over 800 other common vulnerabilities and exposures.
Relying on cloud is like sitting on the shoulders of giants. Our applications in the cloud get a better “on air” rate.
Most GCP services show an uptime of 99,95% at least. BigQuery’s, for example, is >99.99%. Special mention to Cloud DNS, with an uptime of 100% (!).
What does it mean? To put it simply, your best service level is determined by the worst uptime of the underlying services of an application. So the higher the service level, the more available your application is.
Let’s say that we are building an application on GKE with an autopilot cluster and Cloud SQL. Cloud SQL uptime is 99.95%. GKE in autopilot mode has the same one. So my overall service level is 99.95%, which translates into a maximum downtime of 4h23 per year.
And… targeting 4h23 per year is really tough. In the public cloud, we can achieve it with a limited investment.
Sustainability is a vital point for Carrefour, and in a way part of the company’s DNA through Act for Food.
Cloud providers demonstrate a very good Power Usage Effectiveness (PUE). This ratio is simply calculated by dividing the total of energy required at the level of the data center by the part dedicated to computing. For example, if your data center consumes 100 000 kWh and 80 000 are dedicated to computing, your PUE is 1,25.
While traditional data centers have a ratio that is close to 1.5, clouders are closer to 1.1, meaning that they would only spend 10% of their energy in something other than computing (like cooling or lights).
Why is this important at all? While public and private cloud providers bring massive computing power to the enterprise, this comes at a cost: today more and more energy is consumed by data centers globally. Studies show that this already represents 1% of the global consumed energy. 1% may seem small, but notice for instance it is the same as the entire power consumption of Australia…
Partnering with Google brings technological benefits, but it is also an educated decision around this challenge of energy consumption. Indeed, Google’s sustainability efforts are focused on:
- Energy optimization. Raising the temperature of its data center rooms, using natural ventilation to cool down the machines and using custom servers to maximize utilization.
- Carbon neutrality. Google was the first major cloud provider matching 100% of power consumption with renewables (since 2017), it is the only cloud provider with a 0% net carbon footprint and targets being carbon-free by 2030.
- Circular economy. Google incorporates circularity from the project inception, at the data center design level down to the compute hardware. For instance, when servers require repairs, parts are replaced with refurbished parts. In 2016, 22% of the components Google used for upgrades were refurbished inventory.
All in all, when running our applications in green data centers, we are significantly lowering our global carbon footprint. Coupling this with the ability to stop machines and services when they are not used, we are achieving a lot for our future! The next step will be to adopt the “serverless” paradigm: applications consuming cloud services fully on demand, being able to scale from 0 to pike usage instantly, thus making huge efficiency savings.
It has never been easier to monitor threats and fix vulnerabilities.
Cloud providers offer a variety of services out of the box that would be very hard to develop on premise. For instance, most clouders tend to provide integrated and very powerful monitoring solutions, in the case of Google the Security Command Center.
Recently, Google Security Command Center helped us to detect an anormal activity on one of our virtual machines. CPU load was abnormally high. The moment we received this alert, we checked on the machine, identified it was used for crypto-mining, and then immediately killed it. The whole process was done in a matter of minutes, when it could have lasted weeks in a traditional environment.
Besides, GCP gives access to computing, storage, virtual machines, engineering and data services and… cloud-native security solutions. Some of these include a Web Application Firewall (WAF), encryption, or Security Information and Event Management (SIEM). In Carrefour, we are now testing some of these, starting with the WAF.
Finally, setting a Disaster Recovery Plan (DRP) in the cloud is relatively easy. Preparing to rebuild an environment from scratch becomes really as simple as scripting and automating it with a few lines of code.
And last but not least… When correctly configured and monitored, cloud can bring substantial savings.
We already mentioned the advantages of being scalable, which helps to better adapt our capacity to what we need, effectively spending less.
Moreover, the fact that cloud is a measured service enables the development of “FinOps” techniques like “right sizing” or “auto switch-off” that deliver immediate value:
- Right sizing. In a traditional data center, we used to size our machines to absorb peaks of activity with the downside of having a low average utilization of resources. In the cloud, you basically scale for the usual workload, having the option to automatically upgrade available resources if needed. In the case of Carrefour, we take advantage of two out-of-the-box services from GCP to help us do this. First, Google advises to downgrade your machines if it detects that they are oversized. All this is happening in the Recommandation Hub. Second, for specific workloads, Google offers custom resources, dedicating the most powerful -and expensive- machines to cover for the most complex workloads. For example, for one of our store applications in Belgium, the team mixed two types of machines, E2 VMs, cost efficient, and few custom ones for big memory requirements.
- Auto switch-off. Did you realize that non-working hours represent 65% of the time of a week? If you keep non-prod machines up and running during this time, you are basically throwing away 65% of their value. In Carrefour we started to move development environments to the cloud. All the non-prod environments have been tagged to be automatically identified, and then, a script shuts environments off.
Notwithstanding the former elasticity and optimization aspects, our experience so far in Carrefour tells us that the cloud model is more efficient for three reasons:
- Hardware and platform resources are often cheaper in the cloud. For example, using IaaS the right way and activating optimizations led us to reductions of up to 40%. The same applies to middleware: When using PaaS solutions like Cloud SQL we move to a pay-as-you-go mode and avoid license acquisition and maintenance costs.
- Cloud allows to focus the energy of tech teams in the right manner, moving away from basics of hardware and network maintenance to higher value tasks. When shifting from traditional data centers to cloud, we noticed up to 40% of our operations efforts can be reused elsewhere…
- We remove fixed costs when we close data centers. If we take a data center down, we remove all management and services tied to it (monitoring, firewall, network connections, routing, energy, cooling, physical security…). As an illustration, we closed two data centers in France end 2019, with fixed costs going from more than €10mn to less than €3mn.
Cloud and data, data and cloud: the chicken or the egg?
Often, data services come to mind when considering cloud computing. Data is of course at the core of our business since retail has the advantage to be rich in information, which can be found everywhere: in products and merchandising, in warehouses and supply chain, in marketing campaigns and media… even at the point of sales and the store. Not to forget that the retail business is by essence fast, which makes a company Carrefour a great place to be for data lovers!
But data extraction, analysis and visualization are greedy in terms of compute and storage. Our personal experience over the last years showed that running big data in a traditional data center is painful and expensive, both in terms of infrastructure requirements to scale and operations intensity. When we signed this partnership with Google, it was clear to us that we should move our data activities to the cloud.
A few years ago, only a few countries at Carrefour were equipped with data platforms. Data savviness was somehow low and, even if the company was sitting on a golden mountain of raw information, it was not exploited at its full potential. The signature of the Google / Carrefour partnership back in 2018 created the opportunity to leverage GCP’s first-in-class data tools to ramp up the whole company on data-enabled services, be it reporting, AI and advanced analytics or even the valorization of the data itself.
Google is known as “the data company” for many reasons. The enterprise has a strong data-driven culture, meaning that Google’s employees prefer to take decisions based on facts vs. hunches. This led them to develop data tools for their internal use, tools that were made available to customers too, usually as a service in the cloud. Assets like Tensorflow or Collab for data scientists, but also their data processing ecosystem including PubSub for events publishing, Dataproc (Hadoop on demand) or Dataflow, an Apache Beam-based serverless data transformation platform.
We must mention here that BigQuery is a real differentiator for Carrefour, and it became the cornerstone of many of our data usages. Basically, consider BigQuery as a data warehouse for which you pay on-demand (cloud…). The “Data Lab”, ignited by Carrefour and Google, is heavily relying on it. Thanks to BigQuery, data scientists can access a very large amount of data and spend most of their time on model development. Worth to mention that BigQuery enables advanced usages but it also allows traditional data experts, SQL fluent, to keep on mining data with the tools they know.
Recently, Google acquired Looker, a modern BI and analytics platform. At this very moment, Carrefour is using it to provide data-based insights to its suppliers and thus optimize the collaboration. Looker helped Carrefour a lot, for it is a cloud-native offering, natively integrated into the Google ecosystem (in our case interfaced with Carrefour’s global BigQuery data lake as we speak). Moreover, its markup language allows us to perform specific data transformations with ease. And naturally, it provides the user experience we need.
And yet another example of the synergies between data and cloud: “Big data” platforms. Last year, we created the architectural blueprint of Carrefour’s first cloud-based data platform (codenamed “Eagle”). It is available for all our geographies as a collection of scripts to deploy at scale (Terragrunt mainly). The idea behind Eagle is to provide each Carrefour entity with a pre-assembled platform we can deploy locally quite fast, making advanced analytics possible for everyone. We really wanted to help our teams concentrate on the business side of data, not on the technical platform, which we consider as a means to an end. Today, Eagle is live in two Carrefour countries, Romania and Belgium, and will be deployed in three more this year (Italy, Poland and Taiwan). By “deployed” we mean having a platform up and running, data flowing in and at least one AI use case running on it and delivering actual business value. For example, thanks to this platform, we are able to automatically determine the best assortment depending on the store location. This use case being based in Belgium, we can for example advise to switch beer brands to better respond to customer demand. We might post an article about Eagle that later in the year… stay tuned!
We hope that these examples convinced you that the cloud is a fantastic booster for data. This is why in Carrefour, to enable both our data and our digital transformation, we decided to put these two key topics under the same umbrella. For almost two years now, I (Yann) am doing my best to leverage my different experiences in data, and learn how to scale them up thanks to the cloud, relying every day on the precious help of many friends in Carrefour and Google.
What comes next
Carrefourians are quite proud of the progress of cloud. Back in 2019, it was impossible to imagine we will be here. However, we also acknowledge we are in a critical moment, in fact in the middle of the journey. Important challenges remain ahead of us, including developing cloud competences, replatforming our apps or keep on developing AI algorithms over cloud data platforms.
And our partnership is also very special. It is interesting to see the excitement of senior Google leaders like Hamidou: “Our partnership with Carrefour around GCP is unique in the market today. We are thrilled to contribute to Carrefour’s strategic objectives by turbocharging productivity, giving powerful insights to business decision makers and accelerating Carrefour’s global cloud adoption”.
What to expect for the future? The journey does not stop here: We are aiming to become a 100% cloud company.
More importantly, we recognize that Carrefour still has a lot to learn when it comes to cloud and this is what makes the journey so fun 🙂 We would like to grasp every opportunity to leverage our partnership to move forward together, transforming the company and making it a bit better for our customers every day.