What running retail at the edge teaches us about scaling enterprise intelligence

Ahead of Google Cloud Next, Rakuten Symphony SVP, Global Head of Sales, Enterprise Business Unit, Anirban Oni Chakravartti shares insights from the company's work helping retailers overcome the limits of centralized IT by moving intelligence closer to the customer. The key takeaways are a blueprint for any enterprise running latency-sensitive operations at physical scale.

Companies with regional or even global footprints know firsthand how quickly inefficiency can scale. Any delay means immediate, measurable business consequences and potential impacts for an untold number of customers.

While retail is a diverse market, it hasn’t seen widespread transformation in years and the cracks in aging foundations are beginning to show. We’ve worked with a range of businesses hitting the limits of centralized IT architectures. A familiar question often arises: What to do when every outcome is exponential, with the entire experience hinging on an optimized sequence of physical actions informed primarily by the best data available at a given moment?

Get the cloud as close to the customer as possible.

Incremental improvement is never the goal. Companies want to implement architectures that both free them from limitations and unlock new capabilities.

Moving the cloud closer to the customer

The most consequential decision we see companies taking is to stop processing data remotely and start running intelligence processing at the store level via distributed cloud. Targeting edge compute deployments at physical store locations is the foundation of robust business continuity. It also increases local speed required to conduct local inferencing and decisioning for real-time decision making that is not dependent on centralized datacenters.

One area where we’ve collaborated frequently with customers is on deployment of software-defined storage architectures that keep data local to safeguard privacy and maximize use for business optimization and improved experiences. So, whether running cloud-based software apps or proprietary AI solutions, the best data always feeds outcomes.

And it’s repeatable. Location by location, city by city. Cloud and edge working in sync with the right workloads running in the right places.

Let’s break down the key architecture strategies customers are deploying today:

• Platform consolidation. Retailers typically run multiple operating systems and databases across store infrastructure, each requiring separate firmware and application upgrade cycles. A Kubernetes-based platform consolidates all of that into a single environment where containers and virtual machines run side by side, transforming fragmented, high-maintenance stacks into a unified system that can be updated in a single operation.
• High availability storage. Hyperconverged storage architectures deliver redundancy and uptime retailers require at store level that when combined with Kubernetes eliminates single points of failure for availability without the downtime risk of traditional master-dependent architectures.
• Application migration without disruption. Security, kitchen management, point of sale, billing and order management all migrate onto a single platform without requiring applications to be rebuilt. Kubernetes supports both containerized and virtual machine workloads so existing applications move as-is while the infrastructure underneath them modernizes.
• Store-level AI optimization. No two locations should serve the same customer in the same way. A store near an office district may run a completely different demand pattern on a Tuesday than one in a residential neighborhood. With AI agents running at the edge, each location's specific footprint is adjusted for real-time requirements like staffing or prep.

Accuracy, availability, anticipation of demand—whatever the metric being optimized, it ultimately begins to define a new customer experience. This is only possible as a result of local compute and data that don’t require roundtrips to centralized data centers.

The strategy behind data available at the edge

It is an increasing imperative to build architectures that never become a hindrance should a failure arise. That is why more retailers are starting to make the decision to run essential applications locally and independently.

Consider a retailer placing its inventory order for the following morning. If connectivity goes down at 7 pm under a centralized model, an order may not get placed. But with cloud at the edge, the decision can still happen instantaneously without the store needing to wait on cloud infrastructure.

In this scenario, centralized cloud becomes the backup for data, not the dependency for decisions.

A store’s local software-defined storage has its metadata connected to the back-end cloud in a deliberate architectural split between what lives at the edge and in central repositories. Moving from multiple operating systems and applications to a single platform makes data across all operations visible and analyzable from a centralized structure to simplify IT demands.

In practice, all of the intelligence generated locally like order patterns, equipment performance, demand signals or real-time inventory stay safely within the retailer’s own ecosystem. Over time, that proprietary data compounds, creating opportunities to glean additional operational insight specific to each store’s actual operations versus generic industry patterns.

What it means for enterprises everywhere

These advancements are within reach for any enterprise. Anyone running latency-sensitive operations at scale will eventually face the same fundamental constraints.

Increasingly, compute speed matters, and where and how data is stored matters. A fundamental question IT leaders are grappling with is which workloads belong at the edge versus centrally, and how they work together to accomplish new business objectives.

When edge compute, local storage and cloud-backend are all in sync, businesses can pursue new opportunities. It is a repeatable model directly transferable to high-volume physical environments like logistics, manufacturing or healthcare. Basically, any business where decisions need to be made faster and downtime isn’t an option.

‍