How Ecommerce Pure Plays Continue to Thrive: Four First Steps for Enterprise Retail CTOs
The COVID-19 pandemic shocked the retail industry system. It wasn’t long before high streets were boarded up worldwide, and many traditional businesses pivoted to fully operating online. As traditional businesses struggled to stay afloat, cloud-based disruptors (ecommerce pure plays) started acquiring brands that had gone under in order to make them available online.
AWS has been supporting various ecommerce pure plays, like Zalando, Mercado Libre, and Depop, along their journeys. This post will distill these experiences into a set of practical suggestions and first steps to inspire Enterprise retail CTOs looking to rethink their technology strategies. While technology utilization is neither the only reason for pure plays’ success nor the main source of failure for closed retail shops, it is becoming an increasingly important agility differentiator and enabler.
Let’s dive in.
Behave Like a Technology Company
Over the last ten years, enterprise retailers have taken a primarily financial view of their technology organizations. IT organizations were described as cost centers that required optimization, and outsourcing was the main lever for achieving this cost optimization.
But ecommerce pure plays have inherited a radically different technology approach. Many began as start-ups that preferred in-sourced, small teams of highly skilled engineers, as opposed to large teams, in order to create a culture of autonomy, automation, and agility. They recognize technology as a key differentiator for success—especially when you can only reach customers via digital channels.
Enterprise retailers also have invested in digital teams, but they are often seen as a bolt-on solution to the “old core,” and they are obstructed by clashes in culture, processes, and architecture. Behaving like a technology company today demands end-to-end modernization across the business—even if parts will be outsourced or utilize third-party packages. This means considering:
- Investing in a highly skilled core engineering team that can set the architectural and engineering foundations to enable integrations with third-party and outsourced applications.
- Growing an engineering and ownership culture that builds and runs applications in tandem, making continuous improvements along the way.
- Bringing the business deeper into the delivery cycle and promoting Agile delivery principles so that the business adopts an agile mindset.
A concrete organizational first step in this journey is assembling stable Product teams rather than more transient project teams. From an architecture perspective, a solid API strategy combined with CI/CD pipelines and a cloud strategy will provide the foundations required for agility.
Ecommerce agility is Paramount
Agility is a must-have in a retail landscape, where many retailers are selling multiple brands to multiple customer segments, and predicting customer behaviors is becoming nearly impossible. Moreover, the need to adopt agility is particularly acute for ecommerce retailers.
Many ecommerce pure plays have embraced a headless commerce architecture—or decomposed technology stack—to enable greater agility. This strategy utilizes the best tool for a given job, whether it is bespoke-built, SaaS, or a third-party offering. Then, the customer experience is assembled by composing services exposed through APIs using a cloud-based composition layer. The MACH Alliance—microservices-based, API-first, cloud-native SaaS, and Headless—is leading the thinking on this concept. This approach drives agility by enabling new feature delivery at pace across different components, rather than sticking to a common release process.
Beyond ecommerce, this approach also remains valid for driving overall IT agility. While ecommerce can be seen as the emerging tip of the iceberg, a long-term strategy for breaking down all monolithic applications is essential.
Speed: An Essential Business Mission—And a Tradeoff
Andy Jassy once said, “Speed disproportionately matters at every stage of your business.” And this is even truer in very competitive industries, like retail, where it won’t be long before competition can simply copy a new idea or product. As a result, marketing teams must experiment quickly with new campaigns.
When utilizing cloud technologies, ecommerce pure plays take a deliberate approach to achieving speed. They generally dial down portability and trade out instance-level architectures for high-level cloud native services (e.g., AWS Lambda or Amazon EventBridge—see full list here). This pragmatism regarding portability cost versus loss of revenue originates in not being able to release fast enough. The engineering overhead of going portable is rarely calculated by traditional enterprises.
A concrete step is opting for a TCO-view on architectural decisions made under the portability banner. Asking questions like, “What is the additional cost in terms of engineering and support of this portability decision?” and, “How expensive would it be to reverse it out at a later stage rather than investing in it upfront?” are vital.
Enable Innovation through Democratized Access to Data and Managed Data Quality
It’s no secret that machine learning (ML) is the heart of many retail business processes—from warehouse optimization and dynamic pricing, to forecasting and personalization. And there are two keys to success that ecommerce pure plays focus on when utilizing this technology: democratized data access and managed data quality.
Democratized data access means that every data type utiized in decision processes (e.g., exhaust data from in-house systems or third-party market data) should be made available via simple APIs and with a well-understood production delivery CI/CD pipeline for the models executing against them. By decentralizing model development, domain level teams (e.g., marketing, supply chain), and implementing a data self-service model, ecommerce pure plays enable innovation to occur at pace in each domain.
Two key aspects are required to enable democratized access to data. First, a core ML team works with local ML teams and focuses on machine learning operations (MLOps). MLOps demand removing friction to deliver models in production, as well as fostering model reuse. This means shorter time-to-market and faster experimentation stages.
Second, data quality is actively managed in each domain by data product managers. Without taking this step, data quality drifts over time and impacts the ML model accuracy. Measures to deduplicate data must be continuously active in order to remove any personally identifiable information (PII) or errors and update gaps. By doing this, everything published to all ML teams will be clean. A data catalogue with solid naming conventions is also essential to enable data discovery.
A critical step here is designating who in the organization will fulfill the following roles: ML model development, MLOps, and data quality management. If the roles don’t exist, then identify the responsible teams, and start to separate these functions. It may be worthwhile to slow down temporarily in order to install these fundamental roles that are required to scale ML in the enterprise.
The journeys that ecommerce pure play have had on AWS look surprisingly similar to enterprises. Technical debt is also there—mostly a side effect of fast growth. Most ecommerce pure plays have transformed and adapted, so there’s no reason why enterprises can’t do the same. We’ve highlighted four concrete steps that CTOs can act on when starting their transformation journey toward agility and speed. For more information, reference our white paper on ecommerce, The CTO Guide to Ecommerce Architectures: People, Process, and Technology, or contact your account team today.