Platform Approach to Enterprise AI: Five key considerations for Retailers

Retailers today are increasingly leveraging large amounts of data to better understand their business, their customers, and the markets they are operating. The abundance of computing power and technology advancements in the last few years enabled them to harness the power of AI/ML to make informed decisions. Retailers over the years invested in a plethora of solutions like data warehouses, data lakes, self-serve analytics tools, data integration solutions, virtualization tools, ML platforms, and so on. We started envisioning scenarios where machines get the first right of refusal in the decision-making cycle of man-machine interactions.

But the key question that remains even after many years of advancements in this space –have we at least succeeded partially? While we may not be there yet in terms of seamlessly orchestrating algorithmic decisions across the entire retail value chain, we already started embracing the enterprise AI capabilities albeit the progress has been slower. While the leaders are in the journey of building the enterprise AI platforms, the others are exploring building this capability by taking baby steps. Even the retailers who achieved some success in building such capabilities ended up creating AI silos to address specific challenges at the functional unit levels e.g. Marketing, Merchandising, Supply chain, Digital operations, and other lines of business.

What does a platform approach to Enterprise AI mean for retailers??

It is essentially the ability of a common AI platform to infuse analytics into the planning & execution systems of Retail by applying algorithms at scale in a comprehensive manner to enable optimal decision making. The key capabilities expected from this platform include but not limited to:

1. Data management: Robust data management to ingest (from processes, devices, internal & external data sources), store, process, serve and govern the data. This encompasses multiple aspects of data concerning master data, data quality, common information models, data integration services, life cycle management, security, and many more.
2. Intelligence layer: Consists of (a) Set of ‘plug & train’ Algorithmic modules to solve common retail business problems across functional areas e.g Pricing, Seasonal forecasting, Assortment management, Logistics network optimization, and so on (b) AutoML and Custom model development workspaces to add enhanced intelligence (c) Tools & services to support different frameworks, hardware, compute options, and manage the end-to-end model lifecycle management
3. Insights as a Service: API catalog to expose insights as a microservices from AI platform to ensure consistency and promote re-use of models across the enterprise
4. Cross-cutting services: These are the services that provide the common engineering tooling for the platform which includes API management, Integration platforms, Observability, Data, and ML operations.

Apart from the above, it is recommended to have additional capabilities around self-service, low code-no code, workflow automation, process automation, and visualization tools with the common AI platform providing intelligence to all these services.?In a true enterprise AI platform, there will be a lot of business applications integrated with the platform to get the intelligence embedded in them.

Here are five key considerations for retailers when they embark on a journey to build Enterprise AI platform capabilities

1. Develop platform vision: In the AI continuum cycle, the Enterprise AI platform appears in the far right so this needs a strong vision & commitment from the leadership. The vision should have the definition of objectives, roadmap in terms of capabilities & business priorities to be delivered, expected outcomes, stakeholder mapping, platform delivery options, and so on.?After defining the vision, the platform can be developed step by step which means the enterprise AI capabilities may be fragmented during the initial stages but eventually the target state can be reached by aligning the intermediate objectives/deliverables to the larger vision.
2. Make Build vs Buy Decision: There are many factors to be considered for this – What are the IT capabilities & talent pool available within the organization; should we build a ground-up solution to avoid vendor lock-in and have the flexibility to make changes at will;?should I keep my AI assets within the enterprise; what are the financial implications of build vs buy decisions. It is to be noted that the Enterprise AI platform brings a variety of services together so off-the-shelf products may not address all the requirements so the retailers end up building bolt-on components. There may also be a need to have multiple such products to be deployed for different lines of business considering the relative strengths of one product over the other. Stitching all these different solutions together and creating an AI fabric will pose its own challenges which retailers should vary of when making build vs buy or hybrid platform decisions.
3. Break the organizational barriers: Transforming the organization into an algorithm-driven decision-making enterprise requires culture change at all levels. This needs to be planned & executed carefully by following change management procedures. Also, inter-departmental collaboration on data sharing and API agreements are very critical for the success of enterprise-scale AI programs.
4. Measure intermediate outcomes: Determining the business value delivered at intermediate milestones is very critical to measure the progress and make any course corrections. These can be business KPIs on Sales, Margin improvements, customer retention rate, campaign effectiveness, price-performance, etc depending on the use case that was resolved through the platform. For a case in point, a retailer would have implemented algorithmic interventions for Pricing in a multi-year enterprise AI journey. In this case, the dimensions of the measurement can be the following (these are just examples, not the complete list) (a) How is my price performance compared to competitors after implementing the solution? (b) Did we able to reduce the markdown dollars? Did we able to execute the markdowns at the right time? (c) Am I applying promotions to the right items/categories?
5. Establish AI council: Enterprise AI platform is a massive commitment and typically it is a multi-year journey. It cuts across multiple functions of Business, IT, Innovation, and various groups within the Retail organization. Also, the roadmap and objectives of the program need to be refined continuously due to evolving organizational needs and technology-related advancements. A high-level management oversight & executive sponsorship from top leaders of different groups is needed for the success of the program. This need to be enabled through a constitution of AI council who periodically meets reviews the progress and sets the direction for the program team

Conclusion

AI is getting adopted in every line of retail business to solve the multitude of use cases. However, the application of AI capabilities is fragmented, and solving specific use cases of functional lines of business without taking full advantage of the Enterprise AI platform has to offer. Retailers can harness the full potential of AI & ML by taking a platform approach for the application of AI across the enterprise.