The retail industry continues to be under tremendous pressure from online competitors and an ever-changing and more demanding consumer base. Retail leaders and early adopters are already applying AI to build and grow customer relationships, improve operational efficiency, and reduce fraud, resulting in increased revenue, reduced cost, and mitigated risk. But the rapid pace of change means retailers must continue to invest in AI or risk being left behind.
Key use cases include customer retention, inventory stockout, fraud detection, and marketing optimization. Deploying AI at scale is a key enabler of the benefits of digital transformation. However, in our experience, many companies have not yet figured out how to fully capture the promise of AI and struggle with implementing the technology in a manner that generates immediate and significant impact.
In the following section, we suggest several steps retailers can take to scale AI successfully.
Set the strategy first :
It is essential to start with a plan, including deciding which business domain to start with, selecting the right people to drive things forward, and choosing the data and technology that will underpin success. Different domains along the value chain can improve the company’s bottom line or customer or employee experiences.
Start with the domain that has the most significant potential impact. One retailer determined that it had nine main business domains that could benefit from digitalization: revenue management, e-commerce, customer experience, store format, store footprint, sales force, operations, logistics, and talent.
There are two main criteria for picking the best domain from this comprehensive list – the quality and composition of the team and the reusability of data and technology. In the first case, it is essential to have an internal business champion responsible for the entire value chain. Senior business executives can then act as ‘product owners’ (people responsible for solution delivery), translators (who bridge the analytics and business realms), and ‘change leads’ (responsible for change management efforts). In addition, a team of AI practitioners, such as data science and engineering experts, designers, business analysts, and a scrum master (all of whom may also be drawn from a central team in the organization) is required. From an implementation point of view, a cross-functional team with representation from the sales force, marketing, and category managers should be responsible for day-to-day activities.
For the data and technology aspect, companies usually have components that different domains can use. Mapping the data and technology and planning on how to reuse aspects where they overlap can dramatically reduce development time and cost; data and technology act as enablers, not obstacles, of progress. It is unnecessary to have the perfect data and technical backbone in place before testing and implementing use cases. On the contrary, starting with the business use case – or the problem you want to solve or improvement you want to make – and working backwards will hone your understanding of the data and digital tools that are required and avoid costly and time-consuming mistakes.
Reimagine business as usual :
Getting the most from AI requires reinventing business models, roles and responsibilities, and operational processes and using new ways of thinking and working. The ultimate goal is to make AI part of business as usual. It is not enough just to try an enhance an existing process using AI. Companies need to rethink the entire process to maximize the benefit of new analytical techniques.
A good example is how retailers allow their store managers to manage assortment dynamically. Today, retailers operate national and international chains with demographically diverse customers and constantly changing channel affinities. This means it is essential to know what customers want, what kind of products to put in the store, and how to allocate shelf space across massive-scale SKUs.
The legacy approach was plagued by store manager guesswork as they tried to estimate customer preferences, with inaccurate forecasting leading to stockouts of popular products and requiring the use of open-to-buy (OTB) dollars to replenish stock. Over-ordering was also commonplace, increasing waste. At the same time, they also struggled to test new products as there was no space on the shelves, or they could not predict customer preferences.
In one company, store managers identified and understood the issues with the existing processes before mapping out what an ideal alternative might look like. They identified problems to solve and improvements they wanted to see. The company then built an AI prototype dashboard by compiling data from point-of-sale (POS) systems, loyalty programs, and syndicated data sources to indicate which SKUs drove each category. Managers were given the opportunity and power to rapidly choose assortments that more precisely aligned with customer needs, as well as access to intuitive dashboards that visualize how many and which products should be offered in each category. They can view information showing how adding or removing an SKU would change category sales. Integrated feedback loops enable AI systems to refine, update, and make product recommendations based on what works, rather than relying on intuition or personal experience.
Adapt to an agile way of working :
In most cases, significant organizational change is needed to adapt to the interdisciplinary collaboration and the agile working methods required to scale AI successfully. Leaders like the CEO and domain managers need to act as role models, reaching across organizational boundaries to make the new behaviour sustainable. Moving to a sound agile operating model requires leadership to transform from ‘masterminds’ who delegate tasks and instructions in a top-down manner to ‘catalysts and collaborators who meet with the team daily and ensure the delivery of impact.
The traditional technology/IT delivery model, with heavy upfront planning and little flexibility, should install agile feedback loops that enable a test-and-learn approach, with constant reiterations refining output. Organizations can then transition from a focus on scheduling and protocols to one that concentrates on producing better products and business models. As a result, businesses need interdisciplinary teams that own a specific product or customer journey and take full responsibility for building the right pathways.4
Leverage Machine Learning Operations to industrialize AI capabilities :
Once we have the AI prototype and process in place and pilots have proven impact, the next important step is industrializing the AI capability. To build, deploy, and manage analytics/AI applications with speed and efficiency at scale, a rapidly expanding stack of technologies and services is required. This enables teams to move from a manual and development-focused approach to one that’s more automated, modular, and fit to address the entire AI lifecycle.
This best-in-class working framework, often called MLOps (Machine Learning Operations), enables organizations to take advantage of these advances and create a standard, company-wide AI ‘factory’ capable of achieving scale. It ensures your AI modeling and implementation withstands the test of time and that the performance of your AI solutions does not degrade to the point of inutility. MLOps is relatively new and still evolving and encompasses the entire AI lifecycle – data management, model development and deployment, and live model operations.
Building an MLOps capability will materially shift how data scientists, engineers, and technologists work as they move from bespoke builds to a more industrialized production approach. The business impact of MLOps is not just about productivity and speed but also improving reliability and reducing risk while refining talent acquisition and retention.
Scale to other business domains :
Once the business sees proof of impact and the organization becomes familiar with the new agile way of working, the company is ready to scale AI to other domains. Ideally, these are domains in which either data or assets can be reused, such as expanding a supply chain across multiple business units, or similar customer journey mapping can be applied to another business area. An excellent example of the latter is typical customer value management levers like next-product-to-buy or churn forecasting.
Successful implementation typically also means fostering a team of ‘advanced analytics’ practitioners, which comes with its own potential pitfalls. Depending on the starting point, it can be hard to hire and develop the right talent and capabilities internally. We suggest that analytics translators, or the people who determine corporate problems that can be solved through analytics solutions and work to implement them, be hired or fostered internally. Analytical modeling for initial use cases can be outsourced to specialized vendors to speed up delivery and impact. In parallel, companies can hire their own analytics talent and build an internal team in tandem with implementing more use cases, gradually scaling existing models, and developing new AI applications. This tends to work better than a “big bang” approach of acquiring a boutique AI firm, where valuations tend to be outsized, and integration issues are myriad
Tap into the Power of AI in Your Retail Business Today
AI is revolutionizing how retailers operate—and for the better. With AI, you put your store in a better position to make smarter decisions, boost sales, and ultimately enhance customer retention. So, now couldn’t be a better time to begin implementing AI in your daily functions.
Consider the above-mentioned steps as you explore how to incorporate this technology into your retail establishment. With the right AI tools, you can be well on your way to achieving a whole new level of business growth in the months and years ahead.
