For Retail AI ROI, Focus on Outcomes Over Hype

Mike Haddon, the VP of Consumer Products, Retail, and Services Practice Lead at Capgemini Invent, outlines why companies should focus on outcomes over hype when assessing the ROI of their retail AI. This article originally appeared in Insight Jam, an enterprise IT community that enables human conversation on AI.

In the race to leverage AI, no one wants to be left behind. But are first movers’ AI investments delivering real, measurable business value? Right now, it doesn’t look like that’s the case. Instead, we see a sense of urgency among firms across the industry to experiment, leading to quick project approvals that aren’t designed with clear business outcomes in mind.

MIT’s 2025 report on the state of AI in business quantified the scope of this problem. “Despite $30–40 billion in enterprise investment into Gen AI […], 95 percent of organizations are getting zero return.” Another study found that 64 percent of retail organizations faced “major” challenges in scaling up their Gen AI pilot programs. As a result, many leaders at enterprises that have invested heavily in AI pilots are now frustrated at the lack of progress in AI implementation, scaling, and measurable impact.

AI does have the potential to deliver real value. The MIT report observed that the 5 percent of AI pilots that are successful are delivering “millions in value” to the organizations running them. For an organization’s AI program to join that 5 percent, leaders need to adopt outcomes-based strategies and address specific use cases within a value-focused framework.

Let Go of AI FOMO

The retail industry’s vision of what’s possible with AI can be skewed by headlines focusing on future capabilities rather than those with proven results. Press releases designed to wow Wall Street with a company’s AI projects can also drive fear of missing out (FOMO) among competitors. This kind of coverage can goad companies into copying splashy initiatives that may not deliver business value.

Meanwhile, we’re not reading as much about successful scaled AI adoption in retail. Often, companies that have achieved competitive and differentiation advantages with AI keep that information confidential. The key for other organizations is to focus on finding their own quiet, yet effective, use cases rather than chasing trends. For example, automating back-office processes with AI isn’t as visible as a customer-facing chat agent, but it can deliver measurable and compounding value quickly.

Strategize for Specific Business Outcomes

Creating a successful AI project portfolio requires a strategic framework that focuses on measurable business outcomes, defines stage gates, and outlines a path to scalability. This framework must rest on a well-organized, high-quality data foundation with appropriate governance.

Define your AI-driven business outcomes.

Clarify what your organization wants to accomplish with AI. Defining the business outcome will help determine which systems of record and data silos your AI will need to access. For example, if you have back-office use cases for a finance or procurement function, your project scope for data and process information is comparatively small. On the other hand, if you want to move the needle on your supply chain with AI, your project scope will be massive. Whatever your business metrics and value drivers are—customer lifetime gross margin improvement, revenue density per square foot of labor lift, experiments in influencing cash conversion cycle, etc.—your AI ambitions, use cases, and projects need to be grounded in tangible outcomes.

Create a data foundation for AI.

AI requires good data to work, so your data foundation is a critical investment in success. For example, agentic AI can only have a tangible impact if it can talk to many different systems and access their data to complete its tasks. Systems of master record must be able to speak to systems of experience, commerce, and service. AI agents, no matter how well designed, will struggle to deliver value if the underlying business data they need is siloed and not standardized. In fact, siloed and disorganized data is among the most common causes of AI proof-of-concept failure.

Governance and policy adherence are other areas where organizations tend to put the AI cart before the data horse. Less than half (46 percent) of organizations using AI say they have AI governance policies, according to a 2025 survey, and even among companies with AI governance rules, close to half say their employees don’t usually comply. Implementing and enforcing data governance policies should be a precursor to AI investments.

Set up stage gates and productivity measures.

Stage gates are crucial for ensuring business value from AI. Your first stage gate should focus on identifying the target process domains, the prioritized (by expected value) use cases, and the appropriate AI tools and technologies. If the projects don’t demonstrate a high likelihood of sustainable (and scalable) benefit, they don’t pass.

Next, decide how you’ll measure actual versus predicted benefit. This means identifying the needed data to both baseline the targeted business metric and measure the AI-enhanced results. If the data is not clean, well-governed, or from consistent, trusted sources, the project cannot pass. You’ve heard it a thousand times: garbage in, garbage out. Don’t trust that an AI tool or platform can overcome compromised or inaccurate data. If the data do not meet the gate requirements, the project cannot pass.

Tools and technology platforms are, of course, important, but should never be your first stage gate. They, and often the people who sell them, tend to oversimplify the effort, data complexity, and attention needed to achieve the promised results. And the promised results are not always based on real-world nuances and the complexity of your business. Demos are always run with simple and harmonized data, even when based on an extract of your data. Once you have established the scope of the desired improvement and the trusted data to power it, you are in a better position to determine the tools and technologies that might deliver. Where in the stage-gate framework your technology sits depends on the partnership between technology and business leaders in your organization.

Subsequent stage gates will be based on incremental performance, with the total number of gates depending on your organization and your desired pace. For example, suppose you have passed all previous stage gates for an inventory visibility AI project. You might want to test and learn with incremental inventory and transaction data loads from a specific region of stores and the supporting supply chain network nodes, representing 10 percent of your overall data. You might decide to test hypotheses and make incremental adjustments to AI parameters to measure results. If there is evidence of larger, sustained value—for example, better visibility leading to increased available-to-promise and on-shelf availability that can be attributed to 300 basis points of regional sales lift—then later-stage gates can be added to ramp up data ingestion.

A final note regarding stage-gating: Identify the technical and business roles with the autonomy and authority to decide how a project portfolio will move from one stage to the next. This step is essential to remain centered on business outcomes because it prevents vanity projects from consuming resources, provides vetted results to executive management, and enables faster and healthier deployment of enterprise-scale AI.

Look Past the Hype to Valuable Outcomes

There is no shortage of excitement, action, or ideas about how enterprises can use AI. The key to meaningful value and real ROI is not chasing tools, technology, or competitors’ claims. It’s channeling enthusiasm and vision into projects that support scalable business outcomes and deliver value quickly, even if they’re behind the scenes rather than visible to customers and the press.