The recent State of AI in Business 2025 report from MIT/Project NANDA opened with a shocking statement: “Despite $30 billion to $40 billion in enterprise investment into [generative artificial intelligence], this report uncovers a surprising result in that 95% of organizations are getting zero return.”

Those numbers made lots of headlines — and probably induced lots of nightmares — particularly in tight-margined sectors like retail, where no one can afford to gamble on tech that doesn’t quickly and measurably produce return on investment.

The 26-page report is more nuanced than the hubbub surrounding it suggests, but according to lead author Aditya Challapally, the core issue in the enormous failure rate with AI is a “learning gap” for both tools and organizations that points to flawed enterprise integration. 

Learning gaps and flaws generally stem from misunderstanding the technology (thinking it replicates humans with automated processes) and focusing on high-visibility rather than high-value functions. This translates to approaching AI the wrong way to solve the wrong problems.

One of the primary questions about integrating AI into a business concerns the human in the loop (HITL), and where and how people are involved in an AI workflow to ensure efficiency, accuracy, safety and accountability in automated processes. You can definitely use AI to augment and automate specific tasks, reduce friction, increase quality and capture impressive efficiencies. You can also use it to produce, as the report demonstrates, absolutely nothing of value.

For effective business use, AI integration requires understanding the outcome the HITL is working towards, the process they own, and their workflow — then pairing the technology’s capabilities to distinct opportunities for accelerating that workflow, process and outcome. A well-integrated tool helps the human to do their work more efficiently. This translates to approaching AI the right way to solve the right problems.

The MIT report detailed several commonalities among the fortunate 5% of organizations achieving returns on AI investment. These include engaging in specialized external partnerships (rather than attempting internal builds), combined with deep customization aligned to internal processes and data, and benchmarking tied to operational outcomes sourced from frontline managers and staff. As the report notes, “this bottom-up sourcing, paired with executive accountability, accelerated adoption while preserving operational fit.” 

Interestingly, the report states, “back-office deployments often delivered faster payback periods and clearer cost reductions,” and ROI was “often highest in ignored functions like operations…. Notably, these gains came without material workforce reduction. Tools accelerated work, but did not change team structures or budgets.”

So what does all that actually look like in the real-world of retail? One unsung operational opportunity for productive AI implementation involves warehousing optimization.

Consider the case of a niche sports and entertainment brand and retailer with a thriving line of business receiving, tracking and reselling pre-owned sports equipment. It operates one large warehouse in Southern California where all inventory arrives and must be processed.

The company’s traditional warehouse setup involves organizing incoming supply through a kind of assembly line, where several people serve as data quality reviewers. The warehouse workers perform an audit on received items and photograph, catalog and classify them: Are they in mint condition? Are there dents or scratches? Does they have high-end features? New modifications? All these details matter, because they determine how the item gets positioned on the website and what price it commands.

The warehouse isn’t set up for full automation — it’s not some high-tech Amazon-like facility. It’s simply a large space with tons of aisles, lots of equipment, and some organizational structure that requires significant human involvement. Scaling this line of business runs into physical limitations: More workers and more space would be required to move more inventory more quickly. 

Their solution is to scale, using AI integration that’s customized to address specific constraints and needs. Warehouse workers still receive, photograph and manage incoming inventory, but they use GenAI computer-vision analysis to augment intake and audit processes. Based on comprehensive data about golf club characteristics, for example, the AI system automatically generates quality data that’s formatted to suit their requirements, including images and item descriptions, condition and damage assessments, and classifications based on traits and materials. The company’s warehouse workers still own and manage the process. But it takes a fraction of the time previously required to move product from intake to resale. The result: same number of people, same space, much faster throughput and way more efficient. 

The company is currently centralizing its data through IT to facilitate expanding and improving these kinds of AI-driven efficiency gains in other internal business units, which ironically creates different scaling limitations. Additional requests to support major new initiatives (from, say, manufacturing or compliance or sales) all funnel to the same small data team, and workflow challenges multiply. The company is a sports and entertainment brand, after all, not a computer science outfit. 

This is where AI expert partnerships can prove beneficial, so long as they’re strategically aligned with company goals and adaptable to internal data processes, as opposed to just building for building’s sake. The key is emphasizing outcomes over outputs: It doesn’t matter how many AI projects and data pipelines you’re building if they don’t support meaningful business objectives and can’t evolve with your operations.

The MIT AI report should serve as both a warning and guide to industry in the age of AI. Retail is a sector where operational and service efficiency directly drive profitability. Companies that focus on efficiency and measure success through operational impact rather than technological sophistication will succeed in turning AI investment into real returns.