AI in retail (part 2): Powering the operational engine

While much of the industry buzz remains on the customer experience, the real transformation is happening within operations.

In Part 1 – Better experiences, but bigger risks, we’ve seen how retailers can invest in AI to elevate the customer experience and increase revenue. In this article, we will examine ways to improve operational efficiency and increase employee productivity using AI. We will also explore the risks of poorly implemented automated processes and how they can affect companies in the long run. It’s not just about doing things faster, but making smarter, value-driven decisions that modernize the entire process.

Where AI is reshaping retail operations

Smart inventory

Overstocking costs money, and understocking loses sales. With the help of AI, retailers can reduce some of the stress of predicting inventory needs.

By combining real-time sales data, seasonal trends, weather patterns, and even social media signals, AI can now predict demand with a level of accuracy that traditional forecasting cannot match. This means fewer markdowns, less waste, and fewer “out of stock” moments that can push customers to a competitor.

Zara is a strong example of a brand that has built its competitive edge around supply chain efficiency. Their AI-driven inventory model uses real-time demand signals to decide what to produce and in what quantities. The result is a leaner supply chain with dramatically reduced overstock.

Beyond inventory, AI can also monitor the entire supplier ecosystem, flagging risks such as shipment delays, supplier instability, or geopolitical disruptions before they cause downstream problems. This form of AI-powered demand forecasting enables retailers to improve inventory management, reduce stockouts, optimize working capital, and strengthen overall supply chain performance.

Catalog enrichment

For retailers, it’s a challenge to manage thousands of articles and keep the data accurate and complete across multiple channels. AI can now automate much of this work. Using a combination of computer vision and natural language processing (NLP), it can analyze product images to extract attributes like color, material, and style, generate or improve product descriptions, standardize categorization across the catalog, and flag incomplete or inconsistent entries for review.

Maintaining well-structured data has a powerful effect throughout the organization. It means better search results, more accurate recommendations, fewer customer returns due to mismatched expectations, and faster time-to-market for new products. For large catalogs, what once required large content teams working manually can now be handled at scale, with humans reviewing and refining outputs rather than producing them from scratch.

Amazon is now using generative AI capabilities to help sellers create better product information. Sellers only need to provide a brief product description, and Amazon generates compelling titles, bullet points, high-quality descriptions, and product attributes for them to review. These new capabilities will help sellers create high-quality listings with less effort and present customers with more complete, consistent, and engaging product information.

Dynamic pricing

Dynamic pricing, which means adjusting prices in real time based on demand, competitor activity, and stock levels, has been used in travel and hospitality for a long time. AI is now making it practical for retail at scale.

What makes AI-driven pricing different from simple rule-based systems is context. Rather than just lowering prices when stock is high, it can factor in customer lifetime value, competitor moves, time of day, and even the device a customer is browsing from. For retailers with thousands of SKUs, this kind of granular optimization would be impossible to manage manually. MediaMarkt uses AI-driven digital price tags to adjust prices in real time based on seasonal and regional demand. During holiday seasons, this strategy has increased sales by about 20%. AI can help retailers capture opportunities and keep inventory needs in balance.

Sentiment analysis

Understanding how customers feel about a brand, product, or experience has always mattered, and now retailers can do so in real time. Sentiment analysis is the process of determining the emotional tone or attitude conveyed by a piece of text. These tools can monitor reviews, social media, support transcripts, and post-purchase surveys, surfacing patterns that would take a human team a long time to identify.

For operations, this has direct practical value. For example, a sudden spike in negative comments about a specific product can prompt the team to conduct a quality review before the problem escalates further. Recurring complaints about packaging, delivery, or checkout friction can be automatically flagged and assigned to the right team. Positive sentiment around a specific product or category can also inform inventory and promotional decisions.

Nike uses Natural Language Processing to understand the emotional drivers behind sports trends and sustainable fashion. Nike analyzes streams from social media and review platforms to identify the emotions associated with a specific athlete’s performance or community movements (like eco-ethical fashion). They use these insights to adjust their ad copy and product design in real-time, ensuring their messaging aligns with the current mood of their target demographic.

Agentic customer support

This topic was also covered in part 1 of the article because of its impact on customer satisfaction, but the benefits extend beyond customers, as it can be a game-changer for employee productivity.

AI can summarize the customer’s full history and act as a coach assistant in real time, suggesting responses and helping handle difficult conversations, which helps avoid escalation and stay consistent with brand guidelines. After the assistance, it can automatically log notes, update records, and trigger follow-up actions, which were time-consuming tasks for assistants. Besides that, since AI can handle a big volume of repetitive requests, humans can now use their time more effectively to handle interactions that genuinely need their judgment.

To address declining customer satisfaction and live chat wait times exceeding 4 minutes, H&M deployed a generative AI virtual assistant. It was developed using real-time inventory data, 10 million past support tickets, and brand guidelines. While this virtual assistant manages easy repetitive requests, human assistants are available to resolve more complex issues. In its first year, the virtual assistant handled 7 million contacts per month, automating 65% of the end-to-end workload. Average wait time fell to 40 seconds, first-contact resolution climbed to 88%, and the call center’s workload shrank by 40%.

Autonomous vendor negotiations

Procurement is a labor-intensive part of retail operations since supplier relationships are built on trust, context, and nuance. But a significant part of procurement activity involves doing repeated tasks such as renewing contracts, adjusting order volumes, or responding to price changes.

AI agents are now capable of handling these lower-stakes interactions autonomously. They can analyze historical pricing data, benchmark against market rates, assess supplier performance, and generate or respond to proposals, all within defined parameters set by the procurement team.

OTTO group is using an AI-powered software that handles renegotiations in the internal purchasing process for suppliers. It started by handling negotiations with smaller suppliers, but soon after, larger suppliers were also integrated. This allowed them to free employees to engage in more personal negotiations, complex contracts, and strategic planning, where human judgment and relationship-building can be a deal breaker.

The risks of getting it wrong

Speed without strategy

Companies are facing pressure to deploy AI quickly. But the ones that move fast without a clear strategy can end up automating processes that were already broken. AI not only doesn’t fix a flawed process, but can make it even more problematic by scaling it. A poorly designed returns workflow, for example, becomes a faster, but more frustrating return workflow.

Although most retailers are now engaged with AI, it doesn’t mean they are ready to use it. The companies seeing genuine good results are those that started with clear problem statements, not with the technology.

The expertise gap

There’s a big gap between how fast companies are adopting AI and how well-equipped their teams are to implement it. This gap creates real risk because when teams don’t fully understand how a system makes decisions, they can’t catch errors before they happen or trust the results too much.

Pricing models can make wrong predictions and impact margins. Demand forecasting can fail in unusual conditions, for example, a sudden trend, a supply shock, or a viral moment that falls outside the training data. Sentiment analysis can also create risks if models misread tone, sarcasm, or cultural nuance, leading to wrong operational decisions.

Building AI literacy across operations teams is foundational, but it’s also important to use transparent systems that inform and keep humans in the loop.

Data quality and model drift

AI systems are only as good as the data they’re trained on. Retailers operating across multiple markets, channels, and legacy systems often have fragmented, inconsistent data, and models built on that foundation will produce unreliable outputs. This is particularly important for catalog enrichment, where poor source data yields poor enriched data, and for sentiment analysis, where incomplete channel coverage creates blind spots. As market conditions evolve, models need continuous monitoring and retraining.

Over-automation and the loss of judgment

Not every operational decision benefits from full automation. Complex supplier negotiations, crisis response, and strategic purchasing decisions still require human judgment, relationships, and ethical consideration. When companies remove those touchpoints entirely in pursuit of efficiency, they lose the contextual awareness that prevents costly mistakes. AI can assist by enhancing our human capabilities, but it can never completely replace human creativity and decision-making.

Getting the balance right

There are plenty of reasons to improve retail operations using AI. The efficiency gains are real and measurable, but it is important to remember that efficiency serves as a tool rather than a final objective. Employees are the company’s greatest assets. Their work can be highly enhanced by AI, but not completely replaced. The retailers that will come out ahead are those who treat AI as a collaborative partner in operations by augmenting human decision-making and helping make employees’ work easier, faster, and more fulfilling. That means investing in the people and processes around AI as much as the technology itself.

A few principles worth keeping in mind:

• Start with the problem, not the tool. Define what success looks like before selecting a solution.
• Invest in data foundations. The quality of your AI output depends entirely on the quality of your input.
• Keep humans in the loop for decisions with significant risk.
• Measure what matters. Efficiency metrics only tell one part of the story. Track impact on customer and employee satisfaction, and long-term margin.

The companies that approach AI with this kind of intentionality won’t just reduce costs, but also build the operational resilience to compete at a higher level.

What comes next?

Are you exploring how AI can modernize your operations without losing sight of what matters? Let’s work through it together.