The hype cycle is loud, the results are quieter
Every retail technology vendor now has an AI story. Commerce platforms, marketing tools, supply chain systems, customer service software: all of them have added “AI-powered” to their pitch decks. The conference circuit is saturated with keynotes about how AI will transform retail. And yet, when I work with retailers on the ground, the gap between what is being sold and what is actually delivering measurable results remains enormous.
This is not to say AI is not useful in retail. It is. But the use cases that deliver real operational value are almost never the ones that get the most airtime. The highest-ROI applications are unglamorous, operationally focused, and rarely make good keynote material. That is exactly why they work: they solve specific problems with measurable outcomes, rather than promising a vague transformation that never quite materialises.
Where AI delivers real value today
Demand forecasting and inventory optimisation
This is the single highest-value AI application in retail right now, and it has been for several years. Machine learning models that incorporate historical sales data, seasonality, promotional calendars, weather patterns, and external signals consistently outperform traditional statistical forecasting methods.
The numbers are meaningful. Retailers implementing modern demand forecasting typically see 15% to 30% improvements in forecast accuracy, which translates directly to reduced markdowns, fewer out-of-stock events, and better inventory allocation across locations and channels. For a retailer doing $200M in revenue, a 20% improvement in forecast accuracy can easily represent $2M to $5M in recovered margin annually through reduced markdowns and stockouts alone.
The key is that this works because the data inputs are relatively clean and structured. Sales history, stock levels, and promotional calendars are data most retailers already have in reasonable shape. The AI layer adds genuine predictive power on top of a solid data foundation.
Customer service automation
AI-powered customer service, specifically conversational AI for handling routine enquiries, is the second most reliably valuable application I see in retail. The technology has matured significantly. Modern implementations handle 40% to 60% of routine customer contacts without human intervention, covering order status, returns, shipping enquiries, and basic product questions.
The ROI calculation is straightforward. If your average cost per human-handled contact is $8 to $12, and AI handles half of your routine volume, the savings scale linearly with your contact volume. A retailer handling 50,000 contacts per month can realistically save $200K to $350K annually while actually improving response times for customers.
The caveat is that this only works well for structured, repetitive enquiries. Complex complaints, emotional escalations, and nuanced product advice still need humans. The retailers who get this right are the ones who design the handoff between AI and human agents thoughtfully, rather than trying to automate everything.
Content generation and creative production
This is the area where AI has made the most dramatic recent progress and where the ROI is most immediately tangible for marketing and merchandising teams. Product descriptions, email copy, social media content, and basic creative variations can now be generated at a fraction of the time and cost of purely manual production.
I have worked with retailers who reduced their product description writing time by 70% using AI generation with human review. Their catalogue team went from producing 50 product descriptions per day to 200, with consistent quality. For businesses managing thousands of SKUs across multiple channels, this is not a marginal improvement. It fundamentally changes the economics of content production.
The practical advice here is to use AI as a first draft engine with human editorial oversight, not as a fully autonomous content creator. The quality is good enough to accelerate production dramatically but not good enough to publish without review, particularly for brand-sensitive content.
Search and merchandising
AI-powered site search and merchandising is another area delivering real results. Natural language search that understands intent rather than just matching keywords, automated product ranking based on conversion data, and visual search capabilities are all production-ready and delivering measurable improvements in conversion rates.
Retailers implementing AI-driven search typically see 10% to 20% improvements in search conversion rates. For an ecommerce business where 30% of revenue comes through on-site search, that improvement drops directly to the bottom line. The technology is mature, the implementation is relatively straightforward, and the results are measurable within weeks of deployment.
Where AI is still mostly theatre
Hyper-personalisation at scale
This is the most oversold AI capability in retail. Every CDP and marketing platform vendor will tell you that AI can deliver truly personalised experiences to every customer at every touchpoint. The reality for most mid-market retailers is far more modest.
Genuine personalisation at the level vendors promise requires three things most retailers do not have: a unified customer data platform with clean, comprehensive profiles; sufficient behavioural data density to train models that can distinguish meaningful segments from noise; and the operational infrastructure to act on personalisation signals across channels in real time.
Most retailers I assess have fragmented customer data, inconsistent identity resolution, and limited ability to activate personalisation signals beyond basic email segmentation. The AI is not the bottleneck. The data foundation is. Investing in a sophisticated personalisation engine when your customer data is spread across six disconnected systems is like buying a racing car before you have built the road.
Start with solid segmentation and targeted recommendations based on purchase history. That alone outperforms most “AI-powered personalisation” implementations I have seen.
Autonomous pricing optimisation
Dynamic pricing powered by AI is another area where the vendor pitch runs well ahead of practical reality for most retailers. The models exist. They work in controlled environments. But deploying autonomous pricing in a retail context involves competitive dynamics, brand perception, channel conflict, and promotional commitments that make fully automated pricing impractical for most businesses.
What works is AI-assisted pricing: models that recommend price adjustments based on demand elasticity, competitive data, and inventory levels, with human review and approval. Fully autonomous pricing, where the algorithm sets and changes prices without human oversight, is appropriate for a very small number of high-volume, commodity-like categories. For most retail, it creates more problems than it solves.
”AI-powered” everything
Be sceptical of any vendor who has retrofitted “AI-powered” onto an existing product without a clear explanation of what the AI actually does and what measurable improvement it delivers. I review vendor proposals regularly where the AI component amounts to a basic rules engine with a machine learning label, or a feature that uses a simple statistical model that has existed for a decade but has been rebranded as AI.
Ask three questions: what specific model or approach is being used? What training data does it require, and do we have it? What measurable improvement should we expect, and how will we validate it? If the vendor cannot answer all three clearly, the “AI” is marketing, not technology.
The data foundation problem
The single biggest barrier to AI adoption in retail is not technology. It is data. Every AI application depends on data quality, and most mid-market retailers have significant data problems that need to be addressed before any AI investment will deliver its promised returns.
The common issues are predictable: product data is inconsistent across channels, customer identity is fragmented across systems, inventory data has latency issues, and historical data is trapped in legacy systems with no clean extraction path. These are not glamorous problems to solve, but they are prerequisites.
Before investing in AI, every retailer should honestly assess their data readiness across four dimensions. First, data quality: is the data accurate, complete, and consistent? Second, data accessibility: can the data be extracted and processed without manual effort? Third, data unification: is there a single source of truth for customers, products, and inventory? Fourth, data governance: who owns the data, and are there clear processes for maintaining its quality?
If you score poorly on any of these dimensions, fix the data first. An AI model trained on poor data does not give you AI. It gives you automated bad decisions.
A practical adoption framework
The retailers I have seen succeed with AI follow a consistent pattern. They do not try to boil the ocean.
Start by assessing your data readiness against the specific use case you are considering. Pick one use case with clear, measurable operational impact and a solid data foundation to support it. Demand forecasting and customer service automation are the most common starting points because the data requirements are manageable and the ROI is measurable.
Run a focused pilot: 8 to 12 weeks, clear success criteria, specific metrics. Measure ruthlessly. If the pilot delivers the expected results, build the business case for scaling. If it does not, understand why before investing further.
Scale investment based on proven results. The retailers who fail at AI are the ones who commit to enterprise-wide AI platforms before they have validated a single use case. The ones who succeed build a portfolio of AI applications incrementally, each one justified by demonstrated ROI from the previous investment.
Evaluating AI vendor claims
When a vendor tells you their solution is “AI-powered,” ask for case studies with specific numbers from businesses similar to yours. Not testimonials. Not analyst endorsements. Actual metrics: what was the baseline, what improved, by how much, over what timeframe.
Ask for a pilot structure with clear success criteria before committing to a multi-year contract. Any vendor confident in their AI capability will agree to this. Vendors who insist on long-term commitments before proving value are selling you a contract, not a solution.
Ask who else in your segment is using the product in production, not in pilot. There is a meaningful difference between a proof of concept and an operational deployment, and many AI products in retail are stuck permanently in the former.
The organisational change angle
AI changes workflows, not just technology. A demand forecasting model that produces better predictions is useless if the planning team does not trust it enough to act on its recommendations. Customer service automation that handles half your volume requires rethinking how you staff, train, and manage your human agents.
The organisational change required to adopt AI successfully is consistently underestimated. Budget for training. Budget for process redesign. Budget for the transition period where the old and new approaches run in parallel. And most importantly, involve the operational teams who will use the AI tools from the beginning of the project, not just at the end.
The technology is rarely the hard part. Getting people to change how they work is.
Next steps
If you are evaluating where AI can deliver genuine operational value in your retail business, or assessing AI maturity across a portfolio of retail investments, get in touch. I help retailers and investors cut through vendor hype to identify the specific AI applications that will deliver measurable results, given their current data foundation and operational maturity. The starting point is always the same: understand the data, pick one problem, prove the value.
What to read next
- Do you actually need a CDP? — Understanding whether a CDP fits your stack, and what data foundations need to be in place first.
- AI in retail: separating what works from what’s theatre — A broader look at which AI applications are actually delivering results in retail.