Stop Throwing GenAI at Everything
Welcome to DX Brief - Retail, where every week, we review industry podcasts and reports to share what’s insightful and what you can do about it.
In today's issue:
Code vs Craft: Is AI replacing merchandising managers?
IBM's Agentic AI Playbook: retailers need to identify high-impact workflows now
Stop throwing GenAI at every problem: Infios' use-case framework for supply chain AI
1. Code vs Craft: Is AI replacing merchandising managers?
Shop Shock: Retail After AI podcast with Chuck Palmer and DeAnn Campbell, Episode: Code vs Craft: Which Delivers Better In-Store Merchandizing? (Nov. 12, 2025)
For: Store operations leaders, merchandising executives, retail strategy teams, category managers, transformation leaders, store experience designers
Background: Grocery retailers are eliminating merchandising manager positions and replacing them with AI planogram systems that optimize shelf placement, reduce waste by 5-10%, and free up significant operating budget. Meanwhile, the best retailers are discovering AI can generate creative merchandising variations, respond to hyperlocal signals like weather and events, and make intuitive creative decisions that historically required human judgment. The debate isn't AI vs. humans anymore – it's which decisions require human craft and which benefit from algorithmic speed.
TLDR:
AI planogram optimization is eliminating middle management: grocery retailers deploying algorithmic shelf planning are achieving 5-10% waste reduction and removing store merchandising manager positions, fundamentally restructuring store operations economics.
The "craft" defense is eroding: AI can now generate creative merchandising variations, interpret trend signals (like "popular purple sweater" cycles), and make contextual decisions that merchandisers assumed were uniquely human capabilities requiring intuition and taste.
The new value is in hyperlocal responsiveness: AI's real advantage isn't replacing human creativity but enabling dynamic merchandising that responds to weather, local events, foot traffic patterns, and real-time inventory in ways static planograms never could.
The framework many retailers miss:
AI is restructuring store economics by eliminating merchandising management. Campbell identifies the trend clearly: "We're seeing grocery stores right now that are eliminating merchandising managers in favor of algorithmic planogram structures." The math: Eliminate manager salaries, reduce waste by 5-10%, and "you've freed up a huge pot of money."
Most retailers think about AI as augmenting human decision-making. Campbell sees it as fundamentally restructuring store operations. The question isn't "Will AI help merchandisers work better?" It's "What happens when the merchandising manager role disappears and algorithms own shelf planning?"
The "humans do creative work" defense doesn't hold anymore. Palmer was trained as a merchant: "I was taught to be very intuitive about it – look at the numbers but interpret them in a creative way. We all know the popular purple sweater is only going to be popular for so long. You need the next thing in the queue."
Initially he thought: "AI can help with the numbers – predictive analytics, SKU rationalization, hard data. The human part is creative intuition, understanding what silhouette makes sense on the street."
But Palmer realizes AI is encroaching: "I could see where AI could start to bleed into areas where we start to give it more of that experimentation. We can generate 50 or 100 variations of merchandising and then test those against each other."
The framework shift: Merchandisers assumed taste, intuition, and trend interpretation were uniquely human. But if AI can generate 100 merchandising variations and A/B test them, it's not replacing human creativity – it's operating at a scale and speed humans can't match.
The defense "humans are creative" only works if creativity can't be generated algorithmically and tested empirically.
But technology only matters after fundamentals are fixed. Campbell's warning: "The basics of retail – quality products, clean stores, timely service – if you don't get these right, focusing on gimmicky things feels pointless."
The framework: Most retailers think technology creates competitive advantage. Palmer and Campbell think technology amplifies fundamentals. If your stores are dirty, service is slow, and products are out of stock, no amount of AI merchandising will save you. But if fundamentals are solid, AI merchandising can create convenience and responsiveness that feels magical.
What to do about this:
→ Evaluate merchandising manager ROI against algorithmic alternatives now. Calculate fully-loaded cost of your store merchandising management team. Model 5-10% waste reduction from AI planogram optimization. Run a 90-day pilot in 5 stores with algorithm-driven merchandising and compare financial outcomes. If algorithms deliver comparable or better results, redeploy human talent to customer-facing roles.
→ Test AI-generated merchandising variations in low-risk categories. Select one category where you have strong sales data and limited emotional attachment (e.g., batteries, paper goods, cleaning supplies). Have AI generate 10 planogram variations. A/B test them across 20 stores for 30 days. Measure which configurations drive higher sales per square foot.
2. IBM's Agentic AI Playbook: retailers need to identify high-impact workflows now
RetailWire podcast, Episode: Leadership Interview with Karl Haller, IBM (Nov. 11, 2025)
For: Chief Digital Officers, retail transformation leaders, enterprise architects, merchandising and operations executives, omnichannel strategy teams, retail technology decision-makers
The headline: Karl Haller, Partner at IBM Consulting, warns for retail executives: the performance gap between retailers implementing AI now versus those waiting is becoming "so great that it's going to be hard to be a fast follower." By 2027, AI-powered shopping will be commonplace, and retailers who aren't reinventing their core workflows now will face a self-fulfilling prophecy where poor business performance prevents the investment needed to catch up. IBM's seeing this play out already among its clients. Here's the framework separating winners from laggards.
TLDR:
Agentic AI takes actions, doesn’t just give answers: The difference between "ask and answer" (GenAI) versus "task" (agentic AI) determines which workflows to transform first. Focus on repeatable processes across hundreds of people and locations where you need consistent execution, not human prompting.
Shift from "little i innovation" to "capital I Innovation" on core workflows: Stop experimenting with chatbots and focus on reinventing merchandising, assortment planning, allocation, and replenishment; these are the workflows where AI creates actual productivity gains and competitive advantage.
The 2026-2027 timeline is real and forcing moves now: Consumer AI shopping will hit tipping points in holiday 2026 and become commonplace by 2027. Retailers implementing agentic AI today are creating performance gaps that competitors won’t be able to close.
The framework many retailers miss:
Understand the business distinction between GenAI and agentic AI. GenAI is "ask and answer": you ask it to create something (words, images, code, analysis). Agentic AI is "task": it takes action on your behalf by ingesting information, reasoning about it, and executing a series of actions in your systems or presenting you with options to approve.
Most retailers are still stuck thinking about AI as a chatbot that answers questions. The strategic shift is thinking about AI as an autonomous system that monitors, reasons, and acts, transforming workflows rather than just providing information.
IBM is seeing a critical shift among leading retailers: moving from "little i innovation" (lots of small experiments with AI) to "capital I Innovation" (reinventing the core business workflows that actually drive performance).
Agentic AI excels here because you need systems that continuously monitor data, reason about optimal actions, and either execute autonomously or present managers with options. Having hundreds of people manually analyze data and type prompts into GenAI tools doesn't scale. "If you need to do things consistently multiple times, having people type prompts into a GenAI system and get answers back is probably not the best way to do it."
Prepare for consumer AI shopping to be commonplace by 2027, with major tipping points in late 2026. Haller's timeline is specific and grounded in adoption patterns from the ecommerce era. He expects steady growth among early adopters throughout 2025, a tipping point around gift purchases in holiday 2025 (early adopters and fast followers), steady growth through 2026 as more of the bell curve adopts, then a major tipping point in holiday 2026 when AI shopping for gifts becomes mainstream.
Why gift purchases? "Gift purchases are notoriously hard and it's a good use case for AI at multiple steps in the shopping process." This mirrors how ecommerce adoption accelerated despite initial skepticism about buying certain categories online. People said no one would buy clothing, cars, furniture, appliances, or glasses online, but it turned out that for almost any product, some people will buy it online and some won't.
The real warning: performance gaps are becoming self-fulfilling prophecies. While consumer adoption will follow predictable curves, internal retailer implementation moves slower but the consequences of waiting are severe. "Given the actual productivity gains we're seeing among retailers and brands who are moving faster, the gap in business performance might get to be so great that it's going to be hard to be a fast follower in this game and it might be hard to catch up."
This isn't about technology adoption difficulty; it's about business performance. If your competitors achieve 20-30% productivity gains from agentic AI in core workflows while you wait, their superior performance generates more cash to invest in further improvements. Meanwhile, declining performance reduces your resources for investment, creating a downward spiral.
What to do about this:
→ Map your major workflows and identify the top 3-5 where consistent execution at scale matters most. Create a list of your core business processes: merchandising and assortment, allocation and replenishment, pricing and promotions, workforce scheduling, supplier collaboration, customer service operations. For each, document: volume of decisions/actions per day, number of people/locations involved, cost of inconsistency or delay, current process bottlenecks. Rank by potential impact from agentic AI automation.
→ Audit whether your current AI strategy focuses on "ask and answer" or "task." Review your current AI initiatives and roadmap. Categorize each as: (1) GenAI for content/analysis ("ask and answer"), (2) Agentic AI for autonomous action ("task"), or (3) Traditional AI/ML for optimization. If more than 50% are category 1, you're likely underinvesting in agentic AI that drives actual productivity gains.
→ Prepare for AI-powered shopping interfaces by auditing your content, commerce, and data infrastructure. As consumers begin using AI agents for shopping, your product data, content, and systems need to be structured for AI consumption, not just human browsing. Audit: (1) Product data quality and API accessibility, (2) Content optimization for AI retrieval (moving beyond SEO to what's being called GEO or AEO), (3) Transaction APIs that can work with third-party AI agents, (4) Inventory visibility and real-time data. Start addressing gaps in 2025 so you're ready when AI shopping scales in 2026-2027.
3. Stop throwing GenAI at every problem: Infios' use-case framework for supply chain AI
The Agile Brand with Greg Kihlstrom podcast, Episode: Infios Chief Innovation Officer Eugene Amigud on use-case driven success with AI (Nov. 14, 2025)
For: Supply chain transformation leaders, operations executives, warehouse optimization teams, retail technology strategists, omnichannel operations directors
Background: When Eugene Amigud joined Infios as Chief Innovation Officer, he inherited a common problem: everyone wanted to use GenAI for everything. His solution? A use-case driven framework that matches the right AI tool to the right problem. The results: agentic bots monitoring carrier deliveries and proactively calling drivers, warehouse optimizers processing 100,000 scans per day in milliseconds, and ML forecasting systems that barely hallucinate at all.
TLDR:
Match AI tools to use-case constraints, not hype: GenAI fails in millisecond-response warehouse environments processing 100K daily scans, but excels at monitoring delayed carrier deliveries and initiating driver calls without human intervention.
Build a comprehensive AI platform but apply tools surgically: warehouse path optimization needs fast optimizers, not thinking models; inventory forecasting needs ML algorithms with historical data, not generative AI that hallucinates.
Real-time operational constraints dictate technology choices: when 100,000 warehouse employees across multiple locations are scanning shipments, you can't have ChatGPT "thinking more to give a better answer"; you need millisecond responses from purpose-built systems.
The framework many supply chain leaders miss:
Start with the use case, not the technology… then match constraints to capabilities. Eugene's approach begins with a simple question: what is the actual business problem and what are its constraints? In warehouses, scanners process 100,000 shipments per day for a single customer during peak season. Response time requirements are in low milliseconds. "I cannot think more to give you a better answer when the person is scanning and I have 100,000 employees across various locations just fulfilling everyone's Christmas presents."
This real-time constraint immediately eliminates certain AI solutions. GenAI models that "think" to provide better answers simply cannot work in this environment. Instead, Infios deploys sophisticated optimizers for warehouse picking paths – AI systems that analyze historical data and current conditions to provide instant route optimization.
Most retailers make the mistake of starting with "we need to implement GenAI" rather than "we need to solve this specific operational problem with these specific constraints." When you invert the question, different problems reveal different optimal solutions.
Deploy agentic AI where human monitoring creates bottlenecks, not where speed is critical. Eugene's team introduced agentic bots for carrier monitoring – a use case with completely different constraints than warehouse operations. When a carrier leaves Albany, New York heading to Chicago and there's no update two hours later, the old process required manual monitoring, call center staff contacting drivers, or drivers updating apps while actively driving.
The agentic AI solution monitors orders continuously, initiates calls to drivers autonomously, and updates systems proactively "with not a single human on the shipper side being involved." This works because the constraint isn't millisecond response time; it's consistent, reliable monitoring across hundreds or thousands of in-transit shipments.
The strategic principle: GenAI excels when you need intelligent, autonomous action on repeatable workflows where slight delays are acceptable. It fails when you need instant responses at massive scale.
Use machine learning for forecasting where accuracy matters more than novelty. For inventory optimization – determining how much product each store needs during Halloween versus other seasons – Infios relies on ML algorithms analyzing historical data. "There are lots of very solid algorithms that guess what? Those algorithms barely hallucinate at all versus GenAI where there's still hallucination."
Most retailers have jumped to GenAI for everything because it's the current hype cycle. But Eugene's framework reveals that comprehensive AI platform capabilities need to include optimizers, ML forecasting, and GenAI. Then apply each where its strengths match the use case requirements.
What to do about this:
→ Audit your current AI initiatives through the constraint lens. For each AI project or proposal, document: (1) What is the actual business problem? (2) What are the response time requirements? (3) Is consistency or creativity more important? (4) What happens if the system hallucinates or makes an error? This audit will reveal which projects are using the wrong AI approach.
→ Start with one agentic AI use case where human monitoring creates obvious bottlenecks. Identify one workflow where staff manually monitor status, make routine phone calls, or update systems after receiving information. Pilot an agentic bot that can autonomously monitor, communicate, and update.
→ Build a comprehensive AI platform strategy, but implement surgically by use case. Don't limit yourself to one AI approach, but don't try to deploy everything simultaneously. Develop a 12-month roadmap that sequences AI implementations based on: business impact, technical feasibility, and learning value. Start with 2-3 high-impact use cases using different AI approaches to build organizational capability.
Disclaimer
This newsletter is for informational purposes only and summarizes public sources and podcast discussions at a high level. It is not legal, financial, tax, security, or implementation advice, and it does not endorse any product, vendor, or approach. Retail environments, laws, and technologies change quickly; details may be incomplete or out of date. Always validate requirements, security, data protection, labor, and accessibility implications for your organization, and consult qualified advisors before making decisions or changes. All trademarks and brands are the property of their respective owners.
