Smart AI, Smarter Retail: Levi’s, Belstaff, and You
Are you over-investing in AI compared to other retailers, or are you falling behind? Take the anonymous “State of Retail DX” benchmarking survey to find out. (Only participants will get the exclusive, segmented data by size)
In today's issue:
New feature: Quick Takeaways
Levi's "Three Smarts" enterprise AI framework
Rethinking contact centers as more effective sales channels in retail
Enterprise retailers face an "Innovator's Dilemma" with Agentic AI
1. Quick Takeaways
The “everything transformation” mindset that saved Belstaff. Most retailers approach replatforming as a system-by-system exercise. When Navid Jilow joined Belstaff in 2021, the 100-year-old British heritage brand, his team took a different approach: they ripped out everything simultaneously. New ERP (NetSuite), new e-commerce (headless Shopify), new point-of-sale (Shopify POS), new integration layer (Patchworks), new processes, even relocated departments from Italy to the UK. Why? Because Belstaff's previous model had everything outsourced to a third party. When you inherit a black-box architecture, incremental change just perpetuates the visibility problem. The only path to control is greenfield.
From: How Belstaff Rebuilt Everything – The Tech Stack That Transformed a Heritage Brand, Patchworks TV (12/8/25)
AI bridges the knowledge gap every specialty retailer knows: customers assume they're getting a 40-hour-per-week associate with three years of experience. The reality? About 75% of staff work less than 20 hours and don't have that tenure. Footwear Guru addresses this by delivering instant expert-level knowledge to team members. A customer asks about the differences between three shoes in the Glycerin family? With the AI tool, team members get a table comparing the shoes in seconds – and then layer on whatever expertise they have. The shift is: less emphasis on memorizing answers, more on listening and asking questions.
From: The Running Event 2025 with Steve DeMoss (Big Peach Running Co.) and Rob Anderson (Run Free Project): The Accessible AI Playbook: Practical Tools to Inspire Retailers (12/8/25)
2. Levi's "Three Smarts" enterprise AI framework
Deloitte AI Institute with Dr. Katia Walsh (Chief Strategy and AI Officer, Levi Strauss & Company): Creating Value: Transforming Consumer Experiences with AI (Dec 11, 2025)
Background: Dr. Katia Walsh built Levi's AI capabilities from zero, starting alone and scaling to a global function. The result: Levi's has had one of its healthiest financial margins in years, grown to 5.3 million US loyalty members, and uses AI to predict consumer demand, brand strength, and optimal pricing. Here's the framework she's used across financial services, telecom, and now retail to drive transformation without getting stuck in "use case" purgatory.
TLDR:
Stop thinking in terms of "use cases": AI should be a core capability like marketing or finance, not a series of isolated experiments.
The "three smarts" framework drives Levi's AI strategy: smart commerce (predicting demand and pricing), smart connections (personalized customer engagement), and smart creativity (augmenting designers with AI).
Diversity isn't a nice-to-have, it's a bias mitigation strategy: diverse teams + diverse data + diverse tools = minimized AI bias.
Start with "why," not "what." Most organizations approach AI by hunting for use cases. Walsh argues this is fundamentally wrong: "We don't have use cases in marketing. We don't have use cases in product development. We don't have use cases in HR or finance. These are established functions. So why should we have use cases in what is now a core of doing business?"
The shift in mindset matters because it changes how you resource, communicate, and measure AI. When AI is a capability – not a project – you build teams, processes, and governance around it. You stop asking "what can AI do?" and start asking "how does AI enable everything we do?"
The "three smarts" provide a complete retail AI blueprint. At Levi's, Walsh's team operates across three domains.
Smart commerce uses AI to predict consumer demand, brand strength, and pricing. Walsh credits this directly for Levi's margin health.
Smart connections powers personalized messaging, experiences, and the loyalty program that grew to 5.3 million US members.
Smart creativity augments designers by processing thousands of images that no human could review alone.
The key insight: during the pandemic, when everyone assumed the world was unpredictable, Walsh's team built a SWAT team that met every day to prove "actually a lot in life is predictable." They predicted what consumers would need when stores reopened, what they would pay, and where demand would shift. That predictive advantage compounds over time.
Build a "data ocean" that marries data sets that have never met. Levi's built a live-streaming data repository that aggregates financial data, transactions, customer data (always with permissions), plus climate data, economic outlooks, epidemiological models, and social media trends.
Walsh says: "When you marry data that have never met each other in the past, the power is immense."
This isn't just about having more data. It's about combining internal operational data with external signals that reveal context. A sales dip looks different when you know there was a weather event or an economic shift in that region.
Upskilling is the diversity and transformation strategy rolled into one. Levi's upskilled 160 people in two years – store managers, designers, marketing people, distribution center workers.
The screening criteria weren't statistics or coding experience. They screened for curiosity, problem-solving, resilience, and perseverance. One of their best data scientists was formerly homeless.
Walsh's philosophy: "You can never feel bad about giving someone a chance. Even if they don't work out, you have given them a chance."
What to do about this:
→ Reframe your AI initiative from "use cases" to "capability." Present AI to your executive team as a function that touches everything – like finance or HR – not as a collection of experiments. This changes resourcing, reporting, and long-term investment thinking.
→ Map your "three smarts." Identify where AI currently touches commerce (demand, pricing, inventory), connections (customer engagement, personalization, loyalty), and creativity (product design, content generation). Prioritize the gap with highest margin impact.
→ Pilot an upskilling cohort that ignores traditional credentials. Select 5-10 employees from non-technical roles who demonstrate curiosity and problem-solving. Train them on data basics. Track how their domain expertise combined with data skills creates unique value.
3. Rethinking contact centers as more effective sales channels in retail
The Frictionless Experience with Joe Megibow, former CEO at Casper: The Biggest Mistake Retailers Make with Omnichannel Strategies (Dec 8, 2025)
Background: Joe Megibow has led digital transformation at Expedia, American Eagle, Casper, and Purple. At American Eagle, his team built a $100 million contact center sales channel that everyone said was impossible. At Expedia, he discovered contact centers convert at 30-40% – compared to 2-3% online. His core insight: most retailers treat omnichannel as a technology problem when it's actually a P&L alignment problem.
TLDR:
Contact centers aren't cost centers, they're conversion engines. Expedia's phone channel converted at 30-40% vs. 2-3% online because humans provide reassurance on high-stakes purchases that digital can't replicate.
Channel P&L misalignment creates customer friction. When a store associate gets no credit for an in-store mobile purchase, they're incentivized to add friction by redirecting customers to the register.
Sometimes adding friction reduces friction. For considered purchases like mattresses, encouraging customers to visit stores actually accelerates the overall purchase journey.
Stop treating contact centers as failure points – they're your highest-converting channel. When Megibow arrived at Expedia from his digital-native TeaLeaf background, he was shocked to discover 30% of sales came through the contact center. "I viewed that as a bug, not a feature. Like, what are we doing so badly?"
Then he started listening to calls. The pattern was simple: customers booking expensive family vacations needed human reassurance. "This hotel is great. You're going to love it. I've heard what you're looking for." The customer would exhale, hand over their credit card, and the sale would close. "It was a moment of human empathy, human interaction, understanding."
The data was stark: 2-3% conversion online versus 30-40% through the contact center. Average order values 1.5x higher. More complex, higher-margin purchases.
At American Eagle, Megibow was "almost laughed out of the room" when he proposed training contact center agents to sell apparel. "Contact center people can't sell jeans." But they staffed 35,000 store associates and trained them to sell. Why couldn't they train 50 agents? By the time he left, that channel was generating $100 million in sales.
Most retailers believe the contact center exists because of digital friction failures. Megibow argues it's the opposite: "I think that is catastrophically wrong. I think removing it is a deliberate introduction of friction for many customers at many moments in their consumer journey."
Fix your P&L structure before optimizing your customer journey. Walk into an American Eagle store, find the perfect jeans in the wrong size, and pull up your phone to buy online. The associate interrupts: "Don't do that. Come with me to the register. I can get that for you here." You resist. "If you do it there, you're going to have to pay shipping. I can waive shipping at the register."
Two purchase paths. Different rules. Different friction. Why? Because if you buy on your phone while standing in their store, that store gets zero credit for the sale. "I just borrowed their real estate and oxygen and tried on their product and they get no credit." Go to the register, and the store gets credit, even though it ships from the same warehouse to the same customer.
"We've created all these channel dynamics around how we operate as a company that have nothing to do with the consumer." Separate buyers for online and stores. Separate P&Ls. Separate inventory views. But it's the same customer, looking at both sets of inventory simultaneously.
The solution isn't better technology, it's organizational alignment. Until your P&L structure treats the customer as a single entity across channels, your teams are incentivized to create friction.
For considered purchases, sometimes adding friction accelerates the journey. When Megibow moved from apparel to mattresses (Casper, Purple), everything he knew about conversion optimization broke. You can't acquire a mattress customer and expect a second or third transaction to justify the acquisition cost. "If you can't acquire someone and sell them a mattress and make money on that as a one and done, you don't have a healthy business."
More importantly, the purchase journey is fundamentally different. "On a considered purchase... friction doesn't mean I got to acquire them to the site and convert them in that first session or I'm done." These customers are noodling for 14-18 days. The best move? Encourage them to visit a store. "If I can get them to do that, I'm definitely going to get the sale."
If you're measuring stepwise conversion on-site, sending customers to stores looks like failure. "That's killing my conversion." But step back to the full journey, and it's the lowest-friction path to the sale.
What to do about this:
→ Audit your contact center for conversion potential. Pull your phone channel conversion rates versus online. If there's a significant gap, investigate whether you're underinvesting in a high-converting channel by treating it as a cost center.
→ Segment your purchase journey analysis by consideration cycle. For high-consideration purchases, measure success across the full 14-21 day journey, not session-level conversion, and test whether deliberate friction (store visits, human conversations) actually accelerates conversion.
→ Test contact center sales capabilities. If you can train 35,000 store associates to sell product, pilot training 20-50 contact center agents with the same knowledge and measure conversion impact.
4. Enterprise retailers face an "Innovator's Dilemma" with Agentic AI
VTEX Live podcast, Episode: Building Trust in the Age of Agentic AI with Sucharita Kodali (Nov. 25, 2025)
Background: While everyone is hyperventilating about "Agentic AI" (autonomous systems that execute complex goals), Forrester’s Principal Analyst Sucharita Kodali warns that most retailers are nowhere near ready for it. Why? Because while we have the tech, we lack the "Trust Protocol." She compares retail AI to the TSA: we have the technology for fully automated airport security, yet we still force humans to check IDs because we don't trust the machine's failure rate. Here is the framework for deploying agents that won't destroy your P&L.
TLDR:
The "1-Hour Task" Limit: Current Agentic AI excels at discrete tasks that take a human ~1 hour. Beyond that, complexity causes failure.
The Trust Gap: You cannot deploy autonomous agents in high-stakes inventory buying without a "Mission-Led" protocol that admits failure.
The Amazon Exception: Amazon’s "hands-off" merchandising works because they are a marketplace with infinite shelf space; physical retailers have limited shelf space, making AI errors fatal.
The "1-Hour Task" Constraint Kodali provides a sober operational benchmark: "Agentic solutions... can take discrete one-hour tasks that normally a human would take about an hour to do and... is pretty good at doing."
Most retailers try to apply AI to massive, multi-variable problems like "Build me a house" or "Plan my entire Q4 merchandising strategy." This fails.
The current framework for success is identifying high-volume, low-risk tasks that fit inside a 1-hour human execution window, like allocating a daily marketing budget across four specific channels.
If the decision tree extends beyond that timeframe or involves cross-system dependency, the "agent" degrades into a hallucinating chatbot.
The "Mission-Led" Agent Protocol The host, Santiago Naranjo, proposes a solution to the "overselling" of AI reliability: Mission-Led Agents. Instead of promising a "set and forget" finance reconciliation agent, builders must deploy agents with a transparency dashboard.
The Framework: "I have 10 agents attempting this task. 6 will succeed. 4 will fail. The system will explicitly tell you which 4 failed."
The Shift: Trust isn't about perfection; it's about accurate failure reporting. If a retailer knows the agent will self-report when it hits a confidence threshold below 90%, they will authorize the tool. If the tool pretends to be perfect, Procurement will kill it.
The Inventory Risk Matrix Kodali draws a sharp distinction between Marketing AI (low risk) and Merchandising AI (high risk).
Amazon uses agentic merchandising because they have an "endless aisle." If the AI prices an item wrong or surfaces a niche product, the cost is low.
For a traditional retailer with physical shelves, an agentic mistake means dead stock and wasted square footage.
The Rule: Do not use fully autonomous agents for inventory purchasing decisions where shelf space is the constraint. Use them only for pricing and allocation where the inventory is already committed.
What to do about this:
→ Audit your AI pilots against the "1-Hour Rule." Review your current automation roadmap. Are you trying to automate "Strategic Planning" (fails) or "Daily Budget Reallocation" (succeeds)? Break every AI project down into 1-hour discrete tasks.
→ Demand "Failure Rate" SLAs from Vendors. When evaluating AI vendors, stop asking "Can it do X?" Ask "What is the self-reported failure rate, and how does the system alert a human when it fails?" If the vendor claims 100% accuracy, walk away.
→ Deploy "Shadow Mode" for Merchandising. For high-risk decisions (inventory buying), run the AI agent in shadow mode for 6 months. Compare its "buys" against your human buyers. Only turn on autonomy when the agent outperforms the human on markdown avoidance for 2 consecutive quarters.
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.
