Amazon vs Walmart: AI Lessons for Retail Ops in SG

Amazon’s physical stores segment is expected to hit US$5.9 billion in Q4 revenue, up 5.4% year-on-year—even as the company keeps reminding everyone it’s a tech business first. That single number matters because it shows a reality most companies don’t like admitting: digital wins don’t automatically solve operational wins, especially in grocery.

Reuters’ report (via CNA) frames the latest move clearly: Amazon is testing a 225,000-square-foot “mega-store” outside Chicago designed to sell groceries and general merchandise while doubling as a same-day delivery distribution point. That’s not a vanity project. It’s an attempt to reduce the one cost that punishes every retailer—last-mile delivery—by getting customers closer to inventory.

For Singapore retailers and e-commerce teams following our “AI dalam Peruncitan dan E-Dagang” series, the Amazon–Walmart fight is a useful mirror. Not because you’re about to build a megastore in Jurong, but because the same operational question shows up here every day: How do you design your fulfilment, inventory, and customer experience to win repeat purchases—profitably? AI business tools can help, but only when they’re tied to the right operating model.

The real battle: last-mile cost and customer lifetime value

Grocery is the repeat-purchase engine. In the article, an Amazon seller consultant points out that grocery and fast-moving consumer goods (FMCG) buyers often deliver the highest customer lifetime value (CLV). I agree with that stance—and I’ll go further: grocery forces operational discipline. If your picking, inventory accuracy, substitutions, and delivery windows aren’t tight, customers don’t “forgive” you. They just switch.

Walmart’s advantage isn’t a nicer app. It’s a physical network: 4,600 stores and the claim that 90% of the US population lives within 10 miles of a Walmart. That footprint makes same-day fulfilment cheaper because inventory is already near demand.

Amazon is trying to manufacture a similar advantage. It said it delivered 4 billion groceries and everyday items same-day or next-day in 2025 in the US. Volume isn’t the problem. The problem is unit economics when distance and variability are high.

What this means for Singapore retail

Singapore’s geography is smaller, but the trade-offs are similar:

• Speed promises are expensive when demand is volatile and baskets are mixed (fresh + ambient + bulky).
• Inventory errors show up immediately in customer trust (out-of-stocks, wrong substitutions, late deliveries).
• CLV is earned operationally, not just through marketing.

AI tools are most valuable when they reduce variability in the system: better forecasts, smarter replenishment, faster picking, fewer returns, fewer missed delivery windows.

Amazon’s store reset is a strategy lesson, not a retreat

Amazon’s decision to close Amazon Fresh and Amazon Go locations and convert some into Whole Foods is easy to misread as “brick-and-mortar failed.” The more accurate read is harsher—and more useful: a store format that doesn’t scale operationally is worse than no store format.

Amazon itself said its prior approach didn’t create a distinctive enough shopping experience to expand at scale. That’s a polite way of saying the model didn’t hit the required combination of:

• repeatable customer value
• predictable store labour
• stable shrink control
• reliable replenishment
• manageable capex per store

Then comes the new bet: a mega-store that also functions like a fulfilment node.

The AI parallel: stop “piloting” tools that can’t scale

Singapore businesses do the same thing with AI:

• pilot a chatbot that isn’t connected to order status
• test demand forecasting on one product category without fixing master data
• add a recommendation engine but keep promotions manual and inconsistent

If the tool can’t be integrated into your operating rhythm (planning, buying, pricing, fulfilment, customer service), it won’t survive budget season.

A practical litmus test I use:

If an AI tool doesn’t change a weekly meeting, a daily workflow, or a KPI dashboard, it’s not operational innovation. It’s a demo.

Walmart+ shows why membership is an operations product

Walmart’s e-commerce used to trail Amazon badly. The article highlights how Walmart+, launched in 2020, helped change momentum. Morgan Stanley research cited puts Walmart+ at 26.5 million members as of 2025, and Walmart e-commerce sales grew 28% in its most recent quarter.

Here’s the part many teams miss: membership programs are often marketed as loyalty plays, but they’re really operations products.

A membership promise (free delivery, fast slots, easy returns) is only profitable when you can:

• batch and route orders efficiently
• keep substitution rates low
• maintain high in-stock availability
• control last-mile costs
• predict demand by micro-area

Walmart’s physical stores do that heavy lifting.

Singapore angle: memberships only work when fulfilment is disciplined

Whether you run a DTC brand, a marketplace shop, or a multi-outlet retailer, “membership” in Singapore (free delivery thresholds, priority slots, perks) should be designed around your fulfilment constraints:

• If you’re store-fulfilling, optimise pick paths and staffing by time-of-day.
• If you’re warehouse-fulfilling, manage cut-off times and slotting for fastest movers.
• If you’re using third-party logistics, negotiate service levels based on your real order profile.

AI makes this measurable by predicting demand spikes (paydays, festive periods, school terms) and by flagging where service levels are likely to break.

A practical AI blueprint for retail ops (built for Singapore)

Operational innovation becomes real when you wire AI into five decisions. Below is a field-tested blueprint that fits the “AI dalam Peruncitan dan E-Dagang” theme: personalised recommendations, demand forecasting, inventory management, and customer behaviour analytics.

1) Demand forecasting that accounts for promotions (not just history)

Most retail forecasts fail for one reason: they treat promotions as “exceptions.” In reality, promotions are the business.

What to implement:

• forecast by SKU x location x day (or at least week)
• include promo mechanics: discount depth, bundle, placement, ad spend
• measure accuracy with MAPE and with out-of-stock rate during promos

Outcome to aim for: fewer emergency replenishments and less dead stock after the campaign.

2) Replenishment that’s constraint-aware

A reorder suggestion is useless if it ignores constraints: shelf capacity, case-pack sizes, supplier lead times, cold-chain limits.

What to implement:

• min-max policies that update using forecast + lead time variability
• safety stock that changes by service level target (A items vs C items)
• supplier scorecards that feed replenishment decisions

Snippet-worthy truth: A “smart” forecast without smart replenishment is just a prettier spreadsheet.

3) Picking and fulfilment optimisation (where last-mile starts)

Amazon’s megastore concept is basically a bet that stores can double as distribution nodes. If you fulfil from stores in Singapore, picking efficiency is your margin.

What to implement:

• batch orders by aisle proximity and promised time slot
• dynamic labour scheduling using predicted order volume
• substitution intelligence: recommend alternatives that preserve basket value

KPIs that matter:

• pick rate (items/hour)
• substitution acceptance rate
• late order rate

4) Personalisation that respects stock reality

Personalised recommendations are great until they push unavailable items. That creates frustration and increases customer service load.

What to implement:

• recommendations filtered by real-time availability and delivery slot
• “next best item” logic for out-of-stocks
• personalised replenishment reminders for repeat-buy items

Outcome to aim for: higher conversion without increasing cancellations.

5) Customer behaviour analytics that connects to CLV

The article’s CLV point is the strategic heart of grocery. In practice, you should measure CLV drivers by cohort:

• order frequency
• average basket size
• margin per order after fulfilment costs
• churn after service failures (late, missing, wrong substitution)

What to implement:

• cohort dashboards by acquisition channel (ads, marketplaces, organic)
• churn prediction triggers tied to operational events
• win-back offers based on what went wrong (not generic discounts)

“Do I need physical stores to compete?” (People also ask)

No—most Singapore businesses don’t need more stores. They need a tighter network.

Amazon is adding physical capacity because it needs proximity at US scale. In Singapore, the equivalent could be:

• using 1–3 outlets as dedicated dark-store zones at peak hours
• creating micro-fulfilment capability inside existing stores
• partnering with fewer, more reliable delivery providers (with shared dashboards)

The winning pattern is consistent: put fast-moving inventory closer to demand and remove variability with data.

What to do next: turn “AI adoption” into operational advantage

Amazon vs Walmart isn’t a story about who has the better website. It’s a story about operating models: physical footprint, fulfilment economics, and how repeat purchases are protected by execution.

If you’re running retail or e-dagang operations in Singapore, take the lesson seriously: AI tools pay off when they reduce stockouts, shorten picking time, and improve delivery reliability—not when they sit in a slide deck.

I’d start with one concrete move this month: map your end-to-end flow (forecast → buy → stock → pick → deliver → service) and identify the single biggest source of avoidable cost. Then choose an AI business tool that directly attacks that bottleneck and ties to a KPI your team reviews weekly.

The forward-looking question worth asking: when your next festive spike hits—Ramadan, 6.6/7.7, year-end—will your system learn and improve, or will you fight the same fires again?

Source article for context: https://www.channelnewsasia.com/business/amazons-physical-grocery-push-deepens-its-fight-against-rival-walmart-5906961