Grocery AI in 2026: Instacart Connected Stores, Walmart Sparky, Albertsons Cooler, and Online Unit Economics

Grocery is the retail vertical where the AI-and-automation investment of the past five years is finally producing recognisable unit economics. The online grocery category lost vast amounts of money through 2020-2023 — fulfilment cost per order, last-mile cost per delivery, and the basket-margin math all worked against the operators — and the 2024-2026 turnaround has been driven by a combination of automated fulfilment, smarter store-level operations, and AI-driven labour and inventory optimisation. The vendors and operators producing the turnaround are a mix of platform players (Instacart), retailer-internal builds (Walmart, Tesco, Coupang), and a tail of specialised vendors selling into the major chains.

This is the practical sort — what’s actually shipping, where the economics work, and what the in-house AI assistants from Walmart and Tesco are actually doing.

The online grocery unit economics in 2026#

A typical 80-dollar online grocery order in 2026 costs the retailer somewhere in the range of 12-18 dollars to fulfill end-to-end — picking labour, packaging, last-mile delivery, customer service, and the cost of substitutions and refunds. On a typical 25 percent gross margin basket, that’s 20 dollars of gross margin and 12-18 dollars of fulfilment cost, leaving 2-8 dollars of contribution per order before allocated overhead. The order-level economics now work; the question is volume per fulfilment node.

Three operational levers moved the math. Automated fulfilment — the Ocado-style and Coupang-style robotic warehouses that lowered pick cost from the 6-8 dollar range to under 3 dollars in mature deployments. Density of routes — the AI-driven dispatch that pushed orders-per-driver-hour up materially over 2022-2025. Substitution AI — the model that picks a customer-acceptable substitute when an item is out, reducing the customer-call cost and the refund rate.

The retailers that captured the math typically combined all three. The ones that captured one or two are still losing money on the channel.

Instacart Caper and Connected Stores#

Instacart’s transformation from a delivery-only marketplace into a broader grocery technology platform was the largest 2023-2025 strategic shift in the category. Caper (the smart cart acquired in 2021) and Connected Stores (the bundling of cart, in-store screens, inventory APIs, and the Instacart Ads platform under one in-store proposition) became the retailer-side story alongside the consumer marketplace.

Caper carts — the AI-enabled trolleys that recognise products as customers drop them in, eliminating the checkout step — are now in Kroger, ShopRite, Wakefern, and a handful of European chains. The unit economics of the cart hardware itself remain hard (the per-cart cost is meaningful), but the through-cart Instacart Ads monetisation closes meaningful share of the cost. The 2024-2025 product additions included on-cart recipe suggestions, real-time loyalty integration, and the partnership with Uber for cross-platform delivery handoff.

Connected Stores as a category sells the retailer on a story that has only recently become defensible — Instacart is now infrastructure for the chain rather than just a competitive delivery layer. Whether that story holds depends on the chain; Albertsons, Wegmans, Publix have leaned in; Walmart and Amazon predictably have not.

Walmart Sparky#

Walmart’s Sparky — the consumer-facing generative AI assistant — launched in 2024 inside the Walmart app and steadily expanded scope through 2025. The current Sparky handles recipe-driven shopping (the assistant suggests a complete shopping list for a recipe), substitution decisions during order fulfilment (the assistant proposes acceptable swaps in real-time during pick), and an increasing share of customer-service queries that previously went to phone or chat agents.

The underlying stack is the internal Walmart AI platform layered on a mix of foundation models — Walmart has been deliberately multi-provider — and the data layer combines Walmart’s product catalogue, customer purchase history, and the in-store inventory feed. The internal claim is meaningful reduction in customer-service contact volume and measurable basket lift on Sparky-recommended orders; the unpublished side is the tuning effort over months and the specific edge cases where the assistant still hands off to humans.

Albertsons AI Cooler#

Albertsons’ AI Cooler — the smart cooler hardware deployed in stores starting 2023 — is the Albertsons answer to in-store fresh and prepared-food merchandising. The cooler combines product recognition (so the inventory updates in real-time as items leave the shelf), dynamic pricing (price displays update based on inventory levels and time to spoilage), and integration with the online grocery fulfilment for stores acting as fulfilment nodes.

The deployment is selective rather than chain-wide; the high-density prepared-foods stores and the meal-kit aisles are the priority. The 2025 disclosure described meaningful waste reduction and basket margin improvement in deployed stores, with the chain-wide rollout pacing reflecting the per-cooler capital cost.

Tesco’s Whoosh#

Tesco Whoosh — the under-an-hour Tesco delivery service launched 2021, scaled 2023-2025 — is the most successful European retailer-led quick-grocery service. Whoosh ships from the existing Tesco store network rather than dedicated dark stores, which fixed the dark-store unit economics problem that broke Getir and JOKR (see the quick commerce post). The AI behind Whoosh handles dispatch, route optimisation, and dynamic basket suggestions to maintain order density.

The honest competitive read on Whoosh against Gorillas (now Getir-acquired and effectively wound down in the UK) and Deliveroo Hop is that the existing-store fulfilment model produced economics the pure dark-store players could not match. By 2026 Whoosh is operationally profitable in mature catchments, which is a sentence the quick-commerce category mostly could not say at any point in 2021-2023.

Coupang’s automated fulfilment#

Coupang — the Korean e-commerce-and-grocery operator — operates one of the most advanced automated fulfilment networks in retail globally, with Coupang Fresh handling next-morning grocery at high reliability across most of South Korea. The fulfilment centres combine the Ocado-style automated grid with extensive on-site AI for pick optimisation, packaging selection, and route planning. The reliability — same-day or next-morning delivery on the vast majority of orders — set a customer expectation level in Korea that competitors have not matched.

The strategic significance of Coupang for 2026 is the Farfetch acquisition (closed early 2024) — the company is increasingly a multi-vertical retail operator rather than a pure grocery and general-merchandise play, and the AI investment is reusable across categories. We covered Coupang’s Farfetch acquisition in the luxury retail post.

Online grocery’s unit economics fix-via-AI#

The three AI investments that moved online grocery from money-losing to break-even-and-better:

Pick efficiency — store-pickers with handheld devices that route them through the store optimally, predict the most likely substitutions before the out-of-stock is encountered, and batch pick across multiple orders. The Instacart Shopper app and the in-house equivalents at Walmart, Tesco, and Albertsons all converge on this pattern.

Dispatch and routing — the driver-side AI that matches the right order to the right driver in the right time window, with predictive ETA for the customer. Doordash, Uber, Deliveroo, and the in-house retailer fleets all run variants of the same optimisation, with the differentiation coming from the density of orders and the integration with the picking flow.

Substitution AI — the model that picks an acceptable substitute when an item is out, trained on customer purchase history and explicit preference signals. A good substitution model reduces both refund rate and customer-call volume; a bad one creates negative reviews and lost-customer events.

The retailers that combined all three captured the unit-economics turnaround. The ones that captured one or two are still subsidising the channel from the rest of the business.

What we recommend in 2026#

For a grocery operator scoping an AI program:

• Customer-facing assistant — pilot the Sparky-style use cases (recipe-to-list, substitution acceptance, customer-service Q&A) on the existing app surface rather than as a separate product. The integration cost is the binding constraint, not the model capability.
• Smart cart or smart cooler hardware — evaluate against existing in-store fixture economics and the realistic ad-monetisation upside. The technology works; the per-fixture cost is the question.
• Pick and dispatch optimisation — invest before any front-end AI ambition. The fulfilment-side AI is where the dollars live.
• Substitution AI — the highest-ROI single model in online grocery; the refund-rate reduction and customer-call reduction are measurable within weeks.
• Automated fulfilment — the Ocado-style or Coupang-style commitment is a multi-hundred-million-dollar question; appropriate for chains above a clear volume threshold, not below.

Related reading#

• Quick commerce economics and the Getir collapse
• Demand forecasting platforms
• Store operations cameras and shelf scanning

Where pdpspectra fits#

We help grocery operators stand up the data layer behind the pick, dispatch, and customer-facing AI — POS, OMS, WMS, driver-side, and customer-app pipelines plus the feature store and serving infrastructure for the substitution and routing models. Our ML and MLOps practice handles the model lifecycle.

---

Online grocery’s unit economics now work — for the operators that captured the AI investments. If you are scoping a pick-and-dispatch program, a smart-cart pilot, or a Sparky-style assistant, tell us about the chain.