Instacart Just Made The Bet Walmart Placed In 2020. The Best Shelf Scanner Is A Person Already In The Aisle.

In November 2020, Walmart ended its contract with Bossa Nova Robotics and pulled shelf-scanning robots out of roughly 500 U.S. stores. The Wall Street Journal, which first reported the decision, cited people familiar with the situation who said the retailer had found that human workers could get similar results, in part because the pandemic-era surge in pickup and delivery orders meant more employees were walking the aisles anyway, picking up inventory information along the way. Walmart said it would keep testing other approaches, and its club division followed through. Sam’s Club completed a chainwide rollout of Inventory Scan camera towers mounted on its robotic floor scrubbers, built with Brain Corp, across its roughly 600 clubs by October 2022.

On July 16, Instacart bought the industrialized version of the argument Walmart’s own store data made nearly six years ago. The company acquired Arpalus, a computer vision firm whose technology turns a short smartphone video of a store shelf into an item-level inventory read that Instacart says exceeds 95 percent accuracy on average, with models built for the conditions that defeat generic image recognition in grocery, including uneven lighting, weak in-store connectivity, and thousands of visually similar products packed close together. Terms were not disclosed, though Calcalist estimated the price in the tens of millions of dollars and noted it is Instacart’s first acquisition of an Israeli company. Arpalus was founded in 2019 and maintains offices in Netanya, Israel, and Tenafly, New Jersey.

Instacart’s Wager Is That The Sensor Network Already Exists

The acquisition announcement doubles as a description of the sensor fleet Instacart believes it already owns. The company says its 600,000 shoppers visit large-format stores more than 15 times a day on average, generating over 10 million unique daily data points, on top of an intelligence base built from more than 1.6 billion lifetime orders and inventory views from nearly 100,000 stores across North America. Until now, most of that shelf knowledge was inferential, reconstructed from what shoppers found, failed to find, and substituted. Arpalus makes the observation direct. Its models run on any smartphone and can coach a shopper in the moment, prompting adjustments to camera angle or distance so a full shelf gets captured accurately.

The captured data feeds more than order fulfillment. Instacart said the technology directly enhances Store View, its real-time computer vision product being piloted by retailers including Sprouts and McKeevers, and extends to Caper Carts, its camera-equipped smart carts, which have scaled to more than 100 cities. Arpalus founder and CEO Ofir Zilberberg said the company built its platform to give retailers “real-time visibility into inventory and store execution,” and the customer list matters here. Instacart’s release describes Arpalus as serving both retailers and CPG brands.

Robots And Ceiling Cameras Got To Chainwide Scale First

Instacart is the newest entrant in this race, not the leader. Simbe Robotics has spent a decade putting its Tally robot into grocery aisles, and the deployments are no longer pilots. BJ’s Wholesale Club announced in March 2023 that it would roll Tally out across all of its then 237 clubs, following Schnucks, which committed the robot to all 111 of its stores in 2021. Wakefern banners run the robots through aisles up to three times per day scanning for stock position, price accuracy, and promotional execution, and Hy-Vee has deployed them as well. Simbe launched its fourth-generation robot this January, with runtime extended to 12 hours and an NVIDIA AI infrastructure stack underneath.

Fixed cameras represent the third model, and the largest documented result belongs to Morrisons. The UK grocer began deploying Focal Systems cameras in April 2024, installing more than 200,000 across 498 supermarkets in eight months, and The Grocer reported the move improved customer availability by more than two percentage points. Each Morrisons store runs 400 to 600 cameras capturing every shelf throughout the day. Jess Dobson, who led the project at Morrisons, told The Grocer that “Focal’s accuracy was higher than we had seen with previous technologies.” The model is spreading within Walmart’s own ecosystem as well, since Walmart Canada rolled Focal’s cameras out chainwide following a 70-store pilot, and Asda began testing the cameras in five stores last year.

The three models buy the same information at different price points and refresh rates. A robot fleet is a per-store capital purchase that sees everything a few times a day. Ceiling cameras cost more to install densely but refresh hourly without moving. Crowdsourced capture adds near-zero marginal hardware cost, but its coverage follows order volume, which means the stores and aisles with the most e-commerce demand get seen most often. The implication for operators weighing these systems is that the choice is less about recognition accuracy than about which coverage pattern matches how a given banner actually loses sales to gaps.

The Capture Method Shapes The Shelf Data Brands Will Be Offered

Whichever device does the looking, the output is the same commodity, on-shelf availability data, which brands have historically purchased late and secondhand through audits and syndicated reporting. The underlying problem is large. When Instacart launched Store View in 2025, it pointed to an ECR Retail Loss Group study finding that as much as 60 percent of retail inventory records are inaccurate, and the company has identified shelf gaps that go unnoticed as a leading reason online grocery customers end up unhappy with their orders.

The implications split by seat. For retail operations leaders, the question is whether shelf visibility becomes owned infrastructure, as at BJ’s and Morrisons, or a service consumed through a platform partner, as in the Store View pilots. Each path trades control for speed differently, and the contracts governing where captured data flows deserve as much attention as the accuracy claims. For CPG commercial teams, the nearer-term consequence is a multiplication of sources, because availability data on the same product will increasingly arrive from robot fleets at one account, ceiling cameras at another, and crowdsourced capture at a third, with no common standard for freshness or coverage. Teams that treat all three as interchangeable inputs to the same OSA metric will be comparing measurements taken at different frequencies with different blind spots.

The infrastructure buildout is already compounding. Morrisons announced in October 2025 that it will deploy more than 10.8 million electronic shelf labels across its 497 supermarkets beginning in early 2026, and said the label system will work in concert with its shelf-edge cameras so that store staff can be pointed toward empty facings and online orders can be picked with fewer errors. The capture layer, in other words, is starting to connect to the action layer at the shelf edge. Asda’s five-store camera trial and Instacart’s Sprouts and McKeevers pilots are the live experiments to watch next, because each one is a retailer deciding, in public, which way of seeing the shelf it trusts.

Conversations On Retail

Conversations On Retail is a gathering place and resource center for retail and CPG executives, built to make it easier to stay current, discover the technologies and solutions shaping the industry, and connect with the people driving it forward.

We publish news, views, and reviews from staff editors, contributing experts, and trusted partners. Some articles are developed internally, while others are submitted by industry contributors or adapted from interviews and recorded conversations with industry leaders.