Store Operations AI: Verkada, Sensormatic, Simbe, Pensa, and the Bossa Nova Lessons
Cameras and robots inside the store — what Simbe's Tally actually does, why Bossa Nova failed, where Verkada and Sensormatic are quietly dominant, and where shelf-out-of-stock detection earns its keep.
Store operations AI is a category nobody owns. Cameras and robots inside the store solve overlapping problems — shelf availability, planogram compliance, price tag accuracy, customer flow, labour deployment, security — and every vendor sells a slightly different combination. The category has produced a few credible scale deployments, several high-profile failures, and one large quiet exit (Amazon’s Just Walk Out, which we covered in the loss prevention post).
This is the practical sort — what the vendors do, what retailers actually buy, and where the deployments are paying back.
The problem the store is asking AI to solve#
Walk through any large-format retailer at 4 PM on a Saturday and you will find: gaps on the shelf where popular items should be, items in the wrong location, price tags that don’t match the POS, and signs of customer frustration where staff aren’t deployed. None of these problems are new. What changed in the 2020-2025 window is that cameras and compute got cheap enough to monitor them continuously rather than via a weekly walk by a district manager.
The economic case is built on three numbers — out-of-stock revenue loss (most retailers carry meaningful unrecorded loss from shelves being empty when product is in the back room), labour deployment efficiency (where you put staff at what time relative to where customers actually are), and planogram compliance (the gap between the merchandising team’s plan and what the store actually looks like, which especially matters for vendor-funded space).
Verkada — the cloud-camera incumbent#
Verkada (founded 2016, valued at 4.5 billion USD in its 2023 round, the 2021 camera-feed breach now well behind it) became the default cloud-camera vendor for mid-market retail over 2023-2025. The product is simple — IP cameras with on-device AI for person, vehicle, and event detection, all managed in a single cloud console, sold on a per-camera subscription. The retail features added in 2024-2025 include customer counting, dwell-time heatmaps, queue-length detection at checkout, and integration with access control.
The Verkada strength is operational simplicity — install the camera, plug it into PoE, the cloud handles the rest. The weakness is that it isn’t deeply specialised for retail. For shelf availability, planogram compliance, or detailed customer-journey analysis, Verkada is a baseline, not a solution.
Sensormatic Solutions — the unglamorous incumbent#
Sensormatic (the Johnson Controls business that includes the legacy Tyco retail portfolio) is the largest retail-specific vendor in the category by a wide margin, and the one most often overlooked in AI vendor reviews because it is not a startup. The Sensormatic IQ platform combines EAS (electronic article surveillance — the tag towers at store exits), traffic counting, inventory intelligence (RFID, where deployed), and an analytics layer feeding store managers and the central merchandising team.
What Sensormatic does well: it is already in most large-format retailers, the data integrations are working, the analytics are credible, and the relationship with the loss-prevention organisation is established. What it does less well: the AI story lags the pure-play startups, and the platform feels more like an evolution of mid-2010s retail tech than a 2026 rebuild.
For retailers asking where to start an AI-in-store program, Sensormatic IQ is often the highest-ROI first step because the data is already there.
Simbe Robotics — Tally the shelf-scanning robot#
Simbe (founded 2014, San Francisco-based) makes Tally, the shelf-scanning robot that roams the aisles capturing the entire shelf state. The robot autonomously navigates a store, photographs every shelf section several times a day, and produces an AI-derived inventory of what is actually on the shelf — out-of-stocks, misplacements, price-tag errors, planogram non-compliance.
The Tally deployments include Schnucks (the regional US grocer that was the first scale customer), Wakefern’s ShopRite stores, Carrefour locations in Europe, and an expanding footprint at BJ’s Wholesale Club. The Schnucks rollout has been publicly discussed for years and is one of the clearest examples of a robot-led store ops deployment generating real measurable value — out-of-stock reduction, faster restocking, and meaningful merchandising compliance improvement.
The honest read on Simbe: the technology is real, the customer satisfaction is high among the retailers that committed to the rollout, but the per-store cost and the operational disruption (robots need charging stations, navigation calibration, and store-staff cooperation) means the technology is a fit for specific store formats, not universal retail.
Pensa Systems — the camera-on-fixture approach#
Pensa (Austin-based, acquired by NielsenIQ in early 2024) attacked the same shelf-availability problem with a different topology — fixed cameras mounted on shelves and ceilings, computer vision instead of robotics. The product runs continuously rather than during scheduled robot rounds, and integrates with the merchandising and supply-chain stack to drive automated reorder and restocking.
The NielsenIQ acquisition was a quiet signal that Pensa’s category — shelf intelligence — was being consolidated into the broader retail-measurement stack rather than competing as a standalone subscription. The 2025 product roadmap merges Pensa’s in-store data with NielsenIQ’s syndicated point-of-sale data, which is genuinely useful for category managers but a different sale than the operational-retail one Pensa was making pre-acquisition.
Bossa Nova Robotics — the cautionary tale#
Bossa Nova was the Walmart-deployed shelf-scanning robot vendor that became the highest-profile cautionary tale in store-operations robotics. The Walmart pilot ran from 2017 through 2020, expanded to roughly 1,000 stores at its peak, and was wound down in late 2020. Walmart’s public explanation was that the robots were not delivering enough incremental value over store associates with handheld scanners; the under-publicised version involves the cost of robot maintenance, customer experience friction (a robot in the aisle is not always neutral to shoppers), and the operational overhead of fleet management.
The lessons the rest of the category drew: shelf scanning has to clear a high economic bar against the baseline of associates with phones, the robot fleet management is a real operational burden that often gets under-modelled in the business case, and the retailer needs an existing supply-chain integration to actually act on the shelf-state data. Simbe survived by clearing all three of these bars more carefully; several other robotics startups did not.
The shelf-out-of-stock business case#
The most credible single ROI story in store-operations AI is shelf out-of-stock reduction. The mechanism is simple — when a popular item is missing from the shelf, the customer either substitutes (lower margin), leaves without buying, or buys online from a competitor. Studies from Nielsen, IRI, and academic supply-chain research put unrecorded out-of-stock loss at a meaningful percentage of revenue at most large-format retailers, with grocery typically higher than general merchandise.
The AI helps in two ways. Detection — fixed cameras (Pensa), roving robots (Simbe), or shelf-edge sensors flag the gap in near-real-time. Action — the alert routes to the closest associate via the labour management system, or to the back room for restock. The retailers extracting the largest value from shelf-OOS AI are the ones that connected detection to action, not the ones that bought the technology in isolation.
The Just Walk Out exit and what it means#
Amazon’s decision to remove Just Walk Out from most of its first-party Amazon Fresh and Whole Foods stores in 2024 was the largest single market signal in store-operations AI in the last decade. The system worked technically — customers could walk in, pick items, and walk out — but the unit economics in full grocery formats were not sustainable, the customer dispute rate on AI-generated receipts was higher than disclosed, and the offshore review staff required to handle the long tail of ambiguous events made the human-in-the-loop cost meaningful.
Just Walk Out continues as a B2B licensing business — stadiums, airports, university campuses, and a handful of third-party retailers — where the format characteristics make the economics work. For mainstream grocery and general merchandise, autonomous checkout remains a 2027-and-later question.
What we recommend retailers deploy in 2026#
For a retailer scoping a store-operations AI program:
- Verkada or Sensormatic IQ as the baseline if you do not already have a current-generation camera and analytics platform. This is the data layer everything else depends on.
- Shelf-availability detection before any robotics ambition — start with the cameras you have, the planogram data you have, and a credible alert-routing process to associates.
- Simbe or equivalent shelf-scanning robotics only after the alert-to-action loop is working; otherwise the robot data has nowhere to go.
- Labour deployment AI — the queue-length, customer-density, and dwell-time analytics that Verkada and Sensormatic both support. This is often the fastest payback because labour is the largest controllable store cost.
- Autonomous checkout as a 2027 question, not a 2026 one, unless your store format matches the Trigo or AiFi sweet spot.
Related reading#
- Loss prevention AI and Everseen at self-checkout
- Demand forecasting platforms
- Grocery AI and Instacart Connected Stores
Where pdpspectra fits#
We help retailers stand up the data plumbing behind store-operations AI — camera-feed ingestion, POS and labour-system integration, alert routing, and the dashboards district managers actually use. Our AI and LLM integration practice handles the model-side; the data engineering team handles the rest.
The store-operations data layer is the actual product. If you are evaluating shelf-scanning, queue analytics, or planogram-compliance AI, tell us about the stores.