Physical retail is expected to operate with the data precision of an e-commerce store. Most physical stores do not have the infrastructure to do so.
Connecting the physical realities of a shop floor—inventory misplaced on the wrong shelf, queues building at the register, unpredictable foot traffic—to a digital system requires more than modern point-of-sale software. It requires translating physical movement into structured data without disrupting the people working in the store.
Focus Area
Smart Stores
Focus Area
IoT Sensors
Focus Area
Computer Vision
Focus Area
Spatial Analytics
Understanding the Reality of Retail
Running a physical retail environment involves managing constant entropy. Inventory arrives, is moved, misplaced, purchased, or stolen. Staff need to be allocated where traffic is highest, but traffic patterns change by the hour. All of this happens in real time, yet the systems used to manage it often rely on batch processing or manual counting—meaning the store is always being managed based on what happened yesterday.
The gap is usually between the transactional data (what was sold) and the spatial data (what is actually on the shelf or happening in the aisle right now). Modern POS systems handle the transaction perfectly, but the inventory record assumes perfection between the loading bay and the register. When an item is picked up by a customer and put back in the wrong aisle, the system still registers it as available. Bridging that gap requires computer vision, sensor networks, and edge computing that traditional retail stacks were not designed to support.
The burden falls entirely on floor staff and store managers. A manager spending hours conducting manual cycle counts instead of coaching their team. A floor associate trying to help a customer, relying on a stock system that says 'two available' when the physical shelf is empty. These are not just administrative inefficiencies; they erode staff morale, create unnecessary friction, and directly damage the customer's trust in the brand.
Solving the Right Problems
We target the specific workflows where manual effort meets its ceiling.
Real-time inventory visibility
Manual cycle counts are time-consuming and prone to human error. By the time a discrepancy is discovered, the resulting stockout has already cost the store sales and frustrated customers.
Spatial analytics and foot traffic
Retailers possess granular data on e-commerce browsing, but physical stores remain a black box. Knowing what was bought does not explain what was looked at, picked up, and ultimately abandoned.
Pricing and shelf operations
Updating paper shelf-edge labels across a large format store requires hours of manual labour. Discrepancies between the shelf price and the POS create friction at checkout and regulatory compliance risks.
Active loss prevention
Traditional CCTV requires constant human monitoring and usually only serves as retroactive evidence. It does not actively identify or prevent shrink as it occurs on the shop floor.
Equipment and environment monitoring
Refrigeration failures, HVAC faults, or energy spikes are often discovered too late, resulting in spoiled perishable inventory, uncomfortable shopping environments, or compliance breaches.
What We Build
Outcomes defined in the language of the people who rely on them.
Computer vision inventory systems
Camera and sensor infrastructure that monitors shelf states, detects out-of-stocks, and flags planogram non-compliance, triggering replenishment alerts automatically.
Spatial analytics platforms
Edge-processed foot traffic analysis that maps store heatmaps, zone conversions, and queue lengths without capturing or storing any personally identifiable information.
Electronic shelf label (ESL) infrastructure
Wireless e-ink display networks connected directly to the retailer's ERP and pricing engine, ensuring pricing consistency between the shelf and the register.
IoT equipment monitoring
Sensor networks covering temperature, humidity, and energy consumption across refrigeration and HVAC systems, with predictive maintenance alerting.
Behavioural loss prevention layers
Vision models trained to recognise specific movement signatures associated with theft, integrated with staff communication tools for discreet intervention.
Unified store operations dashboard
A single, role-based interface combining inventory alerts, traffic data, and equipment status—giving store managers one place to understand current operational realities.
Honest AI for Retail
Specific, grounded applications—no hype. We use machine learning for tasks that are repetitive, high-volume, and data-dependent.
Computer vision in a busy retail environment has to account for occlusions, varying lighting, and dense crowds. It is not perfect. We design systems to flag exceptions for human review rather than making autonomous decisions that could disrupt the shopping experience or wrongly accuse a customer. AI in physical retail supplements the store team; it does not replace the need for human judgment on the shop floor.
The most frequent and valid concern regarding retail tech is shopper privacy. We build spatial analytics and vision systems using edge processing—meaning the video feed is converted to anonymous coordinate data locally, and the video itself is immediately discarded. We do not use facial recognition to identify individuals, nor do we track personally identifiable information without explicit, opt-in consent (such as through a loyalty app). We ensure the architecture complies strictly with GDPR and CCPA requirements.
Shelf-state recognition
A vision model analyses images from shelf-facing cameras to recognise which products are present, their orientation, and how many remain facing outward. It compares this against the expected planogram and updates the inventory record or alerts staff to restock a specific aisle.
Predictive demand and labour forecasting
A forecasting model trained on historical sales, local event calendars, weather data, and foot traffic patterns provides a probability-weighted view of expected volume. This allows store managers to align shift schedules with actual predicted demand rather than relying purely on last year's figures.
Behavioural anomaly detection
Rather than requiring a security guard to watch dozens of camera feeds simultaneously, a model monitors for specific physical movement patterns—such as sweeping multiple items into a bag at once—and surfaces only the flagged video snippets for human review.
How We Work
We map the physical workflow of the store before we propose any technology. A system that requires floor staff to change their natural behaviour will ultimately be bypassed.
We observe the shop floor before we assess the tech stack
We spend time understanding your current operations—how inventory is unloaded, how staff manage restocks during peak hours, and where the daily friction actually lives. Technology built in a lab often fails in a stockroom. We design around the reality of your physical space.
We integrate with existing systems first
A new sensor network is useless if it requires staff to check a completely separate application. We scope the integration with your existing POS, ERP, and inventory management systems as the foundation of the build, ensuring new data flows into the tools your team already uses.
We pilot in representative environments
We do not pilot exclusively in pristine flagship stores. We deploy initial technology in locations with representative challenges—older layouts, typical staffing levels, and standard network constraints. If the technology cannot survive there, it cannot scale.
We measure impact through operational metrics
We measure success not by technical uptime, but by operational outcomes: hours saved on cycle counts, reduction in out-of-stock incidents, and decreases in shrinkage. The technology is only marked as successful when the store manager confirms it has made the store easier to run.
We do not sell proprietary hardware that locks you into a single vendor ecosystem. We build the integration and analytics layer that connects industry-standard sensors, cameras, and ESLs to your existing retail management systems, maintaining your architectural flexibility.
98%
Inventory accuracy
achieved using automated computer vision tracking
< 5 min
Store-wide price updates
following implementation of ESL infrastructure
~75%
Reduction in stockouts
grocery chain using automated replenishment triggers
Stories of Change
Real scenarios where manual bottlenecks were replaced by continuous visibility.
A regional grocery chain was running manual inventory counts that took roughly 40 hours per week per store, with an accuracy rate of around 70%. Stockout complaints were increasing, and perishable items were frequently expiring on shelves because staff were unable to rotate stock efficiently.
A computer vision inventory system using shelf-mounted cameras and a lightweight edge-processing server. The system was integrated with their existing ordering software to trigger replenishment alerts automatically when stock dropped below defined thresholds.
Inventory accuracy improved to approximately 98%. Stockouts were reduced by around 75%. The 40 weekly hours previously spent on manual counting were reallocated to customer-facing tasks, and perishable waste decreased noticeably over the following two quarters.
A large-format electronics retailer wanted to reduce the time staff spent on manual price changes and address discrepancies between shelf prices and the POS, which were causing friction at the register and drawing regulatory scrutiny.
A store-wide electronic shelf label (ESL) network integrated directly with their central pricing engine. Price updates were pushed via a wireless gateway, ensuring the physical shelf always matched the central database.
Store-wide price updates that previously took several staff members a full morning now complete in under five minutes. Pricing discrepancies at the register dropped to near zero in the first month, entirely eliminating the associated customer friction.
A fashion retailer had robust e-commerce data but no visibility into physical store performance beyond final sales. They could not determine whether poor sales of a new line were due to low foot traffic, poor placement, or lack of customer interest after viewing.
A spatial analytics platform using anonymous overhead sensors to map foot traffic, dwell time by zone, and path-to-purchase. The data was fed into a dashboard comparing zone traffic against POS data to calculate 'conversion by area.'
The retailer identified that a heavily promoted line was placed in a low-traffic blind spot. After adjusting the store layout based on the heatmaps, engagement with the new line increased by roughly 40%, directly correlating with a measurable lift in sales for that specific category.
Nuance by Retail Segment
The core problems are similar, but the operational environment dictates the solution.
Grocery and supermarkets
Fresh produce monitoring, expiration tracking, temperature compliance for refrigeration units, and automated reordering to prevent food waste without requiring staff to manually check every aisle.
Fashion and apparel
Inventory tracking across complex size, colour, and SKU variants. Spatial analytics to measure which displays capture dwell time, and RFID integration for rapid stockroom location.
Consumer electronics
High-value item tracking, behaviour-based loss prevention, and customer dwell analysis by product category to help security and merchandising teams work from the same data.
Home improvement and big box
Inventory management for large-format and variable-weight items, layout optimisation for vast footprints, and automated stockroom replenishment for high-turnover materials.
Pharmacy and health
Strict temperature monitoring for medications, controlled substance tracking, and spatial analytics designed to protect customer privacy in high-sensitivity consultation areas.
Convenience retail
24-hour automated monitoring, rapid restocking alerts, and foot traffic analysis for high-volume environments where speed and availability determine the entire customer experience.
How to Start
A predictable path from initial assessment to scaled deployment.
Store environment audit
A two-week review of your current operations, existing hardware, network infrastructure, and where the measurable operational gaps are. We produce a clear picture of what is technically feasible and what is worth addressing.
Single-capability pilot
A focused 6–8 week build covering one specific capability—such as spatial analytics or environmental monitoring—in one or two representative stores. We define success metrics upfront and measure against them rigorously.
Multi-store platform rollout
A 12–16 week deployment across multiple locations. We manage the integration with central systems, coordinate hardware installation, and provide the training required for store teams to adopt the new workflows.
Ongoing operational partnership
Continued involvement after launch—retraining vision models as store layouts change, adjusting alert thresholds, and expanding the sensor network as the results justify further investment.
Security & Compliance
Built for rigorous retail environments where privacy and availability are non-negotiable.
Every solution assumes a high-stakes environment. Data is anonymized at the edge, encrypted in transit, and secured by default.
Privacy via edge processing
Spatial analytics and computer vision systems process video feeds locally on edge devices. Video is converted instantly into anonymous coordinate data, and the raw footage is discarded. No personally identifiable information (PII) is captured or stored.
IoT network security
Sensor networks operate on segregated VLANs, distinct from the store's primary POS or corporate networks. All telemetry data in transit is encrypted using TLS, and devices are authenticated using unique cryptographic certificates.
Regulatory data compliance
Data collection architectures are mapped strictly to GDPR and CCPA requirements. Automated monitoring for temperature-sensitive products maintains immutable logs to satisfy food safety and pharmaceutical compliance inspections.
Underlying Technology
Enterprise-grade architecture capable of processing physical store events in real-time.
Computer vision and edge computing
Visual analysis infrastructure for inventory tracking, spatial analytics, and anomaly detection
- OpenCV and custom PyTorch models for object detection and shelf monitoring
- NVIDIA Jetson or equivalent edge devices for low-latency, on-premise inference
- DeepStream SDK for processing multiple high-definition video streams concurrently
- Anonymous coordinate extraction pipelines for privacy-compliant tracking
IoT and sensor networks
Device management and telemetry layers for store environments and equipment
- AWS IoT Core or Azure IoT Hub for secure device registration and management
- MQTT protocol for lightweight, low-latency sensor communication
- Integration with major ESL providers (e.g., SES-imagotag, Pricer) via proprietary gateways
- Time-series databases (InfluxDB or TimescaleDB) for historical sensor data
Analytics and integration layer
Data processing pipelines connecting store telemetry to operational decisions
- Apache Kafka for managing real-time event streams from the store edge to the cloud
- REST and GraphQL APIs for bidirectional integration with existing ERP and POS systems
- React-based operational dashboards with role-based access control for store managers
- Python (scikit-learn) for demand forecasting and predictive maintenance alerting
Common Questions
Ready to close the gap?
Every retailer is dealing with a unique mix of legacy systems, physical constraints, and operational bottlenecks. Whether you are looking to automate inventory, understand store traffic, or simply modernise your pricing infrastructure, we are glad to hear where you are. No presentation. Just a conversation about what you are working through.