A cheap field visit becomes structured data
A guided phone walk reconstructs lane pitch, row pitch, shelf depth, and occupancy—without mounting hardware on the fixture.
Phone-based physical shelf measurement
Walk the aisle. Measure the shelf.
ShelfSense turns a guided 30-second phone pass into an auditable record of a structured shelf: what is stocked, where the gaps are, what the scan could not read, and how many units may remain behind each facing.
No fixed camerasNo shelf sensorsExplicit unknownsEvidence attached
Physical observation
Guided phone captureBeverage bay · 10 lanes
FrontDepth row 6
StockedGapUnknown
Capture qualityAccepted
Shelf geometryRecovered
OutputEvidence-backed
The measurement gap
Sales data says what moved. ShelfSense measures what is physically there.
The product is not another generic shelf-audit dashboard. It is a lower-cost way to create a defensible physical observation for fixtures that have no installed sensing.
Every answer keeps its evidence
Observations, model inference, confidence, and validation stay separate, making the result inspectable instead of a black-box score.
Uncertainty is part of the product
Poorly seen positions remain unknown. Hidden inventory is reported as a probability-backed range, not an exact count the scan did not earn.
From aisle walk to evidence
The phone guides the capture. The system earns the answer.
Capture
The app coaches height, tilt, pace, and coverage, then rejects a pass that cannot support a trustworthy measurement.
Reconstruct
Camera evidence, depth, and the fixture's own geometry become a lane-by-depth digital representation of the bay.
Measure
Each slot is returned as stocked, gap, or unknown. An optional confirmed lane map turns the same reconstruction into per-product count ranges.
Verify
A content-addressed evidence record preserves the inputs, model version, decisions, and uncertainty behind the result.
Phone positionHeight + tilt guided
Live quality checksWithin capture envelope
One evidence backbone
Useful outputs for operators, analysts, and data platforms.
ShelfSense is designed to produce a portable physical-measurement record—not trap the result inside one dashboard.
Slot-level status
A lane-by-depth report of stocked, gap, and unknown positions, with confidence and source evidence kept alongside every call.
Count ranges
Probability-backed unit ranges by lane. With a confirmed lane map, those ranges can roll up to known products without pretending open-world SKU recognition is solved.
Fixture geometry
Measured lane spacing, row depth, cap plane, and shelf extent create a lightweight spatial twin for audits, revisits, and downstream systems.
Evidence before promises
A real blind result, reported with its limits.
On a recorded bay whose prediction was frozen before the hand count existed, ShelfSense correctly classified 93.6% of the slots it answered. It deliberately left deeper rows unknown when the view did not support them.
Capture 2A · frozen predictionBlind result
Slots answered140 / 180Rows 1–7 answered; deeper positions remained unknown.
Accuracy on answered93.6%Correct among the 140 slots the system chose to answer.
Whole-shelf accuracy72.8%Unknowns remain in the denominator.
All-occupied baseline52.2%The comparison for calling every slot stocked.
Count-range research: across two recorded bays, an 80% lane range contained the hand count in 32 of 40 lanes, with 0.58-bottle mean absolute error. Encouraging evidence—not yet a customer-calibrated claim.
Trust is a product feature
ShelfSense is built to say “we don't know.”
Physical measurement has failure modes: a pass can be too low, depth rows can visually collapse, and a product can hide behind another. ShelfSense makes those limits visible and blocks unsupported certainty.
MeasuredFixture geometry and visible evidence
InferredOccupancy probabilities and count ranges
Needs confirmationProduct identity from a lane map or operator
RejectedCapture outside the evidence envelope
UnknownPositions the scan cannot support
Seeking a design partner
Put ShelfSense against a real measurement workflow.
We are looking for a partner with structured gravity-feed fixtures and an existing cost of observation. A paid proof compares ShelfSense with that process on field cost, elapsed time, usable-observation yield, and technical quality.