How it works
Record, recognise, reconcile.
A forklift you already own carries a camera unit down the aisle at normal travel speed. AI reads every pallet label and registers every location, including the empty ones, at up to 3,000 units per hour against 50 to 100 by hand scanner. The result is reconciled against your WMS as a discrepancy list, with the recording behind every line.
The position
There is no drone and no autonomous robot.
Most automated counting is sold as a vehicle. You buy or subscribe to a machine, you find somewhere to house and charge it, and you work around it. Sentispec Inventory is not a vehicle. It is a self contained camera unit that a forklift already working your site picks up on its forks and puts down again.
- Nothing is fitted to your vehicles. No brackets, no wiring, no vehicle downtime, no per forklift cost. The unit moves between forklifts and between sites because it is not attached to any of them.
- Nothing changes in your racking. No new labels, no reflectors, no markers, no fixed infrastructure. The unit is taught to read the labels you already print.
- Nothing is installed in the building. There are no anchors, no charging bays and no network hardware to commission, which is why a site is live in five days rather than a quarter.
- There is no capital equipment required. Because there is no machine to own, the price is a setup fee and an annual fee. Full warranty and ongoing support are included. See pricing.
The mechanism
Three steps, one drive down the aisle.
A complete stocktake goes from three to ten people over two to three days to one person in eight to twelve hours. Those hours are one drive down each aisle, and the three stages below all come out of that single pass.
Record
A forklift lifts the recording unit and drives the aisle at normal travel speed, one shelf level at a time, until every position is covered. The driver is doing one thing only: driving.
Recognise
AI reads every pallet label and identifies the position it sits in, down to the individual location in the racking. Empty positions are registered as findings, not as gaps in the data.
Reconcile
The count is matched line by line against your stock file. You get the exceptions, not the inventory you already trust, and each exception opens the recording that produced it.
Why the recording matters more than the count
A manual count produces a number. When the number is wrong, nobody can say why, so the only way to settle it is to recount. That is the real cost of manual stocktaking, and it is not in the labour line.
A camera count produces a number and the image behind it. When a discrepancy is challenged, the operator opens the image, looks at the label and closes the question in seconds. Auditors get evidence rather than an assertion. How the accuracy figure is measured, and how you would check it, is on accuracy.
What it reports
Four discrepancy types, and what each one costs you.
Reconciliation returns four kinds of finding. Each has a different owner and a different fix, so each is reported separately rather than rolled into one error count.
Found pallet in wrong location
The pallet exists and the stock is real, but a picker sent to the recorded position will find nothing. A put away error, corrected in one move.
Found pallet not present in WMS
Physically in the racking, invisible to your system. Stock you are holding, insuring and not selling.
Empty location where pallet expected
Your WMS is promising stock that is not there. This is the finding that becomes a short shipment if nobody catches it first.
Expected pallet missing in recording
Not seen where it should be. It becomes a targeted search of a handful of candidate positions rather than a full recount.
What the split tells you. A single accuracy percentage tells you how bad the week was. A split by discrepancy type tells you whether the problem is put away discipline, receiving, or a process that is writing stock off the system while it is still on the floor. Sites use the split to fix the cause, which is how accuracy stays at 99%+ instead of being restored once a year. See continuous cycle counting.
What it takes to run
A forklift, a driver and a stock file.
A 20 minute run covers 500 to 1,000 pallets and is normally driven during a break or between tasks. Counting happens during normal shifts, so a full audit does not need the operation stopped.
Integration effort is zero to begin with. Send a stock file and you get a full discrepancy report back, which is enough to run counts indefinitely. Connect the standard API when you want corrections to flow into your system automatically, and that is a decision you can take months after go live. The detail per system is on WMS integrations.
The limits
What Sentispec Inventory does not do.
Four jobs the recording unit will not do. If one of them is the job you need doing, this is the wrong product.
- It does not count cases or cartons inside a partial pallet. Sentispec Inventory counts at pallet and location level. It will tell you a position holds a labelled pallet and which pallet it is. It will not tell you that the pallet holds 34 cases rather than 48. If case level accuracy inside a broken pallet is the problem you are solving, this product does not solve it.
- It does not verify what is inside a pallet. The unit reads the label and the position. A pallet correctly labelled but holding the wrong goods reports as a match. That is a receiving and put away control rather than a stocktaking one, and it is covered properly on accuracy.
- It does not produce a discrepancy report without a stock record. Reconciliation needs something to reconcile against. With no record you get a complete, evidenced physical inventory of what is where, which is genuinely useful and is often the fix in itself, but it is not a discrepancy list. See counting without a WMS.
- It does not read a damaged or obscured barcode. A label turned inwards, wrapped over or torn is the same obstacle for Sentispec as it is for a handheld scanner. Neither reads what is not there. The difference is that an unread position arrives as a finding with the image attached, so somebody can look at it instead of walking the aisle again.
The comparison
Why not drones, and why not autonomous robots.
Both count warehouses today. The difference is what you buy, what you house and charge, and what you have to change in the building to use it.
A vehicle you fly and charge
An indoor drone counts without a driver, which is a real advantage. It also has to fly, and a dense, tall frozen aisle is a poor fit for flight. A camera on a forklift rides a vehicle already rated for the room, and is in the cold only for the length of a run.
Ground robotsA vehicle you house and charge
An autonomous scanning robot is capital or subscription, plus a charging position, plus a facility that suits it. It counts unattended overnight. It is also one more vehicle in your aisles and one more thing to work around.
Sentispec adds no new vehicle to your aisles. The recording unit rides on a forklift that is already part of the traffic your site manages, driven by an operator working to the rules you already have.
What that avoids is everything the autonomous route asks you to arrange first. Five day deployment. Nothing you have to own or depreciate. No facility change, no charging infrastructure, no flight envelope to keep clear. Zero integration effort on day one. And performance in the environments that suit a camera on a forklift better than a vehicle in the air, which in practice means cold stores and high racking.
The economics follow from that. Across the ten sites we publish, break even lands between one month and six months or more, and the fastest recorded is one month. The full set of figures, with the method stated, is on the deployment benchmark. If you are still weighing counting by hand against automating it, start with the comparison with manual stocktaking.
Questions about the mechanism
What people ask once they have seen it run.
Does the driver have to do anything except drive?
No. The driver picks the unit up on the forks, drives the aisle at normal travel speed one shelf level at a time, and sets the unit down at the end. Recognition and reconciliation happen without operator input, which is why one person covers a warehouse in eight to twelve hours.
Can we count while the warehouse is working?
Yes, and that is the normal pattern. A 20 minute run covers 500 to 1,000 pallets and is driven during a break or between tasks, so counting fits inside normal shifts rather than replacing them.
How high can it read?
The unit reads the full height of the racking from the floor, one shelf level per pass, which is what removes the need to raise a person to a high location to read a label. Sites with cold stores and high bay racking use exactly this to take scissor lift work out of the count.
What happens to positions with no pallet in them?
Empty positions are recorded and reported, not skipped. Knowing which locations are genuinely free is the same data set as knowing where stock is, and it is reported on empty and occupied locations.
Next step
See a recording from a warehouse built like yours.
A 30 minute demo walking the same three steps on your racking geometry and your label format, ending in a real discrepancy list. Bring your inventory manager and your WMS lead.