Nobody Cycle-Counts Rack 14 in August. Inspection Drones Cut That Walk by Roughly 90%.

When did anyone last put eyes on the top of the bulk rack?

Not scan it. Not adjust it in Epicor Eagle after a picker reported a shortage. Physically look at levels three and four of the bulk racking, where the exhaust, the radiators, the fenders and the overstock filters live, twenty-two feet up. In most distribution centers the honest answer is: at the annual physical, in a two-person order picker, on a Saturday, by somebody who was counting his fifth hour and wanted lunch.

That is not a criticism of your warehouse crew. It is the shape of the job. Cycle counting is the first thing that gets pushed when the will-call window is three deep, when a national account calls a stock order in on a Thursday, and every August when half the DC is on vacation and the A/C season is still running hot. The count gets skipped, the on-hand drifts, and by October you have a number in the system that nobody trusts.

What drifted on-hands actually cost you at the will-call window

A wrong on-hand does not stay a paperwork problem. It becomes a phantom pick: the counter sells a part the system says is there, the picker cannot find it, and now a shop with a car on a lift is being called back and told no. Sometimes you cover it from another branch. Sometimes the sale walks. Either way your first-call fill rate — the number your national program group asks about, the number your installers judge you by without ever using the term — slips.

Then there is the other direction. Stock you have and do not know you have, so the buyer orders it again. That is cash sitting in the rack twice, and in a business where a good year is a few points of net, cash in the rack twice is the most expensive mistake available to you.

The count is also a safety and insurance item. Every hour somebody spends at height in an order picker with a scanner is an hour of exposure, and your workers’ comp carrier is aware of it even if the count schedule is not.

flowchart TD
  A["Friday close: on-hand file exported from Epicor Eagle"] --> B["Unit one flies aisles 1 to 12 overnight"]
  A --> C["Unit one flies aisles 13 to 24"]
  A --> D["Unit two reads bulk rack levels 3 and 4"]
  B --> E["Label reads matched line by line against on-hand"]
  C --> E
  D --> E
  E --> F{"Variance over 2 pieces or 250 dollars?"}
  F -->|Yes| G["Warehouse lead re-counts that location Monday at 6 a.m."]
  F -->|No| H["On-hand adjusted, nobody goes up in the picker"]

What changed: the walk-around went autonomous, and the output is a file

Drones in warehouses are not a 2026 idea. They have been in pilots since about 2019 and most of those pilots died for the same three reasons: they could not hold position indoors without GPS, they could not read a beat-up label at an angle, and the output was video that somebody then had to sit and watch, which is not a saving.

What moved this year is that autonomous inspection crossed out of pilot and into production. Hyundai put AI-powered autonomous drones to work at its Georgia plant and cut inspection time by roughly ninety percent. Humanoid platforms moved onto real facility floors in 2026 as well. The reason it works now is dull and important: the units navigate indoors on their own, dock and recharge themselves, read labels reliably enough to be trusted, and hand back a variance list rather than footage.

The change worth understanding is not that a drone can fly through your warehouse — it is that the drone returns a count file your system can act on, so a human only touches the locations that disagree.

Take Hyundai’s ninety percent as directionally useful, not as your number. Their environment is an assembly plant with controlled conditions and consistent parts. A parts DC has mixed-SKU pallets, half-empty cartons, and a returns area that looks like a garage sale. Expect less. Expect enough.

A Saturday count that does not shut the branch

Here is the sequence a mid-size DC actually runs. Friday at close, the on-hand file goes out of Eagle or Vision. Overnight, with nobody in the building, the units fly the aisles and read the location labels and the carton labels level by level. By 5 a.m. Saturday there is a variance list — not thirty-eight thousand bin locations, but the four hundred that disagree with the system.

Your warehouse lead comes in at six, works the variance list with one other person, and is done before lunch. The order picker goes up maybe forty times instead of four hundred. Will-call opens on time. The buyer gets an accurate on-hand for the Monday stock order instead of a number he has been mentally discounting by five percent for two years.

The other thing that changes is frequency. The reason nobody cycle-counts the bulk rack weekly is that it costs two people and half a day. When the count is a scheduled overnight flight, weekly on the A-class fast movers and monthly on the rest stops being a fantasy schedule pinned to the office wall and starts being what actually happens.

Hours, and the fill rate you get back

Illustrative assumptions for a 110,000-square-foot distribution center with roughly 38,000 bin locations. Get your own quote on equipment — do not take a price from a columnist — and put it against these hours.

Now the part that matters more. Suppose the DC ships 1,200 lines a day and 0.4% of them are phantom picks — five a day, 1,250 a year. Suppose one in four of those ends in a lost sale rather than a scramble, at an average line of $46 and thirty-five points of margin. That is 313 lost lines, about $5,000 of gross profit, plus the goodwill you cannot put a number on with a shop that had a car on the lift. Cut phantom picks in half with an on-hand you can trust and you have another $2,500 and a reputation improvement.

What the drone cannot tell you

It reads labels. It does not open boxes. A carton labeled as a pair of front struts that contains one strut and a Chinese newspaper is going to count as a pair, and you will find out at the counter. Physical verification of high-value, high-theft lines — catalytic converters, wheels, electronics modules — still needs a person putting hands on it.

It also will not judge condition. A leaking shock, a rusted rotor, a crushed corner on a radiator core, a wet carton under a roof leak, a swollen battery — those are eyeball calls, and in this business condition is half of your inventory value. Dirty core sorting is not going anywhere either. Nobody has built a machine that can tell a rebuildable core from scrap the way a core clerk with fifteen years in can.

And there is the boring operational reality: something has to charge, maintain and troubleshoot the units, and in a forty-person distributor that becomes somebody’s extra job. Ask, before you sign, who that is and what happens the week they are on vacation.

Frequently asked questions

Will this fly with my insurance carrier and with OSHA?

Talk to your carrier before you buy, not after. The argument you are making is a good one — fewer hours at height in an order picker is a lower exposure, not a higher one — but overnight autonomous equipment in an occupied building is a conversation, and indoor flight in a commercial building has its own rules. Have it once, in writing, up front.

Our racking is a mess and half the location labels are torn. Does that kill it?

It hurts more than anything else on the list. Label quality is the single biggest driver of how well this works, and cleaning up location labels is a week of a temp’s time that pays for itself whether or not you ever buy a drone. Do that first regardless.

Is this only for big distribution centers?

Today, mostly yes. If you run four jobber stores with a back room each, the economics are not there and a disciplined weekly cycle count on your A-movers gets you most of the benefit. The break-even sits somewhere around a real DC with high bulk racking, where the count requires equipment and two people every time.

What do I do with the hours I get back?

Do not plan on cutting heads — warehouse labor is hard enough to keep. Put those hours into the returns and core room, which is where most distributors are quietly losing money and where nobody has time to work the backlog.

Where to start on Monday

Run one aisle, one night, with the vendor’s unit, against a count your own crew does the same week by hand. Compare the two lists line by line. You are not testing whether drones are impressive; you are testing whether it reads your labels, in your building, on your racking. If it matches your crew on one aisle it will match them on twenty-four.

When your on-hands get accurate, something predictable happens at the front of the building: your phone gets busier, because counter pros and shops start believing the availability you quote. That is the part CallSphere handles — voice and chat agents on the counter and will-call lines that answer after hours, take the year, make and model, and book the callback or hold the part. It does not count your racks. It makes sure the calls your accurate count earns you actually get answered.