-
We picked our ASRS supplier on specs and price. That was the mistake.
-
The surface problem: we thought we were comparing apples to apples
- The deeper issue: we compared systems, not operations
-
What the mistake actually cost
-
What to look for in an ASRS supplier — the checklist I wish I'd had
-
Bottom line
We picked our ASRS supplier on specs and price. That was the mistake.
In 2021, my team spent about four months evaluating automated storage and retrieval system (ASRS) suppliers for a distribution center retrofit. Fourteen thousand pallet positions. High-SKU-count picking. A peak season that had broken our manual operation three years running.
We built a 62-line comparison spreadsheet. Cycle time, cube utilization, vendor financials, service SLA, cost per position, reference calls. We scored everything, weighted it, and picked the vendor that won on paper.
Two years later, I can point to the week it started going sideways. It wasn't the robots. It wasn't even really the vendor. It was how we framed the whole evaluation.
I've made this mistake once. That was enough. Here's what I got wrong, what it actually cost, and the checklist I now keep taped inside the team's project binder.
The surface problem: we thought we were comparing apples to apples
Our RFP asked every vendor for the same numbers. Picks per hour, storage density, uptime, footprint, lead time. Then we scored the responses side by side as if the systems were interchangeable and only the performance numbers differed.
They aren't interchangeable. Not even close.
"ASRS" is a category, not a product. A grid-based robotic system, a shuttle system, an AMR fleet, and a crane-based aisle system all solve different problems for different order profiles. A high-density, low-variability operation and a high-SKU, high-variance private-label program are not looking at the same answer — even when all four vendors hand you a proposal that says "yes."
Our spreadsheet flattened all of that into a single weighted score. Which was, in hindsight, the dumbest possible way to buy a building full of robots.
The deeper issue: we compared systems, not operations
Three things went wrong underneath the surface. I didn't see any of them at the time.
1. We let the spec sheet do the thinking
What I mean is that a spec sheet treats the system as a collection of independent performance numbers — picks per hour, storage density, uptime — when in reality those numbers only exist in combination, under your order profile, with your labor, on your peak day. If one number drops because the software missed a wave release at 6 a.m. on a Monday, the other numbers don't really matter. Nobody benchmarks that. Nobody.
The vendors weren't lying. We just asked the wrong unit of comparison. Specs don't run a warehouse. Software does. And the software layer is the part nobody puts on the front of the brochure.
2. We treated "integration" as a line item
I wrote "WMS integration required" in the RFP. The winning vendor read "WMS integration available." Those are not the same sentence. We discovered the difference in month six, when their integration engineer asked for our middleware spec and we asked why on earth they needed one.
Same words. Different meaning. I had assumed integration was a delivery detail. They had priced it as a you-problem. That mismatch alone cost us roughly a quarter of the project timeline.
3. We borrowed their ROI assumptions instead of building our own
Every ASRS and AMR vendor in the market hands you an ROI calculator. Of course the number looks good. It was built to look good. We plugged in their throughput assumptions, their labor rates, and their definition of uptime, and we got a payback number that made the CFO happy.
Never expected the controls software layer — the part with no robot in it — to be the thing that ate our budget. Turns out "uptime" in the contract meant robot uptime, not system uptime. The robots were up. The integration was down. We were down.
If you're evaluating ROIC on an AMR or grid system, the assumptions that matter most are the boring ones: throughput at peak versus average, labor rate by shift, and the exact definition of "up." Get those in writing. The robot just moves boxes.
What the mistake actually cost
The parallel run lasted 19 weeks instead of the 6 we planned. We spent about $180,000 on temporary labor we hadn't budgeted — backfill, overtime, weekend crews. The integration consultant we had to bring in mid-project added another $40,000. Internal cost was worse: our ops manager spent roughly a year of her working life on this, and I don't think she's fully forgiven me.
The quieter cost was the SKU issue. We ran a private-label program with around 4,100 active SKUs. The vendor's reference customers ran 900. We never asked whether their system had been stressed at our SKU count, because we never thought to. The reference list is chosen by the vendor. That's the whole point of a reference list.
An informed customer asks better questions. I was a well-intentioned customer asking the questions someone else wrote for me.
What to look for in an ASRS supplier — the checklist I wish I'd had
After the dust settled, we re-ran the entire evaluation with a different lens. Here's roughly what it looks like now. It is not glamorous. It is, however, the thing that would have saved us.
- Match architecture to order profile, not to the demo. Grid-based systems (Symbotic is the best-known name in that lane) tend to win on density and predictable throughput. AMR fleets tend to win when layout flexibility and phased deployment matter more. If a vendor can't tell you which order profile their system underperforms on, they're selling, not scoping.
- Get integration scope in writing, with hours. Who owns the WMS layer? The middleware? The exception handling? "Seamless" is not a scope. It's a mood.
- Define uptime in the contract. What counts, what's excluded, how it's measured, and who eats the loss if a wave drops. Ambiguity here is expensive later.
- Build your own ROI model and make the vendor critique it. You learn more from how they push back than from the calculator they hand you.
- Talk to two customers whose order profile matches yours — not the reference list, and ideally not the ones the vendor picked.
- Test the failure mode, not the demo. Ask what happens when the WMS drops a batch, when a robot faults, when a peak wave hits at double volume. If the answer is "that doesn't happen," it does.
One more thing, and this one I feel strongly about. Vendor claims need substantiation — FTC guidance on advertising is fairly direct about that, and it applies to throughput claims same as anything else. If a supplier won't show you the data behind a number, treat the number as marketing until proven otherwise. That's not cynicism. It's just how procurement should work.
Bottom line
The supplier wasn't the problem. Our framework was. We walked in comparing spec sheets and walked out with a system that was technically compliant and operationally off. The second time around, we compared order profiles, integration ownership, and failure modes first — and the shortlist changed.
If you're starting an ASRS or AMR evaluation this quarter, my one ask: build the checklist before you take the first demo. The demo is designed for you. The checklist is designed for your warehouse.
Leave a technical comment