A vendor's pitch deck says the fleet pays for itself in twelve months. The finance committee approves it on that basis. Two years later the robots are running fine — and the payback line still hasn't crossed zero. This is the single most common disappointment in warehouse automation, and it almost never comes from bad robots. It comes from a business case built on four optimistic assumptions that quietly stretch a 12-month promise into a 30-month reality.
Here is where the projections break, and a checklist to sanity-check one before you sign.
1. Labor savings are inflated at both ends
Most models start by multiplying the number of pickers a fleet "replaces" by an hourly wage. Two things go wrong. The wage is understated: a $18/hour picker actually costs $25–$30/hour fully loaded once benefits, workers' comp, and turnover are counted, per integrator ROI modeling from CxTMS. That cuts *for* the project. But the headcount is usually overstated the other way — automation rarely deletes whole roles. It absorbs peak surges and redeploys people to exceptions, replenishment, and QA. Our view: a credible model counts *hours removed at the fully-loaded rate*, not *heads eliminated at the base wage*. If a business case leans on cutting named positions, ask which ones, and whether your operation can actually run without them on a bad day.
2. Throughput is taken at the vendor's face value
Rated picks-per-hour are measured under ideal conditions — full batteries, clean aisles, cooperative SKUs. Real warehouses have congestion, charging windows, and awkward inventory. The practical correction that experienced buyers apply is to use 70–80% of the vendor-quoted rate and to budget 4–8 weeks of ramp before the fleet reaches steady-state performance (per CxTMS). Skip that and the first quarter's numbers will look like a failure when they are simply normal.
3. The integration and facility "tax" is left out
This is the big one. The robot hardware is often the *smaller* half of the project. Integration with your WMS/WES, plus racking modifications, fire-suppression upgrades, and permitting, routinely add 20–40% on top of hardware cost (CxTMS), and Armstrong, a separate integrator, puts facility modifications alone at 20–30% and warns that a cheap AGV fleet needing "custom middleware and constant troubleshooting can drain more cash than a premium solution that plugs in seamlessly." Two independent sources landing in the same 20–40% band is the tell: if a projection shows integration near zero, it is not conservative, it is incomplete.
4. Maintenance is undercounted
Annual maintenance and support contracts typically run 10–15% of purchase price (CxTMS) — every year, compounding across the asset's life. A five-year hold at 12% adds well over half a year of the original hardware cost back onto the ledger. Vendor payback charts frequently model year one and stop.
What realistic looks like
Stack those four corrections and the picture changes. Against a vendor's 12-month headline, documented real-world ranges run 8–14 months for high-velocity e-commerce but 18–30 months for cold storage, with the "36-month disappointment" outcome landing exactly when labor savings were inflated, integration was ignored, and maintenance was skipped (CxTMS). Armstrong frames the honest spread as automation that "pays for itself in eighteen months or drags on for five years." Our view: for most mixed operations, underwrite to a 2.5–4 year payback and treat anything under 18 months as a claim to be proven, not a plan to be funded.
A five-minute sanity-check
Run any warehouse-automation projection through these questions before it reaches the finance committee:
- Labor line: Is it hours-removed at the fully-loaded rate ($25–$30, not $18), or heads-eliminated at base wage? The latter is a red flag.
- Throughput: Is the model derated to ~70–80% of the quoted rate, with a 4–8 week ramp built in?
- Integration: Is there a line worth 20–40% of hardware for WMS work, racking, fire suppression, and permitting? If it's blank, the model is incomplete.
- Maintenance: Is 10–15%/year carried across the full hold, or does the chart stop at year one?
- Utilization: Does the case assume the fleet is busy at your *average* volume, or does it survive your slow season too? Seasonality kills utilization-dependent ROI.
None of this argues against automating — the labor math is real and the technology works. It argues for buying on a model that survives contact with your actual dock, your actual SKUs, and your actual off-season. The projects that disappoint are rarely the ones that automated the wrong task. They are the ones that funded the right task on the wrong numbers.
For the categories where these economics play out, see our warehouse robot and autonomous mobile robot pages.
