Every automation evaluation eventually collapses to one architectural fork: fixed-infrastructure storage-and-retrieval (AS/RS, shuttles, cube storage) or flexible mobile robotics (AMRs). Vendors on both sides will tell you their side wins. The decision framework is shorter than the sales process suggests.

What each architecture actually is

AS/RS (automated storage and retrieval systems) is fixed steel: mini-load cranes, shuttles, or cube systems (AutoStore) that bring inventory to stationary pick stations. It maximizes storage density and per-station productivity, and it commits you to a building, a layout, and a throughput ceiling for 10-15 years.

AMRs (autonomous mobile robots) move inventory or workers through existing racking: goods-to-person bots (Locus, 6 River), transport bots, forklifts. It deploys into brownfield sites in weeks, scales by adding units, and trades peak density for flexibility.

The decision turns on four variables

  1. Order profile stability. If your SKU mix and order lines are stable for 5+ years, AS/RS density pays. If e-comm growth, channel shifts, or M&A make the profile a moving target, AMR flexibility is insurance.
  2. Building constraints. Clear height under 24 feet, mezzanines, odd columns, or a lease under 7 years: AMR territory. New build with 36-foot clear and owned land: AS/RS is on the table.
  3. Throughput ceiling. Sustained 5,000+ lines/hour points to fixed automation with goods-to-person stations. Under ~1,500 lines/hour, AMR fleets usually carry it at lower capex.
  4. Capital posture. AS/RS is a $5M-$30M+ commitment with 3-5 year paybacks. AMR programs start under $1M (or RaaS monthly) with 1.5-3 year paybacks. Balance-sheet tolerance is a real constraint, not a footnote.

The failure modes

AS/RS projects fail on integration overruns and demand forecasts that don't survive contact with reality — the steel can't resize. AMR deployments fail on change management and on expectations set by demo-hall throughput that a congested floor can't reproduce. Both fail when the WMS layer is older than the decision.

The operator's list

Before shortlisting either architecture, get three numbers in writing from your own operation: peak lines/hour by hour-of-week, five-year SKU-count trajectory, and your real loaded labor cost including turnover. Every credible integrator will ask for them; a vendor who quotes without them is selling, not engineering.

Independent automation industry analysis — not investment or procurement advice. Capex decisions warrant site-specific engineering diligence.