WMS and warehouse automation - conveyors, AMR, pick-to-light. What makes sense at what scale

The warehouse automation market looks like a jungle, but there are really only a handful of technologies worth a look in a typical distribution warehouse:

• Conveyors (roller, belt). Moving cartons and totes between zones - from picking to packing, from packing to shipping sortation. The oldest and most easily quantified automation there is; it eliminates the "empty runs" of people carrying a carton in their arms.
• Sorters (cross-belt, pop-up, tilt-tray). Automatically routing parcels to shipping doors, or orders to workstations. They start to make sense at thousands of parcels a day.
• Pick-to-light / put-to-wall. Lights indicating location and quantity during picking, or splitting out a multipick. Cheap compared with the rest, quick to install, and good for a double-digit productivity gain in dense small-item picking zones.
• Voice picking. Voice-directed picking - hands and eyes free, works well in cold stores (gloves) and when handling heavy packages.
• AMR (autonomous mobile robots). Robots that move around the floor with no guidance infrastructure - they navigate by camera and lidar. Two main flavours: goods-to-person (the robot brings a rack or tote to the operator) and transport (the robot hauls pallets or totes between zones, the human only loads them).
• AGV. The AMR's older cousin - it runs along fixed routes (induction loops, magnetic tape, reflectors). Cheaper for simple, repetitive point-to-point flows, but changing a route is a project, not a click.
• AS/RS - miniload, shuttle, pallet stacker cranes. High-density automated storage. The most expensive shelf in the shop, and it makes sense where floor space is expensive (or in cold stores, where a cubic metre costs a fortune) and volumes are large and stable.

Rough brackets, but tested against the numbers from real rollouts, measured in picking lines per day:

• Up to ~500 lines a day: physical automation almost never pays for itself. That volume is optimised with slotting, multipicking and a good WMS - the cost of those changes is a fraction of the cheapest conveyor, and the effect is often similar.
• 500-2000 lines: pick-to-light in the fast-moving zone, voice in the special zones, trucks with terminals and well-planned paths. The first transport AMRs, if the building is long and people are "hauling air" over distances of 60+ metres.
• 2000-5000 lines: conveyors between picking and packing, put-to-wall for e-commerce, a fleet of AMRs on permanent internal transport. This is usually where the first serious ROI calculation lands with a result worth having.
• Above 5000 lines: sorters, goods-to-person, AS/RS - capital projects with budgets in the millions and payback horizons measured in years. At this level the analysis is done with an integrator, not on a blog.

One caveat: these thresholds assume a reasonably stable volume. A seasonal operation that runs 800 lines for ten months and 6000 in November is the hardest case of all - a machine bought "for the peak" sits idle the rest of the year. That's exactly why renting robots (RaaS - robots as a service) is becoming more popular in seasonal businesses: the fleet scales up for a quarter and goes back to the supplier once the season is over.

The most common investment question in mid-sized warehouses today. The short version: an AGV needs guidance infrastructure and runs rigid routes - but it's simpler and cheaper for a repetitive flow like "production → warehouse, the same run 200 times a day". An AMR navigates on its own, works around obstacles, and changing a route is a software configuration; you pay for that flexibility with a higher unit price. A person with a truck stays unbeatable in chaos: odd dimensions, exceptions, "needed yesterday" priorities - everything you can't capture in a rule.

The maths is done on annual costs. The fully loaded cost of a warehouse worker - overheads, overtime and turnover included - is realistically 90,000-130,000 PLN a year per head in Poland in 2026, and internal transport in a large building can eat up 2-4 heads per shift. A transport AMR bought outright is an outlay on the order of 150,000-300,000 PLN plus maintenance; rented, it's a few thousand zloty a month per robot. On a two-shift operation the robot "works" 16 hours with no breaks and no resignations - and it's the two-shift pattern, not the hourly rate on its own, that most often decides the outcome of the calculation.

An automated warehouse has three layers of control, and it's worth understanding the division of roles, because more projects die at the seams between those layers than anywhere else:

• The WMS manages the business logic: stock, orders, priorities, waves, documents. It knows what needs to be done and in what order.
• The WCS (warehouse control system) drives the equipment in real time: conveyor motors, sorter diverters, tunnel scanners. It knows how to physically carry out the move here and now. It's usually supplied by the automation vendor.
• The WES is an intermediate layer found in larger installations - it balances work between the automated and manual zones, for example metering out orders so the sorter doesn't starve the packing stations.

In a mid-sized installation the most common model is a simple one in practice: the WMS issues transport tasks over an API, and the AMR fleet system (or the conveyor's WCS) reports back that they're done. Standardisation helps - VDA 5050, as a common language for talking to robot fleets from different vendors, is making its way into warehouses too, though outside automotive the vendors' native APIs are still more common. The key question to ask before buying: "what API do you have and what exactly can I task through it" - and the answer "we've got a CSV export every 15 minutes" should end the conversation.

The payback case for automation usually breaks down in three places. First, people compare the cost of the machine with the hourly rate alone, ignoring overheads, turnover (recruiting and onboarding a new operator realistically runs to 10,000-20,000 PLN), absences and peak-season overtime. That understates how well automation pays. Second - the other way round - the costs around the machine get left out: service (typically 8-12% of the installation's value per year), parts, inspections, training for the maintenance team, spare capacity for breakdowns. Third, people assume the volume from the sales spreadsheet rather than from history - a machine sized to an optimistic growth forecast has an ROI on paper and a loss on the floor.

A healthy payback horizon for conveyors and pick-to-light is 18-30 months, for AMR fleets 2-4 years, for AS/RS 5-8 years. If the spreadsheet shows a return in 12 months, something is most likely missing from it. And one thing you only see once it's live: automation stiffens the process. A conveyor bolted to the floor defines the flow for a decade - a change in the business profile that's a week of rearranging for a manual warehouse can be a multi-million project with heavy automation. Flexibility has a value that no ROI spreadsheet contains.

Most projects in Poland aren't a new facility built around automation (greenfield), but retrofitting a working warehouse. The list of things to check before anyone shows you a quote:

• The floor. AMRs tolerate unevenness of roughly 5-10 mm over 2 metres depending on the model; an old floor with expansion joints and pot-holes can add hundreds of thousands of zloty of grinding and screeding to the project.
• The Wi-Fi network. A robot fleet needs full, stable coverage with seamless roaming - the same thing that plagues handheld terminals, except that a robot stops dead when it loses the connection. A radio survey before the project is mandatory.
• Power and charging. Fleet charging stations, conveyor power, protection devices - in older buildings the supply sometimes needs to be upgraded.
• Peak throughput, not the average. A machine designed for the daily average will choke at peak hours. You design for the 85th hourly percentile, with headroom for growth.
• Operating during installation. Installing a conveyor in a live warehouse is 2-4 months of work in zones, at night and at weekends - with a fallback plan for every Saturday. The real cost of that gymnastics is rarely in the integrator's quote.

A sensible way into automation looks much the same every time. Data first: the order profile from the WMS (lines, units per line, hourly distribution, seasonality) and a flow map - from where to where, how many times a day, what the bottleneck is. Then picking the single process with the greatest potential - most often that turns out to be internal transport over long distances, or the packing zone. A pilot of 3-6 months: two or three rented robots, or pick-to-light on a single rack module, with the KPIs measured before and after. Only after the pilot do you decide on scale.

A rented pilot costs a few tens of thousands of zloty and answers the questions no quote will: how the machine copes with our specific mess, how much it really demands of the maintenance team, how people react to it. That last point can be decisive - a team that treats the robot as a threat can quietly kill the project; a team the robot has relieved of its most hated runs across the building will defend it as one of their own.

• Automating before you've put the process in order. If picking is limping because of bad slotting and no scanning, a robot won't fix a thing - it'll just move the consequences of those same mistakes around faster. The order is: process, then system, and only then the hardware.
• Choosing the technology before analysing the order profile. "Let's buy AMRs because the competition has them" - while the order profile shows that 70% of the time is eaten by packing, not transport. Data first, catalogue later.
• No plan B for a breakdown. The machine stops - now what? An operation with no manual workaround path (even a slower one) will halt shipments for a day at the first serious sorter failure. The workaround is designed alongside the machine, not after the fact.
• Underestimating the integration. The budget has the robot and the installation in it, but not the WMS-WCS interface, the testing, the exception scenarios (what happens when the robot is carrying a tote and the order has just been cancelled?). Integration and testing are realistically 15-25% of the project budget.
• Automating everything at once. One process, a pilot, a measurement, then scale. "The whole building in a year" projects have an ugly track record of slippage - and a warehouse halfway through a rebuild is less productive than it was before it started.

Warehouse automation has stopped being exotic - conveyors, pick-to-light and rented AMRs are now within reach of mid-sized companies. But the order stays the same: first an orderly process and data from the WMS, then analysis of the order profile, then a pilot of one technology in one process, and scale at the end. You automate a measured bottleneck, not a trend that's fashionable at the trade show.

Weaver WMS works with automation over a REST API - the system issues transport and picking tasks, receives completion confirmations and keeps stock consistent regardless of whether the task was done by a person with a terminal or by a robot. We've delivered integrations with conveyors, sorters and AMR fleets - and from that experience, one piece of advice to finish on: before you invite an automation integrator in, pull the hard data on your own flows out of the system. A conversation that starts from numbers instead of from a catalogue ends with a cheaper and better-matched project.