Automated warehouse: how to choose between AS/RS, goods-to-person, vertical lift modules and AMRs

"We want to automate the warehouse" is a sentence that, in technical terms, can point to at least eight different projects. Investment varies by an order of magnitude, installation times range from a few weeks to more than a year, and reversibility goes from total to almost none. The most serious risk is choosing the wrong technology family and locking in for ten years an operation whose mix, channels and volumes will change in the meantime. Overpaying for a system is a smaller problem by comparison. Anyone who has to defend an automated warehouse in front of management needs three things: to understand which physical problem is being solved, to have the numbers that drive the choice, and to be able to calculate the return with written assumptions.

Eight families of solutions, and what they really solve

Before talking about suppliers it pays to understand which physical problem you want to solve, because each family buys one thing and pays for another. The reviews by Roodbergen and Vis (2009) on automated storage and retrieval systems and by Azadeh, de Koster and Roy (2019) on robotized warehouses describe the same taxonomy from the research side. The table translates it into a buyer's questions.

One clarification heads off half the misunderstandings at tender stage. Transport and storage are two separate problems: AGVs and AMRs solve transport and do not add a single pallet position to the building, while AS/RS, vertical lift modules, cube storage and pallet shuttles solve storage and density. Goods-to-person sits in the middle and solves picking productivity. If the stated problem is "nothing fits anymore" and the proposed solution is a fleet of transport robots, there is a mismatch to clear up before signing. The differences between the two vehicle families are covered in our comparison of AGVs and autonomous mobile robots.

The nine parameters that drive the choice

The choice is made on data, and there are nine figures. The most decisive is order lines, per day and in the busiest hour, split between each, case and pallet picking: they size stations, robots and shuttles, and they have to be measured over at least twelve months, not estimated. Next come the number of items actually moved in the last year, which differs from those simply present in the system, and the ABC turnover curve, to be rebuilt from your own data because it decides which part of the warehouse is worth automating. Often class A deserves goods-to-person and class C stays on static shelving.

Then come the building and the goods. Clear height under the roof structure, net of fire protection and lighting, decides whether a system that uses height has anything to use. Required density, in pallet positions or locations per square meter, only matters if covered floor space is expensive or scarce. The weight and format of the load unit are a technical precondition: many projects get complicated because a small share of items is not compatible with automation. Three context parameters complete the picture: the ratio between the busiest day and the average day, whether the building is existing or new, and the five-year growth horizon with its uncertainty.

The thresholds in lines per day below which goods-to-person struggles to pay back, and above which it almost always does, are worked out step by step in our article on when a goods-to-person automated warehouse pays off, with the assumptions stated.

Where the cost sits, and what changing your mind costs

The breakdown row is the one business cases underestimate most. In an AS/RS with one stacker crane per aisle, a failure of that machine puts the whole aisle's stock out of reach. Stated availability, calculated as A = MTBF / (MTBF + MTTR), therefore has to be read together with the contractual restoration time and the availability of spare parts in Italy. A fleet of mobile robots degrades gradually instead: losing one robot in twenty removes 5% of capacity without blocking a zone.

On timing there is a distinction to make. For standard goods-to-person systems the manufacturer DIPITECH represents states a physical installation of two to three weeks, as reported on the goods-to-person automated warehouse page. Data analysis, WMS integration, training and fine-tuning extend the overall project, and they belong in the schedule from the start.

How to calculate the return: a worked example with written assumptions

The method has five steps, and the third separates a serious business case from an optimistic one. Measure the labor hours of the current process on the flow to be automated, then estimate those of the automated process using the system's contractual performance, not the brochure figures. Next apply a conversion factor, because not every hour saved becomes cost removed: only hours that turn into fewer hires, less temporary labor, less overtime or redeployment to value-adding work count. The resulting hours are valued at the fully loaded company cost per hour, which includes social charges and shift premiums and is not the gross wage. Finally, subtract the system's annual costs and run at least two volume sensitivities.

Here is an example in hours, with our own assumptions written to show the method. A distribution center runs two shifts for 240 days a year and ships 3,600 order lines a day, picked by the each. Assume a measured manual productivity of 65 lines per hour per operator, which means 55.4 labor hours a day. With goods-to-person, assume 200 lines per hour per station, a figure the supplier should guarantee by contract, which means 18 labor hours a day. The gross saving is 37.4 hours a day, 8,976 hours a year. With an 80% conversion factor, 7,181 hours remain that can actually be removed. Simple payback is the investment divided by the value of those hours at your fully loaded hourly cost, net of the system's annual costs. If volumes fall by 25%, removable hours fall in the same proportion and payback stretches by at least a third. So the business case should be presented as a range rather than as the best number.

Two items stay out of the calculation because they vary too much from case to case, and they often improve the picture. The first is fewer picking errors, whose full cost has to be measured internally. The second is the value of freed space, which becomes dominant when the real alternative is renting or building another warehouse. Honesty also means counting the items on the other side, such as electricity, software fees and training. For the mobile robot side, the full analysis of recurring costs is in AMR ROI, TCO and payback.

WMS, WCS and integration: where projects get complicated

The division of roles has to be settled at quotation stage, because it is the first source of disputes. The WMS manages item data, stock, locations, waves and orders. The WCS (warehouse control system), or WES when it includes execution logic, orchestrates the physical devices: it assigns missions to robots, sequences totes arriving at stations, and manages queues and recoveries. The recurring mistake is assuming the existing WMS can do real-time sequencing, which almost no general-purpose WMS does well. Before choosing, check three points: whether the WMS exposes documented APIs or message interfaces, whether it handles dynamic locations, essential with mobile shelves and cube storage, and whether it can work both in waves and in continuous flow. Integration belongs in the budget from the start, because discovering it halfway through always costs more.

On the fleet side, when vehicles come from different makers, interoperability becomes the issue. The VDA 5050 specification, which reached version 3.0.0 in March 2026, was created to let AGVs and AMRs from different suppliers talk to a single fleet control. Requiring it in the tender protects against lock-in even when you buy from a single maker today.

Energy and network: how many vehicles you really need

In mobile robot systems the energy strategy decides how many vehicles are needed. Opportunity charging docks the robot at stations during the natural pauses of the cycle: nobody has to handle batteries, but it needs well-distributed stations and a fleet control that plans charging as a mission in its own right. Battery swapping maximizes machine time over three full shifts, at the price of a manual task or an automatic swap station, a stock of spare batteries and a dedicated room. On chemistry, Battery University's reference data credit lithium iron phosphate with more than 2,000 cycles and better thermal stability than NMC, against lower energy density. The figure to ask for in a quote remains the vehicle's usable availability on the real mission profile, charging included, and station sizing is explained in our article on AMR fleet charging.

Many faults blamed on robots are really network coverage problems. The specification should require minimum signal levels and maximum roaming times between access points in every operating area and a dedicated network separate from the guest one. It should also call for a check with the racks full, because goods attenuate the signal far more than the empty structure surveyed on site.

Safety, racking and fire protection

Driverless vehicles, AGVs and AMRs alike, fall under ISO 3691-4:2023, harmonized as EN ISO 3691-4:2023 for CE marking, which covers person detection, laser scanner protective fields, speed management, protective stops and the preparation of operating zones. Stacker cranes have their own reference in EN 528, now in the 2021 edition with the A1 amendment of 2022, and vertical lift modules and power-operated mobile racking in EN 15095:2007+A1:2009. Risk assessment follows ISO 12100, and on the European market the whole system remains machinery subject to CE marking, with Regulation (EU) 2023/1230 applicable from January 20, 2027.

Three points deserve attention at design stage. In shared areas, scanner protective fields must be set on the real speed and stopping distance of the loaded vehicle, because mass changes stopping distance considerably. In goods-to-person and AS/RS systems the robotized zones are usually segregated, and interface points, maintenance access and safe-entry procedures have to be defined precisely. Installation tolerances and periodic racking inspections follow EN 15620:2021 and EN 15635:2008, and in an automated warehouse a deformed upright can push a whole shuttle-served aisle out of tolerance. In compact high-density systems, finally, the fire protection criterion is not in the robot standard: the topic is developed in our piece on fire protection and sprinklers in automated warehouses. On ergonomics, removing the walking and bringing picking into the ergonomic zone reduces exposure to musculoskeletal disorders, which EU-OSHA ranks among the most common work-related health problems in Europe.

Five mistakes that sink projects

The first is sizing on the worst day of the last three years, which means buying capacity that sits idle most of the time. It is better to size on an agreed high percentile, for example the 95th, and cover the tail with flexible levers such as overtime, backup manual stations and pulling work forward into the previous days. The second is ignoring receiving and shipping: automating picking while leaving the docks unchanged moves the bottleneck downstream. A system that prepares thousands of lines a day with two shipping docks saturated in the afternoon has solved nothing.

The third is designing on old turnover data or an unusual year, which produces systematically wrong sizing. You need at least twelve rolling months of real lines, with seasonality visible and discontinued items excluded. The fourth is underestimating the floor, the most banal and most frequent cause of delay on site: mobile shelves and stacker cranes need tight flatness tolerances, and even AMRs, more forgiving, suffer from joints, level changes and steps. The instrumental check belongs before signing. The fifth is treating the wireless network as an IT detail, with a survey run in an empty warehouse that does not describe operating conditions.

The recommended path: data, simulation, pilot, growth

The most effective way to reduce risk is to proceed in phases, deciding at each step whether to continue. Start with data analysis: twelve months of order lines, ABC curve, load unit master data, layout and flow mapping, and current labor hours by phase. Without this step any quote is a guess. Then comes discrete-event simulation of the flow, which checks behavior on the busiest day and in transients, sizes stations and fleet, and brings upstream and downstream bottlenecks to light before they become a physical problem.

A pilot on one zone or product family validates data and integration, with measurable targets set in advance: actual lines per hour, error rate, availability, ramp-up time. Then growth happens in phases, carrying the pilot's lessons into the specification. Modular technologies, AMRs and goods-to-person, grow by adding units. Fixed ones have to be sized right at the first installation, which is why simulation is mandatory for them in practice.

What to send us for an assessment

The nine parameters double as the list of what we need to tell you which automated warehouse makes sense, and which doesn't:

• order lines per day and in the busiest hour, over twelve months, split by each, case and pallet picking;
• number of items moved in the last year and the ABC turnover curve;
• floor plan with clear height under the roof, columns, docks and building services;
• load units: formats, weights and the share of non-standard items;
• current density and the space you lack, or that costs you;
• ratio between the busiest day and the average day, with the peak periods;
• existing or new building, and acceptable downtime constraints;
• expected volume growth over five years;
• WMS or ERP in use and integration points.

You can send them by email to info@dipitech.com or on WhatsApp. DIPITECH works as a technology partner for selected foreign manufacturers on the Italian market. Its contribution focuses on reading operating data, turning it into verifiable specification requirements, and choosing technology and maker consistent with those requirements. The available solutions are described on the pages about the goods-to-person automated warehouse, robotic forklifts and AMRs for pallets and industrial automation and material handling robotics.