The ROI of Warehouse Automation

Warehouse automation vendors are not shy about quoting dramatic ROI figures. 300% returns in 18 months. Labour savings of 70%. Throughput increases of 10x. These numbers come from somewhere — usually the best-case deployments in the most favourable conditions — but they are not representative of typical outcomes.

This report is based on data from 84 automation projects across North America (41 projects), Europe (28 projects), and APAC (15 projects) completed between 2022 and 2024. The projects cover five automation categories: autonomous mobile robots (AMRs), voice picking systems, conveyor and sortation, automated storage and retrieval systems (AS/RS), and goods-to-person (GTP) systems. All figures are post-implementation actuals, not vendor projections.

The honest conclusion from this data is that automation ROI is real but varies significantly by deployment type, facility size, and labour market. The businesses that achieve the best returns are those that choose the right automation for their specific operation — not the most technically impressive option — and those that invest in the organisational change required to make automation work.

AMRs — robots that navigate warehouse floors autonomously to support picking, replenishment, or transport tasks — have seen the most rapid adoption of any automation category in the past five years. The drivers are clear: the capital cost has fallen significantly (from roughly $40,000 per robot in 2018 to $15,000–25,000 in 2024 for a mid-range unit), the deployment time is weeks rather than months, and the system can scale up or down by adding or removing robots.

Across the 34 AMR projects in our dataset, the median payback period is 18 months. The fastest payback was 9 months, in a single-storey food distribution centre with high throughput and significant overtime costs. The slowest was 38 months, in a multi-storey facility where the AMR footprint had to be restricted to one floor.

Labour displacement ranges from 18% to 62% of direct pick labour, with a median of 31%. The large range reflects the extent to which facilities had already optimised their manual pick process before automation — a well-slotted, well-organised manual operation leaves less room for AMR improvement than a poorly organised one.

Voice picking — directing pickers using spoken instructions through a headset, with voice confirmation of picks — is one of the most mature and accessible automation technologies in the warehouse. It does not require infrastructure investment (no conveyors, no racking changes) and can be deployed in weeks. The main cost is the headset hardware and the voice system licence.

In 22 voice picking deployments in our dataset, pick accuracy improved from a baseline of 98.1% to a post-deployment average of 99.7%. This is consistent with the widely cited industry figures from Honeywell Intelligrated and Zebra Technologies. The accuracy improvement is driven by the hands-free, eyes-free working mode: pickers are not distracted by a paper list or a scanner screen, and the voice confirmation step forces them to verify the pick before moving on.

The productivity impact of voice is more variable than the accuracy impact. In environments with complex pick lists (many single-unit picks across a wide area), voice typically improves productivity by 10–15%. In environments with simple, high-density picks, the improvement may be closer to 5%. The reason is that the voice interaction itself adds a small time overhead per pick, which is only offset when the elimination of paper handling and the reduction in errors (and therefore re-picks) is significant.

Goods-to-person (GTP) systems — where storage units (totes, trays, or containers) are automatically brought to a stationary picker at an ergonomic workstation — represent the most significant transformation of the picking operation. They are also the most expensive: a full GTP deployment for a mid-size operation typically costs £2–5 million in capital, takes 6–12 months to implement, and requires significant facility changes.

The returns, when the deployment is well matched to the operation, are correspondingly large. In the 12 GTP projects in our dataset with at least 18 months of post-go-live data, the average throughput improvement was 3.4x per pick station compared to the equivalent manual operation. Labour per unit picked fell by an average of 58%, and pick accuracy exceeded 99.9% in 10 of 12 deployments.

The critical success factor for GTP is SKU profile fit. GTP systems work best with a high volume of small, relatively uniform items — eCommerce fulfilment, pharmacy dispensing, electronics accessories. They work poorly with bulky, irregular, or very heavy items that cannot be stored in standard totes. Before investing in GTP, analyse your SKU profile carefully. If more than 30% of your volume consists of items that do not fit standard tote dimensions, GTP is likely not the right choice.

The CFO's key questions for any automation investment are: what is the payback period, what is the NPV at our cost of capital, and what are the risks if the project runs late or over budget? Answering these questions honestly requires conservative assumptions, not optimistic vendor projections.

Use actual labour costs including employer on-costs (National Insurance, pension contributions, holiday pay), not just basic wage rates. Model the productivity ramp-up period post go-live — in our data, it typically takes three to six months to reach 80% of projected throughput after a major automation deployment. Build in a 15–20% contingency on capital cost. And do not assume that displaced labour translates directly to headcount reduction — in most operations, displaced pickers are redeployed to other tasks rather than made redundant, which affects the labour saving calculation.

The automation projects that get CFO approval and deliver on their promises share a common characteristic: they were built on detailed, site-specific analysis rather than generic benchmark data. The benchmarks in this report are useful for sanity-checking assumptions and understanding the range of outcomes. But the business case for your facility needs to be built from your own labour costs, your own throughput data, and your own operational constraints.