AI in the warehouse and e-fulfilment: order picking, inventory and the staff shortage in 2026

Why the warehouse feels the pressure first

The staff shortage in logistics is not a temporary spike but a structural feature of the labour market. ABN AMRO notes that the shortage is returning and is hitting across the board, with tens of thousands of unfilled vacancies in transport and logistics once again. In the warehouse, that means in practice unfilled shifts, heavy use of agency staff and a peak period that is more of a struggle to cover every year.

That is the central point to keep clearly in view with every story about AI and robots: no algorithm picks an order and no model puts an employee on the payroll who is not there. The gain lies elsewhere: with the same number of people or fewer, you get more orders out of the door without errors, and the work becomes less vulnerable to absence and staff turnover.

What automation and AI really do in the warehouse

It is useful to separate two things that are often lumped together: physical automation and AI software. Physical automation is the tangible part, such as robots and systems that bring goods to the picker. According to Warehouse Totaal (a Dutch warehousing trade publication), order picking, palletising and internal transport are the areas where demand for warehouse robots is growing fastest. AI software is the decision-making part around it: routing through the warehouse, task allocation, forecasting busy periods and predictive maintenance of the equipment.

In practice, the two reinforce each other. A system that optimises pick routes and allocates work intelligently among people already takes walking time and search time out of the day, even without expensive robots. And a warehouse that does robotise needs good software all the more, to keep those machines fully occupied. Note that research in the sector shows that it is not the cost but the complexity of implementation that is the biggest hurdle. That is an argument for starting small and focused rather than converting the whole warehouse in one go.

Three places where the gain is tangible

The practical gain rarely lies in a spectacular conversion. It lies in removing waste that is currently hidden in day-to-day operations. Three places stand out:

• Order picking: smarter pick routes and task allocation raise picking speed and lower the error rate, so that you can absorb peak days without a proportionate increase in people. This is often the shortest route to a visible result.
• Inventory and demand forecasting: by using historical order data, software forecasts demand for each item more accurately, so that you have fewer lost sales and leave less dead capital on the shelf. Better forecasting also makes your workforce planning less hectic.
• Returns: in e-fulfilment, the returns flow is a large, labour-intensive cost item. AI can help to assess returns faster and route them back into stock, provided your record-keeping is in order. Start here only if returns really are a large part of your operation.

Measure it in euros, and be critical of vendor claims

The effect of smarter software or automation varies widely by warehouse, by product range and by order type. A fulfilment company with many small, varied orders gets something different out of it than a wholesaler with regular pallet loads. Generic savings percentages from a sales pitch therefore say little about your own situation. Ask a vendor for a trial with your own order data rather than for a marketing figure.

Make the calculation explicit in euros and hours. First record how many orders you pick per hour, what your error rate is, how many metres a picker walks and how many agency staff you bring in during the peak. That is your baseline measurement. Only against that background can you judge whether an investment in software or robots pays for itself. Without a baseline measurement, no promise about speed or savings can be verified.

The human remains responsible, even in a smart warehouse

Whichever pick route or inventory decision a system proposes, someone remains responsible for what actually happens on the floor and in dealings with the customer. A plan that is optimal on paper but ignores safety, working conditions or an agreement with a customer is not a usable plan. Good software makes clear what a proposal is based on, so that your team leader can make well-founded adjustments.

Moreover, Article 4 of the EU AI Act (Regulation (EU) 2024/1689) has applied since 2 February 2025. Since the amendment by Regulation (EU) 2026/1744, in force from 27 July 2026, it has asked organisations to take measures that support the AI literacy of employees who work with AI systems. They do not have to guarantee a specific level. For a warehouse, what matters in practice is that team leaders and planners understand what an AI proposal is based on, where the assumptions and limits lie, and when they must intervene manually. That is not a formality, but exactly what keeps a smart warehouse manageable.

Where to start in practice

Do not start with a large automation project, but with the question of where your current operation loses the most time, makes the most errors and uses the most agency staff. Choose one process, for example order picking on your busiest line, and establish how many orders go out per hour, what the error rate is and how many agency staff you bring in during the peak. That immediately gives you the baseline measurement against which you set every subsequent step.

Our free Practice Scan helps you choose that first process and sharpen the baseline measurement, so that any investment in smarter software or automation rests on real figures from your own warehouse and not on a general promise.