Retail will have to cut the amount of work needed to serve its own offer. Automation will help, but decisions about assortment, logistics and the scope of services will matter just as much. An AI model can prepare the calculation. The assumptions, and the choice of what a company is willing to give up, will belong to management for a long time yet.

Today's process can be well documented and executed exactly as written, and still stop paying for itself. All it takes is a change in the cost of the person doing it, in the margin, or in the amount of work per basket served.

The pressure shows in the data. According to Poland's statistical office, the average gross monthly wage in retail rose from 6,335.35 zloty in 2024 to 6,811.24 zloty in 2025, an increase of 7.51%. That is a nominal increase in the average for surveyed enterprises with at least ten employees. It is not the employer's full cost, nor the rate of a cashier in a small store. GUS, Statistical Bulletin 7/2026, table 19.

None of this assumes shop assistants will suddenly run out. The Occupational Barometer for 2026 forecasts a national balance for shop assistants and cashiers, with shortages among warehouse workers and qualified accountants, among others. The pressure depends on the role and the location. Occupational Barometer forecast 2026.

The problem starts when the margin stops covering the staffing needed to do the work. At that point one more amendment to the procedure will not help.

The challenge: count how much work the offer itself creates

Every additional item needs handling. Changing a display means physical work. A promotion has to be applied, explained and settled. Fast order collection means preparing the order when the customer wants to pick it up, not when the store happens to have a free hand.

Consider a campaign that lifts sales in a category. The shelf has to be replenished more often, an extra delivery received, the display changed. Part of the purchases shift forward from next week, and some customers pick the discounted product instead of the higher-margin one. In the spreadsheet the extra work does itself, of course.

In the store its cost can eat the benefit of the campaign outright. If AI makes it easier to prepare more promotions like this, the company will generate work for itself faster. Coming up with the idea gets cheaper. Carrying it out physically still takes people, time and space.

So the offer has to be created together with the plan for executing it. That can mean fewer display changes, different delivery packaging, concentrating order picking in selected locations, or dropping services the customer is not funding with their margin. Labour intensity should enter the commercial decision before the store starts, to put it plainly, running on fumes.

Fewer items does not automatically mean a better result. A labour-intensive product can attract a customer who leaves margin across the whole basket. That is why the scope of the offer has to be judged together with customer behaviour, availability and the cost of service, not by the margin on a single item.

Some local decisions should simply disappear

If people in many stores approve similar purchase requirements every day, it is worth checking what they are really deciding. Some cases require knowledge of the local situation. Others are one more execution of the same rule.

For the latter, approve the rule, set its boundaries and leave the exceptions to a person. An agent clicking through every former approval preserves a process structure that is no longer needed.

In DEM-02, the manual path for releasing a requirement in ORPR, you can see this concretely. Moving a defined class of requirements onto a permitted automatic path reduces the share of manual work. The right to purchase and the limits of the commitment still have to be established, otherwise automation will simply multiply a bad purchase.

A similar change awaits administration. Handling correct documents centrally reduces the need to rebuild a back office in every location. Specialists concentrate on discrepancies and on the result. The condition is trustworthy data about execution, because a digital invoice does not yet confirm that goods were received.

A large chain finds it easier to spread the cost of integration, devices and supervision. Part of that advantage can be made available to independent stores through shared services. A buying group, one of the six operating models described in ORPR, from an independent store to a fully owned chain, can extend that back office with data, settlements and know-how. For the owner of a small store the benefit also includes time recovered. Work done after hours is not free just because it does not appear on the payroll.

The social costs will be real. Some positions or hours will lose their justification, and the employees who remain may end up with a larger share of complaints and difficult exceptions. Automation does not guarantee better work. Workload has to be watched deliberately, and a path has to be left by which beginners gain the experience they will later need in order to supervise.

Saving time is not the same as saving money

A few minutes less per task does not always allow you to reduce staffing. Keeping things running despite an unfilled position is a different benefit from cutting cost. Serving more stores without growing the administration is yet another result.

In every case supervision has to be counted too. A camera that detects a gap does not restock the product. An agent that reports a discrepancy does not supply the person who will resolve it. A queue of exceptions can grow faster than the capacity to handle them.

ORPR is, for me, the map of that kind of redesign. It helps show where the performer changes, where an obligation moves to, and which conditions still have to hold. An organisation can change how it works while keeping responsibility for goods, money and the promises made to the customer.

Even today you can establish where the offer creates the most work and which decisions are repeatable. Then change a selected part and check the result of the whole change, including costs arising outside the place that was automated. Only then do you know which tool to buy. Buying in the reverse order merely cements a process that has stopped paying for itself.