H&M: teaching circular services without the green magic trick

Someone arrives with a bag of unwanted clothes. They want to do something useful and reclaim the part of their wardrobe currently serving as a museum of optimistic sizing. It is tempting to say, 'We will recycle everything.' The difficult word is everything. Good service should not begin with a promise the associate cannot verify.

Practise precision rather than enthusiasm

The first scenario asks what a customer can bring to this particular store. The learner must identify the market, location and current conditions, then check relevant exclusions. The AI should not supply one universal answer for a global network. Here, local rules matter more than an inspiring sentence about the planet.

A second customer asks what their old shirt will become. A competent answer separates possible processing routes from the eventual destination of that individual garment, which the associate may not know. The simulator flags absolute promises and asks for a clearer explanation. Confidence is useful until it impersonates a tracking system.

The third exercise involves declining an item that does not meet local conditions. The associate explains why, shows the applicable guidance and offers an available next step. AI can simulate disappointment; a manager evaluates whether the conversation remains respectful rather than becoming a competition over who cares most about sustainability.

Put an expiry date on the answer

The training library needs approved rules, local exceptions, partner information and permitted environmental claims. Each answer must carry its source and update date. Outdated documents should prevent definitive advice. Questions about measured programme outcomes go to the responsible team instead of being completed by a language model for rhetorical effect.

The first trial could cover three common questions in stores within one market. Short practice before a shift is more realistic than removing an entire team for a day. Anonymised difficult enquiries can refresh the exercises. There is no reason to retain identifiable customer recordings when a stripped-down scenario will do.

The measures are accuracy against local rules, repeat contacts caused by unclear explanations, and the frequency of unsupported environmental claims. Human reviewers assess a sample of conversations. Bags collected may be interesting operational information, but that number alone proves neither training quality nor environmental benefit.

H&M has a documented garment collection programme. My proposal uses that type of service as the basis for AI-supported staff learning. It is an independent operating idea, not a description of H&M's internal training. The objective is accurate explanation, not producing highly enthusiastic environmental speechmakers.

This is a version of AI in retail that makes language slightly less spectacular and considerably more trustworthy. Customers know the next step, colleagues can admit uncertainty, and the brand does not rely on a universal 'it will all be fine'. Sometimes the most useful technology simply helps us remove one unnecessary word.