Douglas: AI cannot smell a fragrance. Remembering that helps

Fragrance has a language of its own. Something always unfolds, a trail becomes sophisticated, and the evening apparently cannot be forgotten. Lovely, of course. But sometimes a person just wants a scent they can wear in a lift without giving their colleagues an unexpected encounter with an entire spice market.

Douglas's fragrance guide discusses staff consultation and using fragrances a customer already likes as a starting point. That is a useful basis for training. My proposed AI role-play is an independent idea, not a description of a confirmed Douglas implementation.

Translate impressions into useful questions

The first skill is understanding context. Where will the fragrance be worn? What has the customer enjoyed before? What would they prefer to avoid? AI can play shoppers with different vocabularies, from someone who knows the notes to someone who simply says, 'Something quieter, please.'

The second mechanism is a deliberately short selection. After asking questions, the adviser offers two or three directions and explains their differences plainly. The trainer checks the reasoning against approved product information. It cannot evaluate a smell or claim to know how a fragrance will develop on an individual's skin.

A third exercise begins after the customer tries a sample and changes their mind. The employee practises not defending their recommendation like a university dissertation. In a real shop, the person's experience outranks an algorithmic match. AI can prepare an adviser for that conversation, but cannot replace the sensory experience.

A separate exercise should cover someone choosing for another person. Their own reaction is only part of the brief. The adviser asks about known preferences and leaves the unknowns unresolved, rather than pretending fragrance notes can mathematically establish somebody else's taste.

No personality diagnosis from a bottle

My three measures would be consultations with an identified use occasion, customer ratings of comparison clarity and the proportion of visitors who feel the decision remained their own. Commercial outcomes can be examined later, alongside repeat visits. They should not be invented from a polished demonstration.

I would prohibit the trainer from inferring preferences from age, appearance or presumed wealth. Expensive clothes do not establish a preference for heavy fragrance. Useful inputs are much less theatrical: current descriptions, tester availability, approved demonstration procedures and preferences the customer has actually chosen to share.

The pilot should cover a limited fragrance selection and a small adviser group. Assess before and after using equally demanding, but different, scenarios. A human reviewer checks whether the context was understood, differences were explained and the customer retained the option not to choose that day. Eloquence earns no automatic points.

This is a grown-up role for AI in sales training. It does not become a master perfumer. It helps an employee remove unnecessary monologue. The customer hears something more useful than an encyclopaedia of notes: a clear explanation of the options and why those options might belong in their life.