The Unstaffed Decision

Who picks the skincare when nobody's there to ask

Walk into a beauty retailer and count the moments where a human helps you choose. For a lot of shoppers, in a lot of stores, that number is zero. Even a strong consultant can only hold a limited number of real conversations per shift, and the store serves many times that number. Online, the ratio isn't lopsided, it's missing. Run it for your own floor. The figure is usually worse than you'd guess.

The decision still happens. It just happens unstaffed.

I'll call it that, because the industry doesn't have a name for it and you can't budget for something you can't name. An unstaffed decision is a purchase made without access to anything, a person or a system, that can translate what the shopper actually needs into a justified pick from what's actually available.

Two things that definition rules out, on purpose. It isn't a complaint about headcount. Staffing at that density was never coming back, and nobody's arguing it should. And it isn't solved by more information. A review, a TikTok, a friend with good skin can all be useful. None of them can match what this shopper needs against what this store actually carries today. They shape the decision without finishing it.

Plenty of skincare gets bought this way now. That's a strange place for a category to sit, because skincare is one of the few things where a wrong pick doesn't just disappoint you, it stays on your face for weeks.

What the industry optimized instead

Beauty retail spent the last decade getting extremely good at two things that both sit next to the decision without touching it.

The first is the shelf. Placement, planograms, packaging, the endcap, the way a serum photographs. All of it assumes the shopper arrives already knowing roughly what she wants, and the shelf's job is to steer a known intent.

The second is the ad, the whole apparatus for getting someone to show up with a brand name in her head.

Both work on the intent. Neither works on the decision. The gap between "I should probably do something about this" and "I'm buying this specific bottle" is where the money gets won or lost, and that gap is staffed by nobody.

You can see it in the returns, the abandoned baskets, the three-quarters-full bottles in every bathroom. Those aren't marketing failures. They're decision failures that marketing paid full price for.

Why the existing tools didn't close it

Diagnostics showed up to fill this. Skin analysis is a real technology and a crowded field. Haut.AI, Perfect Corp, ScanSkinAI and a dozen others do serious work on image-based assessment, and some of it is genuinely good.

But most of it stops one step short. It tells you what your face is doing. It doesn't tell you what to buy, from the assortment actually in front of you, right now.

That's not a small gap. It's the whole thing. A shopper who learns she has "moderate dehydration and uneven texture" isn't closer to a purchase. She's further from it, because now she has vocabulary and still no answer.

The second problem is ownership. When a diagnostic tool belongs to a brand, it recommends that brand. Shoppers work this out in about four seconds, and the recommendation loses whatever authority it had. A tool that can only ever arrive at one answer isn't recommending, it's packaging.

So the missing piece isn't better analysis. It's a decision layer: something that sits between the assessment and the assortment, stays neutral about which brand wins, and ends in an answer the shopper can act on by herself.

Neutral needs a definition too, or it's just a vendor saying nice things about itself. Neutral means products get ranked against the shopper's stated needs and the retailer's available assortment, with no commercial weighting toward a preferred brand. That's testable, and the test works on any vendor including mine: change the commercial relationship with a brand, hold the shopper's needs and the product data constant, and see whether the ranking moves. If it moves, the layer isn't neutral. The system shouldn't care which brand wins. The ranking should only care which product fits.

What shows up when the layer is there

I work on one of these, so treat what follows as a report from inside rather than a study. SKINBOT is a neutral decision layer running in beauty retail, and here's what early deployments show, with the limits attached, because numbers without limits are marketing.

At a six-location Moscow retailer, roughly four out of five people who start the flow finish it. That's high for any multi-step consumer flow, and it's high for a specific reason: the payoff at the end is an answer rather than a report. A multi-tenant deployment in Estonia shows the same broad pattern on completion.

One more observation, handled carefully. Recommendations concentrate into a much narrower set of products and brands than the visual breadth of the shelf suggests. That could reflect duplicated positioning across the assortment. It could equally reflect catalog composition, price constraints, the distribution of skin concerns among shoppers, or gaps in product data. There isn't enough evidence yet to separate those causes. But the observation itself is new information, because it can only be seen from a layer that sits across the whole assortment at the moment of choice. Diagnosis can't see it. Point of sale can't see it either, since it only records what got bought, never what should have.

Here's what none of this proves. These are early deployments, not a study. They're in Russian and Estonian language markets, not Western Europe or the US. They're independent and mid-market retail, not a flagship with heavy traffic and a staffed floor. And what's measured is completion, not conversion: transaction attribution isn't mature enough for me to publish a rate I'd defend. Six-month retention and return rates, the metrics that would actually settle the argument, aren't in hand at all.

So the honest version is narrow. Completion at that level says the flow is well built and that unstaffed shoppers will finish something that ends in an answer. It does not yet prove the category thesis. I'd rather hand you the small true thing now than the large claim I'd have to walk back.

The same decision, one step upstream

The unstaffed decision isn't confined to the aisle or the product page anymore. It's moving upstream, into the assistant conversation that happens before the retailer ever sees the shopper.

People now type "what should I use for this" into ChatGPT, Claude or Perplexity, and some of them never reach a store or a product page with an open mind about it. The consultant didn't get replaced by a kiosk. It got replaced by a system answering in a context the retailer can't see or measure.

Which means the decision layer has two surfaces. One faces the shopper in the store. The other faces the model, and it only works if the tool is legible to the systems doing the answering. That's a different discipline from SEO, and most of the category hasn't noticed it yet.

What this means if you run a category

Three things follow, and none of them require buying anything.

Count your unstaffed decisions. Take your traffic, take your realistic staffed-conversation capacity, subtract. That number is your exposure. Most retailers have never calculated it, and most are surprised by it.

Check whether your diagnostic ends in an answer. If your tool produces a skin report and then hands the shopper back to the shelf, you bought a diagnosis, not a decision layer. Those solve different problems. One produces information. The other produces a choice.

Ask who the recommendation is allowed to favor. If the answer is "us," your shoppers already worked that out, and the tool is doing less work than the invoice suggests.

Run those three on supplements, haircare, pharmacy or anything else with a wide catalog and a confused buyer, and you'll get the same reading. Skincare is just where the punishment for a wrong pick is most visible.

None of this is really a question about software. It's whether the shopper standing in front of the shelf, phone in hand and nobody to ask, gets a justified answer before she gives up or guesses.