Case 001 · UK Ground Coffee

What the AI found

We simulated 256 coffee buyers across 16 behavioural dimensions. One dimension predicted Amazon's best sellers list. It pointed to a product combination that exists on Amazon — but isn't selling at volume. The question is whether the demand is real.

The Match

One behavioural dimension out of 16 correlates with what actually sells

We built four behavioural dimensions with four levels each — how people relate to coffee, how they shop, what drives their lifestyle, and what their food world looks like. We scored Amazon's top 16 ground coffees against each level and measured correlation with the actual best sellers ranking.

15 levels showed no meaningful correlation. One did.

Correlation with Amazon BSR rank · 16 behavioural levels

The dimension was Food Identity. The level that matched:

"I eat what's easy and familiar. I'm not adventurous with food. I know what I like, I stick with it, and I don't spend a lot of time thinking about it."

r = +0.479, p = 0.060. For this food identity, the AI's product preferences track the actual Amazon ranking.

The Signal

This group's top-scoring product: medium roast, organic, certified, value-priced

We scored every possible product combination against this group. Their highest scorer combines a medium roast, organic beans, Rainforest Alliance certification, and a value price point.

That exact combination doesn't appear anywhere in Amazon UK's top 50 ground coffees.

The closest best seller — by Amazon's own-brand Crema Med at BSR #3 — matches on three out of four. It's medium roast, Rainforest Alliance, and value-priced. But it's not organic.

The Price Map

On Amazon's best sellers list, organic and cheap don't overlap

We checked the price of every organic ground coffee and every by Amazon product in the BSR top 50.

Price per 100g · Amazon BSR Top 50 only

The cheapest organic coffee in the top 50 is £2.33/100g. The most expensive by Amazon product is £1.50/100g. No organic product in the best sellers list competes at the value price point.

The Check

But the product does exist on Amazon. It's just not selling.

When we searched Amazon UK beyond the best sellers list, we found products that match the combination the model identifies. They're on the platform. They're just not ranking.

On Amazon · Not in the top 50

Der-Franz Crema Organic · 4×500g£1.57/100g
Organic, Rainforest Alliance (UTZ), medium roastPage 1 search
Austrian brand. Sold by Amazon.Not in BSR top 50

At Tesco · Not on Amazon

Löfbergs Organic Medium Roast · 450g£1.11/100g
Organic, Rainforest Alliance, medium roastTesco & Ocado
Swedish brand. Entered UK 2024.Not on Amazon

Both products match the combination. Der-Franz is on Amazon at £1.57/100g — close to the by Amazon price range but not breaking into the best sellers. Löfbergs is at Tesco at £1.11/100g — cheaper than every by Amazon product — but not on Amazon at all.

The Question

The AI says there's demand. The market says it isn't converting. Which is right?

The model identifies a behavioural dimension that correlates with what sells on Amazon. That dimension's ideal product — value-priced, organic, certified, medium roast — has supply on the platform (Der-Franz) and off it (Löfbergs). But neither is selling at volume on Amazon.

There are two possible explanations:

1. The demand is real but the product has a discoverability or positioning problem. Der-Franz exists on Amazon but may lack the visibility, reviews, or price perception to compete with the by Amazon range. The product is right; the go-to-market isn't.

2. The AI model overpredicts demand for this combination. The simulation's behavioural profiles may overweight organic and certification signals relative to how real convenience buyers actually make decisions. The correlation with BSR is real (r = 0.479) but the ideal product it generates may be a simulation artefact.

This is a testable hypothesis, not a confirmed gap. The AI generated a specific, falsifiable prediction: that people with a convenience-oriented food identity will score organic, certified, value-priced ground coffee higher than what's currently in their basket. Fielding the study with real respondents will confirm or reject that prediction.

If real respondents confirm it, the opportunity is specific: either help Der-Franz break into the Amazon best sellers, or bring Löfbergs onto the platform at a price point that undercuts the by Amazon range.

If they don't, the model has identified a bias in how AI simulates purchase behaviour — and that's a finding worth documenting too.

How we got here →

1. Decomposed Amazon UK's top 16 ground coffees into four product dimensions with four options each — roast profile, bean type, trust credentials, value positioning.

2. Built four behavioural dimensions using Meta Ads Manager interest and behaviour parameters. Four levels each. 256 unique simulated profiles.

3. Simulated each profile with AI. Each evaluated 24 coffee descriptions on a 5-point purchase intent scale.

4. Scored all 16 Amazon products against each behavioural level. Correlated with BSR. One level — Food Identity: Convenience — was the strongest predictor (r = +0.479, p = 0.060).

5. Enumerated all 256 possible product combinations for this group. The top scorer — medium roast, organic, Rainforest Alliance, value price — was absent from the BSR top 50.

6. Searched Amazon beyond the top 50. Found Der-Franz Crema Organic — matching the combination at £1.57/100g. On Amazon but not ranking.

7. Searched the wider UK market. Found Löfbergs Organic Medium Roast at Tesco — £1.11/100g. Not on Amazon.

Next step: field the study with real respondents. If this behavioural dimension emerges in human data with the same product preferences, the hypothesis is validated. If it doesn't, the model has a measurable bias that needs correcting.

The full trail — how we got here

This is what discovery looks like.

A structured experiment that generated a specific, testable hypothesis about a real market. If you want this for your category, reach out.