CASE STUDY 01ButterHow we workUnited States

Two of every five things we collected, we threw away. That is the work.

A premium international dairy brand asked us to explain the US butter market. We worked from what people had already done — what they typed into search engines, what they wrote in reviews after buying, and what they argued about in public. Deciding what did not belong took longer than finding what did.

32,795

reviews, posts and comments read and sorted — from 7 websites, in 2 languages

931

search terms checked against real monthly search counts before using them

3,192

customer reviews read across 27 competing butter brands

What happened to everything we gathered
Thrown out — wrong product, wrong country, repeats
Kept — real butter buyers, writing about butter

40% of everything we gathered was thrown out before any of it was analysed.

Three things we had, and did not report

01

A big number we would not use on its own

Two thirds of the Spanish-language search terms we found were one word that usually means a different food altogether. Rather than quietly leaving it in and reporting a bigger market, we reported the smaller figure without it, next to the bigger one, and said which was which.

02

A measure that read zero every time

One set of figures came back as zero on every single round of research. A measure that never moves is broken, not a discovery about shoppers. We told the client the tool had failed instead of publishing the zeros as a finding.

03

Anything resting on fewer than 30 items

Where a finding was based on fewer than 30 pieces of writing, we described which way it pointed and gave no percentage. Below about 30, one item added or removed moves the number by several points, so a precise-looking figure would be false precision.

The client got a shorter report with fewer numbers in it — and could act on all of them. Where the evidence simply did not exist, we said so and named the research that would answer the question, instead of producing a figure that looked confident and wasn’t.

EVIDENCE BASE

US butter market · five rounds of research · English and Spanish · 2022–2026. Everything here comes from what people searched for and wrote after buying, not from sales data. Their words are described, not quoted. Client name withheld.

URCHIN INTELLIGENCE · THE SCIENCE OF REVEALED REALITY

Start with a brief.

A finished question, a half-formed idea, or just a category you want to understand.

Talk to an analyst