The brief
A major beauty retailer was spending heavily on digital acquisition. The campaigns were well-designed. The creative was strong. The targeting was standard: demographic segments built from purchase history, loyalty data, and third-party audiences. Results were acceptable. Not exceptional. The cost per acquisition was rising quarter over quarter, and the marketing team suspected they were reaching the same customers repeatedly while missing others entirely.
They came to Intent with a specific question: could behavioural signals identify customers who were ready to buy, not just customers who had bought before?
The problem with demographic targeting
Demographics describe who someone is. They do not describe what someone wants right now. A 34-year-old woman in London who bought moisturiser six months ago might be in the market for foundation today. Or she might not. The demographic profile cannot tell the difference. So the retailer was spending to reach everyone in the segment, hoping that a fraction would convert.
This is the core inefficiency of traditional targeting. It optimises for resemblance to past buyers rather than proximity to the next purchase. It treats all members of a segment as equally likely to act, when the reality is that intent is distributed unevenly and changes constantly.
The behavioural signal approach
Intent deployed its on-device intelligence to identify behavioural patterns associated with purchase readiness in beauty retail. This was not simple browsing history. The model processed sequences of behaviour: app usage patterns, content engagement depth, time-of-day signals, cross-category browsing that indicated research mode versus decision mode.
The key distinction was between interest and readiness. A customer comparing foundation shades across three apps in one evening is exhibiting different intent than a customer who glanced at a single product page last week. Both would appear in a traditional segment. Only one is ready to buy.
All processing happened on-device. No personally identifiable information was exported. The retailer received anonymised intent signals, not identity data.
The test design
The retailer ran a controlled test over eight weeks. The control group received the existing demographic-targeted campaign. The test group received the same creative, the same offers, and the same budget allocation, but targeted using Intent behavioural signals instead of demographic segments.
The only variable was the audience selection method. Everything else was held constant. This was deliberate. The retailer wanted to isolate the impact of signal quality from creative quality.
Results
The behavioural signal group delivered 1.6x the click-through rate of the control. More significantly, it delivered 2.56x the conversion rate. The gap between click-through and conversion is the important number. It means the behavioural signals were not just finding people willing to click. They were finding people willing to buy.
Cost per acquisition dropped by more than 40 percent. The total campaign spend was identical. The return was materially higher.
Why the difference was so large
Demographic segments are static snapshots. They describe a person as they were when the data was collected. Behavioural signals are live. They describe what a person is doing now and what that behaviour predicts about what they will do next.
The beauty category is particularly sensitive to timing. Purchase cycles are irregular. A customer might buy foundation once a quarter and skincare monthly. There is no predictable cadence. The only reliable indicator of readiness is current behaviour, not past behaviour.
By reaching customers at the moment their behaviour indicated readiness, the retailer stopped paying to interrupt people who were not in the market. Every impression went to someone whose on-device signals suggested they were actively considering a purchase.
The privacy dimension
The retailer operates under GDPR and was increasingly concerned about consent rates and regulatory exposure from third-party data. The Intent approach eliminated both concerns. Because all processing happened on-device and no PII was transmitted, the legal and compliance risk dropped to near zero.
This mattered to the retailer beyond the legal dimension. Their brand positioning was built on trust and self-expression. Surveillance-based targeting was increasingly at odds with the brand values they communicated to customers. Behavioural intelligence allowed them to be effective without being invasive.
What this means for the category
Beauty retail is a category where relevance and timing matter more than reach. The customers are there. The demand exists. The problem is not awareness. It is precision. Brands that can identify the moment of readiness and respond to it will outperform brands that broadcast to segments and hope for the best.
The 2.56x conversion uplift is not a ceiling. It is what happens when you replace demographic guesswork with live behavioural signals on a single campaign. Compounded across a full media plan, the efficiency gains are substantially larger.