Data listing for Snowflake and Databricks

GLP-1 Journey Personas

Individual-level GLP-1 behavioural profiles, built from what consented participants searched, browsed, asked an AI assistant, and bought. One row per person, with treatment state, adoption barriers and journey timing derived from observed activity rather than claims data.

194
Features per profile: 6
Treatment states: GB, US
Markets: Quarterly
Refreshed: 12 months
Window

Request access Request a sample

What this is

The pre-prescription window

The value is in what clinical data cannot see: the cost research, dosing questions, side-effect worries and competitor comparisons that happen months before anything reaches a pharmacy record.

Why it is different

Auditable, not a black box

One inference sits on top of the observed topics, and a companion column names the rule that produced it, so every label can be checked.

Barriers people raise themselves

Salience of cost, side-effects, needles, stigma and supply, measured from what a person brings up rather than what an assistant told them.

Timing, in days

Days from first signal to first cost question, first dosing discussion, and naming a specific drug.

What is in each profile

194 features, grouped by prefix

Treatment: Six-state treatment_state with treatment_state_basis, plus treatment_days_to_first_cost_signal, _dosing and _named_drug.
Barriers and topics: Seven barrier salience scores and thirteen observed conversation themes as flags and month counts.
Brand and drug: brand_drug_primary, injectable and oral mentions, and brand_ai_introduced_brand_flag for a brand the assistant raised first.
Coverage: coverage_cohort_depth for scoping to aware, engaged or on-drug, plus a relevance_cf confidence score.
Commerce and context: Amazon purchase behaviour, fourteen app-category flags, and month-over-month trend and seasonal blocks.

How teams use it

What it supports

Demand-stage sizing

Split the cohort into researching, cost-seeking, treatment-detail and confirmed groups.

Barrier quantification

Rank what people actually raise, by stage and market, as a messaging input.

Journey timing

Decide when in the journey a message should land.

Adjacent-category demand

See what else these people buy, from protein and meal replacement to skincare.

Sample query

Barrier salience by demographic and treatment state

SELECT
    identity_age_group,
    treatment_state,
    COUNT(DISTINCT user_id)                     AS profiles,
    ROUND(AVG(barriers_cost), 3)                AS cost_salience,
    ROUND(AVG(barriers_side_effect_fear), 3)    AS side_effect_salience,
    ROUND(AVG(trajectory_at_risk_score), 3)     AS avg_at_risk
FROM glp1_personas_v4
WHERE coverage_cohort_depth <> 'aware'
GROUP BY identity_age_group, treatment_state
HAVING COUNT(DISTINCT user_id) >= 20
ORDER BY profiles DESC;

Scoping away the `aware` tier drops one-search passers-by, without which every rate is diluted.

Specifications

What ships

Dataset	`glp1_personas_v4`
Grain	One row per person per market, unique on `user_id`. A user-month panel is available for trajectory work.
Field count	194 columns across 20 prefixed blocks
Surfaces	ChatGPT, Gemini, web browsing, search, Amazon purchases and Amazon search, plus a mobile app panel.
Demographics	Age band, gender, education, household income and device, on each row and effectively complete.
Geography	United Kingdom, United States. Identical schema, so the markets stack or compare.
Window	12 months, June 2025 to May 2026
Refresh cadence	Quarterly
Format	Native tables on Snowflake and Databricks
PII	None. No names, emails or device identifiers.

Privacy & provenance

Aggregated features only

Every column in this dataset is a derived, aggregated behavioural feature: a state, a score, a rate, a count. Raw event rows are never exposed and are not part of what ships. All data comes from participants who opted in through the Measure app and browser extension. Participants are compensated, are told what is collected, and can withdraw at any time.

No personal identifiers are included: no names, emails or device IDs.
`user_id` is a pseudonym. The mapping that could resolve it to a person is internal and is not shared.
Treatment state is an inference from text, not a medical fact. No clinical or diagnostic use is supported.
Handled in line with GDPR, CCPA and applicable data protection law.

Request access

Talk to us about scope, sample data and licensing.

Snowflake Marketplace, Databricks Marketplace

Get in touch