Consent-Based In-App Data Sharing for AI Training
Data Collection
Client
An AI/ML training-data company
Objective
The objective was to enable AI model training using real, in-app behavioural and usage data, sourced directly from consenting respondents rather than synthetic or scraped sources.
Methodology
Respondents were recruited on the basis of having a minimum of 5 mobile apps installed with 2+ years of historic data, and at least 50% being apps of specific interest to the client — with the remainder non-client apps, preserving a natural usage context around the apps of interest.
Explicit respondent approval was captured on a per-app basis before any data was extracted — no app's data was accessed without individual consent.
Data extraction itself was run as a distinct downstream workstream, separate from recruitment and consent capture, allowing each stage to be quality-controlled independently.
Separate data extraction training sessions were held with the respondents.
Key Learnings
Blending client and non-client apps within the quota preserved natural, unbiased usage data rather than a client-app-only sample.
Treating data extraction as its own discrete process, separate from fieldwork and consent, made the pipeline easier to scale and audit.
This is an ongoing project in India, and under consideration for other countries including the USA and SEA countries.
Looking for real, consent-based behavioural data to train your AI models? Market Xcel can design and manage compliant, per-app consent-based data programmes — including expansion into new markets. Get in touch to discuss scope.
