Description
Regenerate data-driven personas each period with an unchanged method and measure what share of the persona set persists, to detect when personas go stale.
Personas are usually built once and assumed stable. Jung, Salminen and Jansen fed monthly YouTube analytics for a publisher with over 580,000 subscribers (October 2016 to September 2018) into an identical NMF-based persona pipeline producing 15 personas per month. Overlap is shared personas divided by the earlier set; change is one minus overlap. Mean monthly change was 2.8 personas (18.3 per cent), yearly change 20 to 27 per cent, and 47 per cent over the whole period. Two months with six or seven changes matched a surge in channel popularity. The rigour lies in holding the method constant so that change reflects the audience, not the analyst, and in summarising the distribution of change.
Attributes
- Discipline
- Research Design & Rigor
- Loss type
- Latent
- Stage
- Post-launch, Definition
- Data
- Quantitative
- Effort
- Weeks
Steps
- Collect user data for each period from the same source.
- Generate the same number of personas per period with an identical, scripted method.
- Match each period's personas against earlier lists using explicit matching criteria.
- Compute overlap: month over month = shared(prior, current) / prior; year over year and first-to-last similarly; change = 1 - overlap.
- Summarise the distribution of change (mean, SD, median, range, counts of periods by number changed).
- Investigate spikes against events in the underlying data and set a refresh threshold for persona updates.
Sources
- Personas Changing Over Time (Jung et al., 2019)
Also known as
- Longitudinal persona change tracking with set-overlap metrics
- Persona drift monitoring
- Persona staleness analysis
- Month-over-month persona overlap
Categories
- Research Operations
- Research Design & Rigor