Information technology / Design software
Canva: Producing year end content for each individual user
- Company
- Canva
- Country
- Australia
- Adoption stage
- In operation
- Source published
- In use from
- Date basis
- The date the source was published. It can differ from the date adoption started.
- How the source was checked
- Read the full source text
The work problem
Canva ran a year in review campaign called DesignDNA in December 2024 that delivered different content to each user. The goal was to read the designs a user made over the year and assign them a personality, but personal designs could not be opened for privacy reasons. The personality therefore had to be worked out from the metadata of the public templates a user had used. Producing writing and images of the same quality by hand across many languages worldwide was also impossible because of the volume.
Technology and data
The company matched users to one of the 7 design trends in its own 2025 Design Trends. Keyword matching on public template metadata covered 95% of the users in the target cohort, and the remaining 5% were handled by iterating on the same keyword method with generative AI expanding the keywords. The community was divided into 10 audience groups based on the most frequent themes in the templates they had used. For each segment pairing a trend with an audience group, Magic Write produced a personality name and description, with AI translation added per locale. Hero images were produced with Dream Lab. Combinations of locale plus the top 3 styles came to just over a million, so the poems were generated at scale by generative AI. For non English locales, the Localisation team reviewed a sample of the poems and the flagged ones were regenerated in repeated cycles.
Results
The company said 95 million unique DesignDNAs were created. It said that without opening personal designs it was able to match 95% of the users in the target cohort to a trend using public template metadata alone. It said writing and images were filled in for more than a million combinations of locale and style, which could not have been produced by hand. It wrote that the quality of the non English writing was checked through a sample review by the Localisation team.
Limits and open questions
Business results such as campaign participation or retention effects are not given. The share of generated writing that the localisation review flagged is also not stated.
Sources
- Behind the scenes of Canva's DesignDNA campaigncanva.dev, Accessed
Compiled from public sources. These are not results from ATF Works customers.