Case study: Canva harnesses the power of machine learning

Once Aussie design giant Canva implemented a data warehouse into their organisation, they were able to harness the power of machine learning.

Greg Roodt, head of data platforms at Canva said before using data warehouse platform Snowflake they were doing mostly offline batch predictions with three to four predictive models running.

“It was an enormous amount of effort, a lot of engineering and a lot of work. Fast forward to where we are now, we've deprecated, all of those old approaches running in batch in our data lake. We now have over 20 models in production,” Roodt said.

“We've gone from two or three clunky processes, lots of effort and tedious work, to now having a situation where everyone wants to try and models. Now we are having to ramp up that the machine learning side to add models wherever we can into the product.”

Roodt explained it has been a step change moving from offline batch to models trained on Snowflake and producing online inference models for the product.

“We're using it in all kinds of places, such as recommendations, search, relevance, and personalisation. Also things like lifetime value prediction, and very interesting use cases all over the place now, so that's exciting,” he said.

For the future, Canva is planning to continue its expansion of the Snowflake data warehouse into different areas in the business and continue to invest in machine learning.

“We're obviously going to continue investing in machine learning and advanced analytics capabilities in the product, that's still our main focus. But we're also looking to get adoption in as many groups as we can at Canva,” Roodt said.

“Infrastructure observability, perhaps next and we might also want to start looking at finance and strategy more. We're trying to expand use of snowflake.”

The company is also looking to expand their data warehouse into China.

“We are looking forward to expanding into China, as Canva has an installation in China. At the moment, we're not able to use Snowflake there, it would be something that would help us,” he said.