Want to learn how to unlock your behavioral data with reverse ETL? 🚀 Get our free ebook to do more with your data today.

How Canva uses reverse ETL to personalize messages to 55M+ users

By using Census to send more data into marketing tools like Braze, Canva has now empowered its data analysts to run segmentation experiments faster, resulting in a more personalized, targeted experience for its 55 million users.

INDUSTRY

Design, SaaS, Product-Led Growth

HEADQUARTERS

Sydney, Australia

use cases

Granular segmentation
Email Personalization 
Ad retargeting

Sources

Snowflake

DESTINATIONS

Braze, Salesforce, Facebook Ads, Google Ads

Results
  1. Sending more data into Braze using Census has improved Canva’s segmentation, resulting in more personalized messaging to its 55M+ users.
  2. Marketing data analysts can put more complex, or ML based segmentation models, in production in a matter of minutes, rather than weeks or months. 
  3. Data engineers spend more time unlocking value, rather than spending weeks building and maintaining integrations.
"A data analyst looks at data, looks at dashboards, finds insights, and is then able to activate it using Census all in one step. It’s really, really powerful and I think we're going to get a lot of use out of it."
Cuong Duong, Data Analyst at Canva
"Census solves two problems for us. It removes the bottleneck to get integrations out of the warehouse. It also centralizes where we look for data going out...so it’s given us a lot more confidence."
Krishna Naidu, Data Engineer at Canva

The company

Canva is a visual communications platform with a mission to empower everyone in the world to design anything and publish anywhere. It has had explosive growth with its freemium design product, boasting over 55 million monthly active users, across 190 countries. With a freemium business model, driving engagement in the product is key to a product-led growth strategy. As a result, Canva has been laser focused on delivering a world class user experience and has invested heavily in developing a data driven culture.

The problem

A growing user base demands more personalization

With a user base of over 55 million, sending targeted and relevant messages is a key priority. Data analyst, Cuong Duong partners closely with growth marketing teams at Canva to understand how different emails are performing for different segments, and uses that data to make recommendations around questions like “What emails do we send to our users? What segments should we send them to? What time do we send them?”

As its user base continued to explode by the millions, a strong need emerged for Canva’s product and growth teams to optimize personalization and segmentation. “We realized we were sending users the same thing, and we really needed to start segmenting more,” says Cuong.

“That user data was only coming in, or it was coming in most easily, into the data warehouse. We just didn't have the link to be able to send that data into Braze.”
Cuong Duong, Data Analyst at Canva

More personalization requires more data in Braze

Canva’s growth marketing teams use Braze (a Census partner) as its messaging platform. To get to the next level with its personalization strategy, the growth teams identified two key needs:

1. They needed to send more data into Braze

Data analysts like Cuong wanted to be able to leverage additional user data to build segments in Braze, but this data wasn’t readily available or easy to get into Braze. “In order to collect this data we’d have to insert Braze tags into different parts of our products that we might not have wanted to,” explains Cuong. That data, however, already lived in the data warehouse. “Some of that user data was only coming in, or it was coming in most easily, into the data warehouse. We just didn't have the link to be able to send that data into Braze.”

2. They needed a way to build more complex segments and push those into Braze

It wasn’t just the availability of data that was hindering Canva’s personalization efforts. Data analysts like Cuong felt restricted by the limitations posed by marketing and CRM tools in terms of building audiences or segments. Although point and click interfaces are great for “creating really simple things quickly,” audiences can quickly become error prone as the segmentation logic grows in complexity. “As you get to the more complicated segments, things that may require a little bit of machine learning, or even decision trees with a few rules, the logic inside Braze becomes really complex,” explains Cuong. Without any reliable way to sync these segments into Braze, the growth teams were forced to abandon their more complex segmentation ideas.

The integration dilemma: to build or to buy

“Not only was it time-consuming for our stakeholders to wait for these integrations to be built in-house, but the maintenance was not the kind of work that data engineers and analytics engineers wanted to do”
Krishna Naidu, Data Engineer at Canva

In order to get more data into Braze, Canva’s data engineering team considered building out integrations to tools like Braze themselves and conducted a thorough build vs buy analysis. “We always try to do it in-house first, because we have the people. But what we found was that not only was it time-consuming for our stakeholders to wait for these integrations to be built in-house, but the maintenance was not the kind of work that data engineers and analytics engineers wanted to do,” explains Data Engineer, Krishna Naidu.

Cuong also recalls one of the data engineers sharing his findings that “it can take a lot of engineering effort and monitoring to maintain even something that appears simple, like sending data into the Braze API.”

“Since Census only syncs what needs to be synced - changes, additions, and deletions - the volume of data sent back to Braze is relatively low”
Cuong Duong, Data Analyst at Canva

Census supported all the destinations that Canva needed from day one. With Canva's large volumes of data, the data team appreciated Census’ incremental syncs since they no longer needed to worry about API rate limits. “Since Census only syncs what needs to be synced - changes, additions, and deletions - the volume of data sent back to Braze is relatively low,” says Cuong. This was also an economic benefit “since there’s often a cost involved with sending data back to the platform.” 

Canva also valued Census’ observability and alerting features. “Being able to see statistics in terms of how many records are being synced and how long it's taking. That kind of stuff is useful for us to know,” says Cuong. Features like that are difficult to maintain in-house and often get dismissed as ‘nice to haves’ with internal integrations.

The solutions

“Census has helped us make the messaging flow a lot nicer and more targeted. We’re now able to create segments, for example is this user a hardcore Presentations user or not, and then send them emails that are part of a specific Presentations campaign.”
Cuong Duong, Data Analyst at Canva

More data = more powerful segmentation = more relevant messages for users

For the initial use case, Cuong tried out Census to send key segments built in SQL inside Snowflake, back into Braze. “Our first task was to figure out whether Census can sync to Braze in a reasonable amount of time for 1% of our users. I was involved in logging into Census and trying to figure out how to connect to Braze. But it was all just extremely straightforward and intuitive. There were really good instructions on the website already.” 

Now data analysts like Cuong work with the growth teams to define segmentation models in SQL, turning the data warehouse into a source of truth for segmentation. Cuong describes this process, “now it’s just a matter of making the right view, going to Census and then clicking sync. Once it appears inside Braze, I then work with the growth marketer to define how we will test the effectiveness of that segmentation with an experiment.”

Cuong points to the benefits of using Census with Braze to optimize Canva’s messaging, “It’s helped us make the messaging flow a lot nicer and more targeted.  We’re now able to create segments, for example is this user a hardcore presentations user or not, and then send them emails that are part of a specific presentations campaign.”

"We actually already had an in-house solution for doing reverse ETL for Facebook and Google Ads. There was a Python script built by an analyst originally. No one was really managing it and no one really wanted to manage it. And so we thought "Oh, this is a prime use case for Census."
Cuong Duong, Data Analyst at Canva

Replacing old Python scripts for sending audiences to advertising platforms

As part of its personalization strategy, Canva is also sending curated audience lists to Facebook Ads and Google Ads for retargeting. After the success of the Braze integration, the data team saw other opportunities for leveraging Census’ automated reverse ETL to improve the processes and efficiency of the team. “We actually already had an in-house solution for doing reverse ETL for Facebook and Google Ads,” explains Cuong. “There was a Python script built by an analyst originally. No one was really managing it and no one really wanted to manage it. And so we thought "Oh, this is a prime use case for Census.”

The outcomes across the company

“The point and click interface for setting up a sync is a really big productivity boost. I think that really empowers and gives confidence to data analysts to just go in and send data to these platforms to test ideas.”
Cuong Duong, Data Analyst at Canva

Teams have become more self service with data

Since that initial period, usage of Census has really taken off with teams becoming self-service for their use cases. For example when Canva’s enterprise group wanted to send product usage data into Salesforce, they knew where to go. “We just noticed that Census was a very good general solution for these use cases where we needed to send information about customers back to the business platforms. And so we naturally just started to use that,” explains Cuong.

Census’ simple UI has been “a really big productivity boost,” he adds, “I think that really empowers and gives confidence to data analysts to just go in and send data to these platforms to test ideas.”

“Now we can immediately say “why don’t we just use Census to test the effectiveness of this model for targeting those ads?". And so it’s an avenue for us to test the business impacts of our insights more quickly”
Cuong Duong, Data Analyst at Canva

Enabling more growth experimentation

Census has also enabled the data team to run growth experiments, and put their machine learning (ML) models into use much faster. “We recently built a machine learning model to estimate the propensity that a user performs a certain conversion on the site. These models help us better understand and segment our users, but it can take a while before a model gets used inside the Canva product. Now we can immediately say, "We have these propensity scores for each user. And we’re also running Facebook ads at the moment. So why don't we just use Census to test the effectiveness of this model for targeting those ads?" And so it’s an avenue for us to test the business impacts of our insights more quickly,” explains Cuong.

“Census empowers data analysts to put their insights into action, and removes the need for dedicated engineering resources or time spent maintaining custom code.”
Cuong Duong, Data Analyst at Canva

Empowering data analysts to put their insights into action

Overall Census helps Canva’s data team to “activate their customer data more efficiently,“ says Cuong. “As we have matured, we have a lot of data analysts coming in who are strong at analysis and insights, and don't want to deal with all these API issues and custom integration scripts that break. Census empowers data analysts to put their insights into action, and removes the need for dedicated engineering resources or time spent maintaining custom code.”

The outcomes for the Data Engineering team

“The turnaround used to be in the order of weeks to months. Whereas with Census it takes minutes to hook something up.”
Krishna Naidu, Data Engineer at Canva

Less time building integrations = more time unlocking value

Krishna points to the time it saves his data engineering team as one of the main benefits, since “the turnaround used to be in the order of weeks to months. Whereas with Census it takes minutes to hook something up.” Data Engineers like Krisha can now spend more time unlocking value from data. Krishna shares his personal experience, “I used to spend time just working on integrations, not interacting with the business at all. I would get a brief that we need a Facebook connector, spend a few weeks building it, and then it just takes away time from improving the data experience of people within the organization, which is a huge focus now. Things like performance optimization of queries and organizing the data warehouse better.”

Centralized governance for data leaving the warehouse

In addition to enabling data engineers to focus on more value-adding work, Krishna describes the governance benefits to centralizing the integrations for data going out of the warehouse. “Census solves two problems for us. It removes the bottleneck to get integrations out of the warehouse. It also centralizes where we look for data going out. Now we can see all the integrations out of the warehouse, and we’ve put some process around that so that external parties can easily do an audit.  So it’s given us a lot more confidence.”

Enabling a whole new virtuous cycle of data

More importantly, using Census has unlocked a whole new data feedback loop within Canva. “Beforehand it was just getting data out and not getting much feedback. But now we have use cases where we get data out, and sometimes it's a two step process, from our warehouse to Salesforce, and then from Salesforce it goes somewhere else like Pardot. And then we get data back from Pardot and that enriches Salesforce. So the use cases are just getting more complicated, and there's more of a virtuous circle going on that we didn't have before,” says Krishna.

Looking ahead

In addition to continuing to experiment with segmentation, Cuong and other data analysts at Canva are also excited to scale operational analytics to as many areas as possible to further optimize operations, for example prioritizing their help tickets.

Finally the data team would like to explore going one step further and pushing data from the warehouse into their products, to be able to “personalize the product experience for users based on insights from the warehouse.”

From your data warehouse to your favorite apps. Without code or engineers.

Start using Census today by booking a demo with one of our experts.

Illustration of a data stack from multiple data sources.