Canva to save 30,000 work hours with agentic AI

Australia’s homegrown graphic design platform Canva is making extensive use of workflow automation tool Workato and rolling out artificial intelligence (AI) capabilities that are expected to save thousands of hours of work.

Canva began using Workato around four years ago to automate business processes, said Michael Denari, Canva’s head of IT, who had previous experience with the tool at a former employer.

Workato is an easy-to-use, low-code/no-code product, he said, and Canva has since created several hundred “recipes” – automated workflows that connect applications – and has trained hundreds of its staff, with around 500 achieving a baseline certification.

The goal is for the IT group to maintain the Workato platform, while empowering end-users to build their own automation projects. The company is about 70% of the way there, with its human resources, finance and sales departments leading the adoption.

Denari said Canva will continue to upskill its staff, noting that the entire workforce recently dedicated a week to learning about and experimenting with AI.

Workato now allows the use of large language model (LLM)-powered steps in integrations, for example, to assemble information from thousands of social media posts. This has been “a real accelerator for us”, said Denari, as it has broadened the use of automation beyond core business processes into general experiments and one-off projects.

More recently, Workato has introduced agentic AI features, which are AI tools that can act semi-autonomously on a user’s behalf. For example, one agent monitors Slack channels for questions about expense reimbursements. The agent attempts to answer the query, which it currently does successfully about 60% of the time. If it cannot, it automatically generates a support ticket in Jira.

Another is a call-preparation assistant for its go-to-market team. It sends a Slack message with relevant information and suggested questions to the employee making the call. Afterwards, it updates the customer’s record in Salesforce based on the conversation. Canva plans to extend this assistant to support its customer success team.

Governance and controls are important considerations when implementing AI-powered functions, and Canva has already invested in that area. Model context protocol (MCP) is currently a big thing and offers some value, Denari observed, but Workato lets users build a customised MCP server that incorporates a knowledge graph and enterprise governance.

The agentic AI projects are expected to save 30,000 person-hours in 2025, with an annualised return of A$1m over the second half of the year. Canva is not limiting the use of AI to projects with high expected returns, as it wants to allow experiments to get the most value from the technology.

The company is aggressively pushing AI adoption. All staff have access to Google’s Gemini, while ChatGPT and other tools are available on request without needing approval. The hope is that AI, especially agentic AI, becomes part of everyone’s mindset, Denari said.

Workato’s original focus was on integrating applications to support workflows, but the goal now is to help customers get the fastest returns on their AI investments, said Bhaskar Roy, the company’s chief of AI product and solutions.

With connections to more than 10,000 applications and systems, Workato’s low-code agents, which are driven by key performance indicators (KPIs), “should be able to orchestrate across everything”, he said.

Prepackaged agents for various lines of business, including sales, HR, finance and IT support, speed up deployment, and can go live with appropriate security and governance controls in as little as three weeks, even where customisation is needed. “We’ve tried to make it really easy,” said Roy.

But he warned that customers who adopt point solutions for agentic projects will face agent sprawl, which he said is going to be 10 times worse than software-as-a-service (SaaS) sprawl.

That doesn’t mean Workato can’t help. Workato MCP allows customers to build agents on other platforms such as Salesforce’s Agentforce and then integrate them with Workato. It takes four clicks to build a Workato MCP server if none of the prebuilt ones are suitable, Roy claimed, adding that “all of our agents can be made into MCP servers”.

For teams unsure where to start with AI despite a company directive, Roy suggested identifying five or six key initiatives where an agent could help, even in a small way. Then, design and build an agent, iterating to the point where it helps achieve one or more KPIs. Management will then notice the success, which should lead to approval of further projects. Organisations are tired of experiments and want to see returns, he said.

While existing agents may seem impressive, Roy believes there is much more they could do. Using customer support agents as an example, he said they are currently little more than chatbots as they cannot see into operational systems such as enterprise resource planning (ERP).

If they gained a deeper understanding of the underlying systems, agents could proactively detect issues and take action. For instance, if a shipment is delayed, an AI agent could ask an affected customer to approve a switch in delivery method to ensure the purchase arrives on time.

“There’s still a long way to go,” Roy concluded.