Canva chose its generative AI tools by letting its employees try them all. It's an argument for consumption-based pricing.

Canva has been working in and around AI and machine learning over the last eight years while building out its flagship product, according to CTO Brendan Humphreys, but the company took a much more measured approach to adopting generative AI for internal purposes after ChatGPT hit the scene in late 2022. In the end, spending a little more money to give employees a chance to try a variety of tools seems to have done the trick.

"We've been playing with vendor tooling space for maybe two years, but we've really only reached an assessment that now is the time to roll these tools out at scale in the last six months," Humphreys said in a recent interview with Runtime. "The early versions of these tools were … you can see the power behind them but they were a little bit gimmicky and quite limited in their applicability."

Companies selling generative AI over the last two years really did themselves a disservice through comical applications of hype and FUD in their rollout strategies. But there was a kernel of truth beneath all the exuberance.

"Our assessment is that there is real disruptive change here that we can't ignore, and we are working very hard internally to help our engineers adopt AI as an accelerant, as a superpower that they can employ to increase their efficiency," Humphreys said.

Canva spent "perhaps more than I'd like to" in order to license similar types of tools from a number of different vendors, he said, and it took a long time before the company found most of those tools suitable for its needs.

Most of that investment went toward coding tools like GitHub Copilot and Cursor, but Canva also encouraged employees to use ChatGPT as a personal assistant.

The price is wrong

Canva's strategy touches on an important shift that is underway in SaaS: the introduction of consumption-based pricing. Traditional SaaS licenses charge customers based on the number of people that need access to the product, but that approach is starting to look outdated. Companies like Salesforce and ServiceNow are introducing consumption-based pricing as they release products based around agentic AI.

"I think that the market currently is saturated, and it seems there are new entrants every day. And I think that consumption-based pricing is a competitive advantage in that market," Humphreys said.

Canva has convinced some vendors using traditional per-seat licenses to give it some flexibility, but it sounds like companies that are struggling to sell generative AI tools the old-fashioned per-seat licensing way — such as Google and Microsoft — might find more success from a consumption-based approach as opposed to forcing generative AI features down every user's throat.

Shadow IT

In its early days, cloud computing was very much a grassroots phenomenon; engineers with credit cards were spinning up cloud servers to try out new features or side projects, and business departments found new SaaS tools much more flexible and user-friendly than older CRM or ERP suites. Right now inside a lot of companies, the generative AI experience has been almost completely the opposite thanks to management terrors about being left behind by another platform shift.

Canva is trying it both ways; strongly encouraging its 5,000 employees (a little less than half of whom are developers) to use generative AI tools, but giving them the space and time to figure out how those tools work best for them.

"We want people to be intrinsically motivated. We don't think, for this particular technology change, that extrinsic motivation works," Humphreys said.

Instead, the goal is to show employees what's possible with generative AI tools and let them experience that "oh, fuck!" moment on their own when they stumble upon a feature that makes them more productive, he said.

"I see my role as shepherding this technology into making sure that we have this positive cultural experience, and it really is a cultural shift for engineers," Humphreys said.

(This post originally appeared in the Runtime newsletter on Feb. 11th, sign up here to get more enterprise tech news three times a week.)