Sobeys’ tech team is no longer the bottleneck

Rupinder Dhillon’s team scoped the work of plugging Sobeys into Uber, Instacart and DoorDash at about 10 months. The first partner went live in 10 weeks, and the time the team got back went into retail questions, including which products to list in the apps and whether the pricing was right.

Dhillon, vice-president of enterprise data and AI at Sobeys, told that story on stage at Snowflake World Tour in Toronto in September to make a point about where the holdup is now.

“We’re used to being the bottleneck,” said Dhillon.

When business teams wanted to move forward and technology lagged, the default assumption was that tech hadn’t delivered.

That has now flipped to the business, she said.

“Because the technology is so much easier to use and so empowering, now it really becomes about ‘Are our business teams ready for it?’” said Dhillon.

Part of what changed was the technology underneath. Sobeys runs its data on Snowflake, the company hosting the event, and when Snowflake’s Nav Anand asked how that had helped, Dhillon said the difference showed up in how her data engineers spend their time.

They used to spend it keeping things running, fixing the connections that move data between Sobeys’ systems and the systems themselves. Snowflake’s platform is an off-the-shelf product that stores and processes a company’s data, and Dhillon said it worked out of the box. Her engineers now spend that time on business problems.

When I covered Snowflake Summit in San Francisco in June, the new AI tools on show needed a company’s data pulled from scattered systems into one governed platform first. That’s work many Canadian organizations still have to do.

According to Dhillon, the delivery launch showed the rest of the company there was “something going on here that can drive a real fundamental shift in terms of how we work.”

The business has to be ready

Dhillon assigns each technical team to one part of the company, so her engineers can learn that area’s business problems in depth.

“I joke with our team all the time that we’re not a technology team. We’re retailers first,” said Dhillon.

Business teams and delivery teams now build side by side, she said.

“This isn’t the ‘give me your requirements, lob them over the fence, and I will build it,’” said Dhillon. “We’re really co-building together.”

Whether an AI tool makes it into daily use now depends on whether the team around it changes how the work gets done, she said.

“The technology productionization is not the challenge it used to be,” said Dhillon.

She wants the business on board from the start. Every big AI project at Sobeys has a business executive as its sponsor, a vice-president or senior vice-president from the part of the company that will use it, who backs the project and answers for it.

“The more [business teams] can understand it, the more that they can believe in it, the more they can see how their problems are going to get solved from the technology,” said Dhillon.

Dhillon called it a balance. Sobeys gives people “the sandbox to play,” she said, and its leaders decide which AI projects get funded based on which will show up on the bottom line. Without those decisions, she said, the AI bill just climbs.

Dhillon’s experience at Sobeys is one Snowflake sees across its thousands of customers, with the technology moving faster than the people around it.

The technical work is getting shorter

I sat down with Bala Kasiviswanathan, Snowflake’s vice-president of AI and developer experience, later that day in a small room down the hall from the main event, in the bowels of the Metro Toronto Convention Centre.

“Most of the challenge comes in just, I think, getting the people aligned,” said Kasiviswanathan.

Even with a strong leader who has a vision, he said, breaking down silos means changing ownership and getting people to come together around it. Sobeys’ co-build model, with business and delivery teams working side by side, is one way to do that.

“I think the technical times are compressing,” said Kasiviswanathan. “I think the human interaction time is something we just have to go with.”

He named three places where that human work can slow a project. The first is agreeing whether to build AI agents or buy them. The second is governance. He said companies need “a very well-thought-out process about governance in this new day and age,” and some are still applying rules “coming from the traditional database world or SaaS world in the past.” The third is an experimentation culture that lets teams “do things in a smaller way” and move faster, which is what Dhillon’s sandbox is for.

“If those three things work very well, I think things move faster,” said Kasiviswanathan.

His governance point fits with what Caitlin Halferty, head of data and analytics at Thomson Reuters, told me this summer, that governing data early let her company move faster on AI.

Speed depends on the organization

In the 10 days before we spoke, Kasiviswanathan had met two financial services customers at opposite ends of the adoption spectrum.

One bank wanted to move everything onto Snowflake’s data platform before trying AI. Another financial customer wanted to go “very aggressive” and give its AI agents context from across the company.

He said plenty of customers sit somewhere in between.

Regulated industries can be slower by nature, he added, and the tight controls over data that drew them to Snowflake are part of the reason.

“We are always the trusted governed platform,” said Kasiviswanathan. “So the strength is also sometimes the slowness that comes in, because these industries chose us because that’s what they wanted.”

When I pointed out that both examples were financial, he took the point.

“I think you’re right to observe that,” said Kasiviswanathan. “I do not want to do the cliche like financial industry is always slow, or public sector is always slow.”

Speed may come down to “the leader and the culture and the DNA of the organization,” he said, along with how willing a company is to change the way it works.

He called Canadian customers “a mixed bag” and held up the Sobeys story from the keynote as an example from the fast end.

“There are customers who are doing almost as innovative, or sometimes even more innovative than what I see in the U.S.,” said Kasiviswanathan.

I heard two different reads on Canada that day. On stage that morning, Dhillon, whose team got its first delivery partner live in 10 weeks, had pointed to the numbers that put Canada behind its peers.

“As Canadian leaders, I also hear the stats of how Canada is lagging behind a lot of the adoption and innovation versus other peer countries. And so I would also challenge this room to think bigger, and to think bolder,” said Dhillon.

She asked companies to use AI to challenge their business models as well as the processes they already run. Her own 10 weeks suggest the technology is ready for that, which leaves Canadian leaders with the question she raised earlier: are the business teams ready?

Final shots

• An organization moves at the speed of its decisions. Someone still has to decide what to build, how the work changes and who owns the result.
• When technical work shrinks from months to weeks, business readiness sets the timeline.
• Canada’s adoption question may come down to whether leaders are prepared to change how their companies work.