Behind the algorithm: How investigative journalists exposed Canada’s listeria oversight failure

Behind the algorithm: How investigative journalists exposed Canada’s listeria oversight failure‍

When a deadly listeria outbreak swept across Canada in 2024, the initial headlines focused on recalls and rising case counts. But for The Globe and Mail journalists Grant Robertson and Kathryn Blaze Baum, something didn't add up.

A closer look at the data revealed a chilling detail: the DNA strain responsible for multiple listeria deaths had appeared nearly a year earlier—raising urgent questions about how it was missed.

Their investigation exposed a systemic failure in federal oversight. By analyzing hundreds of pages of regulations and speaking with whistleblowers inside the Canadian Food Inspection Agency, they found that an “risk-assessment” algorithm was allowing facilities to go years without inspection.

Robertson and Baum joined Chris Arsenault to break down the realities of investigative and trauma-informed reporting: from building trust with grieving families to navigating confidential government sources.

Read Grant Robertson and Kathryn Blaze Baum's award-winning investigation here: https://www.theglobeandmail.com/canada/article-cfia-food-safety-algorithm-listeria-outbreak/

Chris Arsenault: How did you decide to investigate this story?

Grant Robertson: I was watching the news coverage of the Listeria outbreak of 2024 unfold and there was something that started to bother me as case numbers started to mount.

Outbreaks happen, but they usually get contained, and they don't spiral out of control, typically, to where people start dying. There's a bigger problem at play here, and the question is, what was it?

Around the time that the products were recalled, public health authorities had done DNA sequencing on the recalled products—the alternative milks, almonds, soy, coconut milk—to figure out where the Listeria outbreak came from.

What they found was quite remarkable. The strain responsible for a large number of people getting sick and dying was responsible for people getting sick almost a year earlier.

Chris Arsenault: How did you identify the victims?

Grant Robertson: Under privacy [rules]... the identities of the people affected by this were really unknown, even the three dead.

The way we reached out to the victims initially was one: you can scour social media to look for mentions about this and start to build a network of sources that way.

The second is going through emails, letters and phone calls to the newsroom that might have been missed, where people might have reached out … and wanted to talk. We were able to start source building that way.

But probably the most important one was I sat down and wrote a letter that was designed to be given to the people whose family members had died in this and sent that to lawyers who are representing people in food poisoning cases.

That letter was really a plea for them to talk, for the importance of Canada.

Chris Arsenault: How do you approach sources out of the blue who have faced the tragedy of a loved one dying, particularly when these are just regular people with minimal experience dealing with journalists?  

Kathryn Blaze Baum: Just being mindful when you're speaking to them that you're not … re-traumatizing them.

You need to balance the public interest of wanting to have a high-impact story that holds a government to account or a company to account, and lays bare the devastation, with the desire to also preserve that victim's well-being, their privacy and their agency.

It requires a lot of back and forth, a lot of texting, a lot of keeping them posted, not wanting to leave them in the dark.

Looking at the case of Sanaya, who's the woman who had a pregnancy loss at 18 weeks, I just wanted her to feel as though she had the power in the reporter-source dynamic. It was up to her whether to speak to us, to be on the record or on background, or not for attribution. It was up to her whether she wanted to be photographed.

Chris Arsenault: In terms of the documents with CFIA, the 400 pages you analyzed—are these Access to Information requests a different dataset? Tell me about the documentary evidence side of this?

Grant Robertson: We didn't have a year on this investigation; we had a matter of months.

You can't really rely on access to information requests in that case, because the system's really broken, it takes so long.

What I started with is, through past investigations, I've got sources within the federal government and those were the first people I reached out to—whether it was within Health Canada, CFIA, public health.

And those people… if you can provide them enough protection and enough assurance… they start to point you in the direction of documents to read… or documents they could leak.

They advised me to read the Food Safety Act. That was key, because what was clear was a lot of this problem was hiding in plain sight. A lot of the information about the algorithm that we had to prove and detail.

Chris Arsenault: Tell me about building up CFIA sources. How do you actually go about developing internal government sources for an investigation like this?

Grant Robertson: Source development is really sort of the boots on the ground type reporting.

These institutions are impenetrable, so you can't go in the front door. The people that work in government have their livelihoods to worry about. They know they can't speak out.

Whether it's through LinkedIn or developing sources on one story and going back to them and saying here's the topic I'm working on, do you know anybody who might talk to me … and giving people to vouch for you …  is one way to do it.  

Chris Arsenault: How did you actually understand and fact-check something as complex as the algorithm?

Grant Robertson: Initially, what you do is … you start going back-channel to anybody who might know about this … and you get them to explain it.

So, what it was was talking to inspectors off the record … saying, explain to me how this algorithm dictates your day, explain to me the plants you do inspect, and which ones you don't expect and how the algorithm influences that process.

And then we had to verify all of this with the CFIA.

In this case, the algorithm was called the Establishment-Based Risk Assessment Model.

And what we found … was that there were 16 key data inputs into the algorithm, and 12 of them were supplied by the companies themselves, and not verified by the CFIA.

Chris Arsenault: Why did you structure the story so the nut graph, or core contention of your investigation, comes after several hundred words?

Grant Robertson: With something of this depth and complexity … you need to basically do some scene setting.

You have to make the reader first care about the people, then the issue, and then you have to hit them with the reveal, and essentially the nut graph on what went wrong.

Because at this point, they would have been reading about the outbreak … and they might be tired of the issue.

So, the way we structured this story is to lead in with Kale Sampson's mom. Basically, introduce you to this person who was quite healthy and then within the span of about 2 days had a terrible turn of her health, ended up in the hospital, on life support, then died.

What you're doing is setting up a medical mystery there, and then you're gonna talk about why the plant wasn't inspected.

Then the first section ends off on the reveal of the story, which is… that the algorithm prevented inspectors from going in.

Chris Arsenault: What was the response after publication?

Kathryn Blaze Baum: The feedback we got [from the victims] was pretty swift and overwhelmingly supportive.

It really showed that they felt as though their story was captured, that it was going to be impactful, that they felt that they had been given a voice.

Grant Robertson: The other feedback that I think was important was an immediate response from the Health Minister's office. They had ordered the Inspector General to investigate the outbreak, but after our story, they ordered a focus on the algorithm itself.

When the Inspector General's report came out, it really validated the reporting. The Inspector General said yes, there's a problem with the algorithm … this needs to be fixed.

This interview has been edited for length and clarity.