"If Your AI Never Says Good Catch, You're Not Challenging It Enough"
IGM Financial's Corby Fine on content supply chains, the compliance lesson learned the hard way, and why human oversight isn't a constraint on AI ambition.
Corby Fine, VP of Digital Marketing and Performance at IGM Financial, came to MACH X Toronto with a framework, a set of hard-won lessons, and a line that stopped the room: if your AI never tells you you've caught something it missed, you're probably not challenging it enough.
It's a deceptively simple diagnostic for something genuinely difficult to manage, the risk that AI adoption becomes, in Fine's words, a race to the bottom. That teams become yes people, approving outputs without scrutiny, and quietly lose sight of the judgment that made the work valuable in the first place. For Fine, avoiding that outcome has been as central to the AI program he's built as any of the efficiency gains it's produced.
The Bottleneck Nobody Was Talking About
Fine's entry point into AI wasn't a technology decision. It was a process observation: the content supply chain—the full journey from brief to production to compliance review to translation to publishing—was proving to be a challenge for marketing, and generative AI was a direct solution to a very specific bottleneck.
"The simplest way to think about a content supply chain is how an organization gets materials to its audience," he says. "That includes everything from briefing, creative production, compliance and legal review, to translation and publishing. For marketing teams, this is core work: getting content out the door."
The advantage IGM had was that many of the tools they already had in place were embedding AI capabilities. That meant the earliest experiments didn't require significant new procurement or legal review cycles.
"We didn't have to spend millions of dollars right out of the gate," Fine says. It was a low-barrier entry point, and it gave the team room to learn before committing to larger infrastructure decisions.
Translation emerged early as the clearest near-term efficiency gain. It’s repeatable, stable, and easy to train a model on without requiring it to think in genuinely novel ways— principles that extend beyond translation. Plus, the market was already trending in that direction.
"If your vendors are already automating that work themselves, that's probably a strong signal you can too," Fine says. His recommendation is to start with what doesn't change often, prove the value there, and expand from a position of demonstrated credibility rather than aspiration.
A Governance Lesson Learned the Hard Way
The most candid moment in Fine's MACH X session came when his team created a solution, but only brought it to compliance for review after it was built. In retrospect, the result was predictable: compliance identified issues that could have been designed around from the start, and the cost of retrofitting was significantly higher than the cost of involving them earlier would have been.
"Map the full process first," Fine says. "Find all the touchpoints and handoffs—especially in regulated industries—and bring in the people who can address the gaps before you've built something they have to unpick."
The stakes of getting this right in financial services are significant, but so is the upside when it works. At IGM, bringing compliance into the process earlier revealed that up to 75% of what was being sent to them for review could potentially be automated or eliminated from their queue entirely. Not because the compliance function was being bypassed, but because the process had been mapped carefully enough to know with confidence what didn't need a compliance gate in the first place.
"It frees compliance teams to focus on strategic work instead of reviewing routine content," Fine says. The observation connects to a pattern that surfaced across the conversations at MACH X: the organizations making the most progress on AI aren't necessarily the ones moving fastest, but the ones who did the process mapping work that’s the prerequisite for knowing which inefficiencies are actually worth automating.
Human in the Loop Is Not a Limitation
Fine's position on human oversight is unambiguous, and it's informed by something more than regulatory necessity. In regulated industries, a human in the loop isn't optional. But Fine's argument is that it shouldn't be optional anywhere, not because AI can't be trusted, but because uncritical acceptance of AI outputs is its own kind of failure.
"If we become yes people and just approve the AI output, we lose sight of the real value," he says. "It becomes a race to the bottom."
The practical test he uses is direct: how often does your AI agent tell you that you've caught something it missed? If the answer is never, the problem isn't the AI, it's the way the team is engaging with it. Real oversight requires real engagement, and real engagement requires teams that feel empowered to push back, question outputs, and bring judgment to bear on what the model produces.
This is where the interaction model—reimagining how roles and team composition work alongside AI—turns out to be harder than either the technology or the governance.
Fine flagged it at MACH X as unexpectedly difficult, and it's not hard to see why. Getting a model to produce content is relatively straightforward. Getting an organization to engage with that content critically, at speed, without either rubber-stamping it or reverting to doing everything manually? That's a change management challenge of a different order. This kind of organizational transformation is the work that sits underneath the technical enablement, and it doesn't resolve itself because the tooling improves.
What the Role Actually Becomes
The dream Fine described in his talk—one person able to ideate, produce, and publish—is closer to fruition than it's ever been. But the reality of what that person's job looks like has altered in ways that aren't always visible in the efficiency narrative.
"The expert's role moves from pure creation to more management and oversight," he says. "You're training the models, adapting to algorithm changes, staying on top of what gets ranked and featured. It's less about making the thing and more about making sure the thing stays good."
This transition is happening fast— faster, Fine acknowledges, than most organizations are prepared for. The industrial revolution was arguably a bigger disruption, but it unfolded across generations. This one is compressing similar levels of change into years, sometimes months. The organizations that navigate it well won't be the ones that automate the most, or the fastest. They'll be the ones that figured out what judgment is actually for, and made sure it stayed in the room.
Leigh Bryant
Editorial Director, Composable.com
Leigh Bryant is a seasoned content and brand strategist with over a decade of experience in digital storytelling. Starting in retail before shifting to the technology space, she has spent the past ten years crafting compelling narratives as a writer, editor, and strategist.