AI Costs vs Human Labor: Canadian CIOs Debate ROI
Canadian CIOs are debating whether AI can be cheaper than human labor, citing challenges in estimating costs and measuring ROI. They share strategies for integrating AI, from token billing to automated scripts, and outline future plans to monetize usage. The discussion highlights the need for gover…
By Felo News Desk · Published
Three Canadian chief information officers—Cecilia Carbonelli of Haventree Bank, John Comacchio of Teknion, and Brigitte Larivière of Bombardier—recently convened to discuss a common dilemma: can artificial intelligence actually save money compared to traditional human staffing? The conversation, held as part of the 2026 Canadian CIO of the Year Awards, revealed that while AI promises speed and scale, the true cost and return on investment (ROI) remain elusive.
Why the Cost Question Matters
Carbonelli opened the discussion by pointing out a practical reality: people get sick, take vacation, and require ongoing training. “When you hire a human, you’re paying for their availability, not just their output,” she said. Yet, when she asked hyperscalers for a clear estimate of AI costs, the answers were vague. “Some of these hyperscalers still to this day can’t answer us on projected costs,” she noted, highlighting a gap between the promise of cloud‑based AI and the financial transparency needed by enterprises.
Comacchio, who runs technology at the Toronto‑based furniture maker Teknion, shared a similar frustration. He has integrated AI into the company’s existing software stack and tracks token usage to gauge spending. “I think it’s one of those necessary evils where we have to spend the money to learn, and we have to hit the wall to understand,” he said. Larivière, on the other hand, emphasized that the value of AI is often intangible—speeding up workflows rather than delivering a neat dollar figure.
Measuring Value Beyond Dollars
All three CIOs have AI tools in use today, but the metrics they track differ. Carbonelli wants an AI agent that can skim a 30‑page property valuation report in seconds—a task that typically takes a human analyst. She is building the underlying infrastructure, setting data controls, and monitoring the agent’s behavior before she can claim a cost benefit.
Comacchio’s approach is more pragmatic. He turned on AI wherever it fit within existing processes, trained staff, and then simply watched the token bill. “We have to spend the money to learn,” he reiterated, implying that the learning curve itself is part of the cost.
Larivière’s strategy is to treat AI like any other consumable—phones, for instance. She plans to roll out usage reports in 2027, followed by billing in 2028 or sooner if costs rise. “I want them to first see the bill, and after that pay for it,” she said. This model acknowledges that some users will generate far more AI usage than others, but the expectation is that the productivity gains will justify the expense.
From Scripts to AI‑Driven Automation
All three executives are moving beyond simple automation scripts toward AI‑augmented processes. Larivière’s team automated over 800 scripts in 2026, focusing on infrastructure tasks like patching and server reboots. The next step is to layer AI on top of those scripts, creating a fixed sequence of operations that can’t be improvised. “Automate everything you can, and then move it to AI,” she advised.
Carbonelli echoes this sentiment, arguing that not every task requires an agentic AI model. “Not everything needs an agentic AI agent,” she said. “Many tasks can be handled by a script someone wrote, and that costs less to run and less to watch.” This pragmatic view aligns with Comacchio’s observation that data governance and workflow mapping—often overlooked—constitute about 80% of an AI project’s effort.
Looking Ahead: 2027 and Beyond
All three CIOs agree that the real work—defining clear metrics, establishing governance, and proving ROI—will take place in 2027. Until then, they are focused on the foundational work that rarely receives public attention but is crucial for success. The upcoming award ceremony in Toronto on October 1 will highlight the winners, but the underlying conversations about AI cost, governance, and value will shape the industry for years to come.
In short, while AI offers undeniable speed and scalability, the challenge remains to translate those benefits into tangible, predictable financial outcomes. The Canadian CIOs’ discussion underscores the need for transparent cost models, robust governance, and a willingness to experiment before committing to large‑scale AI deployments.
Key facts
- AI cost estimates from hyperscalers are often unclear
- Human labor offers predictable availability but incurs ongoing costs
- Token billing provides a tangible measure of AI spending
- Governance and workflow mapping are critical for AI success
- Future plans include usage reporting and billing to justify ROI
Why it matters
Understanding the true cost and ROI of AI is essential for enterprises to make informed investment decisions and avoid hidden expenses.
Frequently asked questions
What is token billing in AI?
Token billing tracks the number of tokens—units of text processed—used by AI models, providing a direct measure of usage and cost.
Why do CIOs hesitate to fully adopt AI?
Uncertainty around cost, lack of clear ROI metrics, and the need for robust governance frameworks often slow full adoption.
How can companies measure AI ROI?
By tracking productivity gains, time saved, error reduction, and aligning usage with business outcomes, companies can build a case for AI investment.
Sources
- [1] digitaljournal.com — originally reported as “Humans might be cheaper than the robots after all”
