Canadian organizations are leading the charge in AI adoption, with IDC forecasting 8.1% IT spending growth in 2025 and 46% of enterprises prioritizing AI initiatives. But there’s a critical gap between AI experimentation and actual business value; one that became clear during a recent partner panel at AWS Summit Toronto.
At the summit, three industry leaders shared insights from the front lines of enterprise AI implementation: Shannon Katschilo, Country Manager at Snowflake; Nitika Sharma, Director at CGI; and Alex Dassa, VP of Strategy, Transformation and Business Value at Coveo. Their conversation revealed something remarkable: Canadian businesses are achieving a 43% return on their AI investments, slightly above the global average. How? By focusing on practical outcomes over flashy demos.
The panel’s central message was clear: the organizations succeeding with AI aren’t those with the most projects, but those building sustainable programs that deliver measurable business impact. Here’s what we learned about moving from pilot to production in the real world.
The Implementation Reality Check
“While anyone can build a prototype in-house and query an LLM,” Dassa explained during the panel, “the real challenges emerge when you need to ground responses in enterprise data, both unstructured and structured, and manage permissions and security. That’s where many organizations get blocked by IT.”
This insight cuts to the heart of why so many AI initiatives stall after the pilot phase. The technical barriers—data grounding, permissions, security, and IT approval processes—aren’t afterthoughts. They’re the core challenges that separate successful implementations from abandoned experiments.
Dassa’s team at Coveo has developed a methodology specifically designed to navigate these challenges. Rather than building pilots in isolation, they start with value creation targets in mind, measuring answer accuracy and rates internally before progressively rolling out to customer traffic segments. The key metrics that matter aren’t technical benchmarks, but business outcomes: self-service success rates, case deflection numbers, and actual cost reduction.
This approach is delivering tangible results for Canadian enterprises. Katschilo noted that according to Snowflake’s research with Enterprise Strategy Group, Canadian businesses are seeing 43% return on AI investments—a figure that reflects not just successful pilots, but scaled implementations driving real business value.
The proof is in the customer outcomes Dassa shared from the panel:
- SAP Concur achieved $8.6 million in annual cost-to-serve reduction with a 31% decrease in case volume
- A global bank saw 99.4% search engagement with a 4.4X answer growth rate
- FleetPride experienced a 25% conversion rate driving 21% revenue growth year-over-year
These aren’t proof-of-concept results—they’re production-scale implementations delivering measurable business impact.
Relevant reading: See more Coveo customer success stories
The Technical Evolution: Beyond the Hype
One of the most compelling stories from the panel was Dassa’s account of Coveo’s strategic migration from GPT to Amazon Nova Lite on Bedrock—a move that exemplifies the kind of technical evolution required for enterprise-grade AI.
“We’re currently making a significant move by migrating our generative answering backend from GPT to Nova Lite on Amazon Bedrock,” Dassa explained. “This is a key milestone in making our platform fully powered by AWS.”
The migration provides customers with:
- Over 30% decrease in hallucinations
- Higher stability and improved data residency
- More LLM options via Bedrock
- Improved answer quality while maintaining response rates
But perhaps more importantly, this migration illustrates a crucial insight Dassa shared: “AI adoption isn’t about single tools; it’s about integrated stacks, outcomes, and trusted partnerships.”
This ecosystem approach is what enables organizations to move beyond pilots. Coveo’s Passage Retrieval API, for example, grounds Bedrock Agents in enterprise knowledge and permission structures, solving the data access and security challenges that derail so many AI projects. As Bedrock continues to evolve with features like AgentCore, these integrated partnerships will become even more critical for organizations looking to deploy AI at scale.
The partnership between AWS and ISVs like Coveo enables customers to access enterprise-grade deployments without having to build everything from scratch. As Dassa noted, “We’ve built our solution such that agents can access and act on the same trusted knowledge and signals—without needing to be individually integrated or grounded into every system.”
Industry-Specific Applications: Meeting Organizations Where They Are
The panel made clear that successful AI implementation isn’t about one-size-fits-all solutions. Different industries face different challenges, and the most successful approaches are those that address specific sector needs.
Dassa outlined how this plays out across key verticals:
- In Finance: Implementing AI that is permission-aware, compliant, and secure
- In Healthcare: Developing HIPAA-compliant navigation agents that improve patient experiences
- In Manufacturing and Software: Creating agent-driven troubleshooting solutions for productivity gains
This industry-specific approach was reinforced by Sharma’s perspective on modernization in financial services. “Financial institutions in Canada are essentially tech companies,” she noted, “and modernization isn’t just about upgrading code—it’s about enabling agility and regulatory alignment at scale.”
Sharma’s team at CGI doesn’t come in with predetermined solutions. Instead, they meet clients where they are and bring bespoke solutions that leverage best practices from both CGI and AWS. A recent example with a Canadian financial institution demonstrates this approach perfectly.
The client wanted to increase productivity in their customer-centric operations using two advanced tools they already had: AWS Connect as their cloud contact center and ServiceNow for ticketing. Rather than replacing these systems, CGI helped layer an AI auto-resolution system for repetitive issues and simple service requests, orchestrating across AWS and ServiceNow while ensuring alignment with the client’s regulatory framework and adoption at scale.
This approach aligns with Katschilo’s emphasis on building the right data foundation. “AI is only as good as the data it uses,” she explained during the panel. “Our platform unifies all of an organization’s structured and unstructured data to provide the essential context for accurate AI.”
The combination of industry-specific expertise, existing system integration, and proper data governance creates the foundation for sustainable AI programs that deliver long-term value.
The Road Ahead: Three Predictions for 2025-2026
The panel concluded with each expert sharing their prediction for how the Canadian AI landscape will evolve over the next 18 months. Together, these predictions paint a picture of an AI market moving toward maturity and practical impact.
Organizational Change: Sharma focused on the structural shifts coming to Canadian enterprises.
“The next 18 months will fundamentally change how Canadian organizations operate. We’ll see AI shift from being a departmental initiative to becoming embedded in corporate governance structures.”
This means new roles, new organizational models, and new ways of measuring success. With the Canadian Sovereign AI Compute Strategy rolling out, organizations need to build their governance and compliance frameworks now, focusing on building the organizational muscle to run AI at scale.
Implementation Maturity: Dassa predicted a sharp divide emerging in the market.
“The next 18 months will be about moving from ‘AI everywhere’ to ‘AI that works.’ We’ll see a sharp divide between organizations that deployed quick AI solutions versus those that built sustainable AI programs.”
The winners won’t be those with the most AI projects, but those who focused on specific, measurable business outcomes. She expects Canadian enterprises to lead in areas like regulated AI implementation and responsible AI scaling.
Data Democratization: Katschilo sees the next phase as being about expanding AI access across organizations.
“While today we’re focused on data scientists and technical teams using AI, by 2026 we’ll see AI capabilities in the hands of business users across every department.”
However, this will only work if organizations invest in what she calls the “data foundation”: governance, quality, and accessibility. Canadian organizations that get this foundation right will see AI driving decisions at every level, from the C-suite to front-line workers.
The Canadian AI Advantage
The insights from this AWS Summit Toronto panel reveal why Canadian organizations are outperforming the global average in AI ROI. It’s not about having the newest technology or the biggest budgets; it’s about taking a systematic approach that combines solid data foundations, practical implementation strategies, and organizational change management.
The three-pillar approach that emerged from the panel discussion — data foundation (Katschilo), practical implementation (Dassa), and organizational change (Sharma) — provides a roadmap for enterprises looking to move beyond AI experimentation. Canadian organizations have an 18-month window to build sustainable AI programs that will define their competitive advantage for years to come.
The key is focusing on measurable business outcomes rather than technological novelty. As the panel made clear, the future belongs to organizations that can build AI solutions that are not just innovative, but trustworthy and business-critical. Canadian enterprises are well-positioned to lead this transition, but only if they invest in the foundational elements — governance, integration, and organizational capability — that enable AI to work at scale.
Ready to Move Beyond AI Pilots? Get the Enterprise RAG Foundation Right
Dassa highlighted a key difference between AI leaders and laggards: “While anyone can build a prototype and query an LLM, the real challenges emerge when you need to ground responses in enterprise data and manage permissions and security.”
This is exactly why Coveo achieved over 30% reduction in hallucinations when migrating to Amazon Bedrock—and why their customers like SAP are seeing $8.6M in cost savings rather than just impressive demos.
Download our guide: “Putting the ‘R’ in RAG: How Advanced Data Retrieval Turns GenAI Into Enterprise Ready Experiences
Learn how to avoid the pilot-to-production gap by building AI on a foundation that handles:
- Complex enterprise data across dispersed systems
- Document-level permissions for secure, compliant AI
- Real-time accuracy that eliminates costly hallucinations
- Continuous learning from user interactions for better results
Don’t let your AI initiative stall at the pilot phase. Get the enterprise-grade retrieval framework that transforms promising prototypes into production-ready solutions delivering measurable ROI.