Enterprise modernization used to be a budgeting problem. Now it is a compounding capability problem.
Harder, because AI raises expectations for speed while exposing the weakest parts of a technology estate. Easier, because AI can justify modernizing foundations that have been neglected for years.
The problem is that many organizations are treating AI as a layer they can add on top of legacy systems. That approach fails in practice. The data is inconsistent, integrations are brittle, governance is unclear, and teams end up stuck in pilots that never scale. A recent MIT report found that only about 5% of generative AI pilots achieve rapid revenue acceleration, while the vast majority stall with little to no measurable impact on profit and loss.
It’s not that the AI models aren’t advanced enough, it’s because of a lack of preparation.
When enterprises are underprepared
Data trust is the first fault line. AI systems depend on the information you can actually access and reuse, not the information you hope exists. IDC reports that unstructured data, such as documents, tickets, emails, and transcripts, will grow at a 30% compound annual growth rate over the next five years. A lot of unstructured data is dark. This means companies don’t know what they have, why they have it, or how long they need to keep it. They’re complex and often require manual intervention and tagging for downstream purposes. This increases the risk for failure if governance and quality are not addressed.
Technical debt is the second fault line. It acts like a tax that a company pays on any development to redress existing technology issues. As per McKinsey research on technical debt, it accounts for about 40% of IT balance sheets. It’s a compounding problem, too: companies pay an additional 10-20% to address tech debt on top of any project costs.
Legacy constraints are the third. Recent Broadridge study on digital transformation found that over 40% of C-Suite executives and their direct reports say they’re still held back by inflexible legacy systems. This is indeed the practical reason many pilots never leave the lab. When an AI prototype meets the realities of legacy integration, timelines blow up and confidence collapses.
What leaders should prioritize over the next 12 to 18 months
- Build a readiness inventory that is brutally honest: Start with a map of your highest value workflows, the systems they touch, and the data they depend on. Your goal is to identify where AI will fail first.
- Modernize the seams before you modernize the core: Most organizations do not need to rebuild everything. They need consistent interfaces, reusable connectors, and data access patterns that let teams ship safely and repeatedly. This is where AI programs either become a capability or become a pile of demos.
- Put governance on rails: Governance must make shipping predictable. Define what data can be used for what purpose, who approves what, how risk is assessed, how models are monitored, and how incidents are handled.
A lower-risk way to move from interest to outcomes
One practical pattern is to bring real enterprise problems into an environment designed for rapid prototyping and validation, then only scale what survives integration, governance, and operational scrutiny.
This is also the intent behind the collaboration between AXL and Compugen. Compugen, Canada’s largest privately-owned and operated Technology Ally, sees the operational challenges inside large organizations. AXL, a Canadian AI venture studio with a mission to deliver human-centric solutions, focuses on turning applied AI research into deployable products. Together, the goal is to shorten the path from business pain point to validated solution, and to reduce the number of AI initiatives that die in pilot purgatory.
This process enables teams to validate what can be delivered at enterprise grade and then ship it.
The bottom line on AI & enterprise modernization
AI is forcing a reset in enterprise modernization. Those who focus on shipping as many prototypes and pilots will be in a race to the bottom. Meanwhile, the organizations that modernize their foundations, treat data trust and integration as first-order problems, and build governance that accelerates delivery are positioning themselves for success.
As teams review their AI implementation approach, ask this question: Are you building a one-off AI demo? Or are you building an enterprise that can repeatedly turn AI into scalable products and workflows?
About the authors:
Dr. Daniel Wigdor is the Co-Founder and CEO of AXL, a Canadian venture studio dedicated to building scalable, high-impact companies based on world-leading AI research. He is a professor at the University of Toronto and one of the world’s foremost experts in human–computer interaction. His innovations power more than a billion devices globally. He also holds over 60 patents and has published more than 130 peer-reviewed papers.
He founded several companies that have collectively raised $100 million in funding and revenue, including Chatham Labs, which Meta acquired in 2020. At Meta, he became the founding director of Reality Labs in Toronto, leading frontier research in emerging technologies. Through AXL, Daniel is now focused on ensuring Canada doesn’t just invent the future of AI, but owns it, economically, strategically, and at scale. Backed by leading researchers, investors, and industry partners, AXL is on a mission to launch 50 AI companies in the next 5 years, transforming Canadian research into enduring global businesses.
Harry Zarek is the Founder & President of Compugen, Inc., Canada’s largest privately-owned and operated Technology Ally. Under Harry’s stewardship, Compugen has grown and evolved alongside a rapidly changing industry, adopting a world view which recognizes and celebrates the profound influence technology has on both business and society. The company partners with mid-to-large private and public sector customers to dream, design, and deliver Connected + Secured experiences around Modern Workspace and Hybrid IT with a keen focus on Service + Delivery.
More Expert Speak columns on AI:
