Deloitte: Technical Debt Is Now a Binding Constraint on AI Adoption

Technical debt has become a binding constraint on AI adoption, and many companies are paying it down in the wrong order. That is the argument of a Deloitte US consulting perspective by Tejas Desai, Siva Muthu, Jayanta Ghosh, Nandan Muralidharan and Aradhana Prema Harikumar. Treat debt as an IT backlog item, they write, and teams fix the most visible problems rather than the ones with the biggest business impact.

The authors put debt on both sides of the P&L. It is an “innovation tax” that slows time to market, and a cost multiplier that keeps cost to serve high even as AI spending grows. AI can add to the pile: they cite Gartner’s projection that prompt-to-app building by citizen developers will raise software defects by 2,500% by 2028. The counterexample is Ramp, whose AI-native financial stack reached 70,000 business customers and $1 billion in annual revenue in seven years.

The method is a sort, not a rewrite. Deloitte assesses debt across five layers — user experience, applications, data, security and infrastructure — around a core of business processes, then places each item on a value-versus-cost matrix: non-negotiables, quick wins, strategic bets, fill-ins and items to deprioritize. Its strategic-bet example is Vanguard’s move from a mainframe brokerage platform to a cloud-native, API-first architecture, which cut infrastructure total cost of ownership by nearly 30% between 2022 and 2025.

BCG’s research on industrial companies shows the same gap from the adoption side. Almost 70 percent of US industrial AI adopters make no organizational, data or process changes, and only 12 percent build new workflows. BCG puts roughly 70 percent of AI value in people and processes, not algorithms.

The move for leaders funding AI is Deloitte’s own: govern debt as a continuous enterprise portfolio with C-suite co-ownership and explicit risk and funding thresholds. Then rank modernization work by the AI outcomes it blocks — not by which system complains loudest.