Deloitte Leader: How AI is Blowing Up the Digital Transformation Playbook

TLDR

• Digital transformation in the AI age is different from past tech revolutions because AI can do parts of the work itself.
• Companies can predict the costs of traditional digital transformation efforts more readily than they can with AI.
• AI is not a silver bullet for overcoming problems with legacy systems.

Historically, digital transformation meant optimizing business systems so employees could work more efficiently. AI is blowing up that playbook because it can do parts of the work itself.

That distinction is forcing companies to rethink not just their technology stacks, but their processes, data, workforce and even how they calculate the economics of transformation, according to Karthik Krishnamoorthy, a Deloitte principal and a leader in the firm’s restructuring practice.

“Historically, when we’ve contemplated digital transformation, it’s largely been around systems,” Krishnamoorthy told The AI Innovator. “You take a customer interfacing system, or you take a finance system, and you optimize it. That was historically digital transformation.”

With generative and agentic AI, three things have changed, he said.

The first is the emergence of what Krishnamoorthy calls a “simulated workforce.” Rather than just giving employees better systems to use, AI also brings in digital workers capable of performing tasks themselves.

Second, the process itself can change. In the past, companies typically used technology to improve an existing workflow. AI can eliminate parts of that workflow altogether. For example, Krishnamoorthy said, a financial analyst previously might have had to spend time pulling data from earnings reports to do an analysis. AI can perform much of that preparatory work, enabling companies to redesign the process rather than merely making it faster.

AI makes costs harder to predict

The third major difference is financial.

Companies previously had relatively established models for determining how much a technology project would cost and how much value it could produce, Krishnamoorthy said. AI has introduced another variable: token consumption. Different AI models and workloads can consume dramatically different numbers of tokens, making costs harder to predict.

“Tokenomics is a new field. We’re all trying to figure it out,” he said. “The token usage can be wildly different.”

Historically, Deloitte might estimate that maintaining a fully automated process would cost about 10% of the original process, he added. With AI, companies cannot make that assumption with the same confidence because costs vary according to how AI is being used.

Deloitte’s 2026 Enterprise Cost Transformation Survey found that 77% of companies expect to invest between 20 cents and 90 cents for every dollar of targeted savings. The wide range reflects the different kinds of transformations companies undertake. Vendor consolidation, for example, may require relatively little upfront investment. Integrating thousands of applications and creating consistent data across an enterprise can require a multiyear investment.

Krishnamoorthy emphasized that much of that spending is intended to be a one-time investment that produces continuing benefits. “You’re investing once to get advantages for many years to come.”

AI runs into legacy systems

That opportunity is colliding with the reality of enterprise technology.

Deloitte’s survey found technology modernization among the five biggest priorities for 67% of respondents, while 66% cited AI-enabled automation. Yet 55% of executives pursuing AI-enabled transformation said their existing technology systems cannot support it.

“Most of the companies in the Fortune 100 ranks still use mainframes,” he said. He also pointed to relational databases and other legacy technologies that remain embedded in enterprise systems.

To be sure, mainframes are also evolving to accommodate AI. IBM, for example, has been modernizing its mainframe systems with AI capabilities, while APIs can allow newer applications to access data and functions housed in older applications. Companies can also connect mainframe data with modern cloud and AI platforms without replacing the systems that continue to process critical transactions.

Other core enterprise systems are changing as well. CRM, ERP and other business software platforms increasingly incorporate AI directly into applications employees already use, allowing AI assistants and agents to analyze information and automate parts of workflows.

The bigger challenge for established companies is often the technology estate accumulated around those systems: fragmented data, thousands of applications, inconsistent architectures and business processes that were never designed around AI.

10,000 pieces of software

Krishnamoorthy said it is not unheard of for a Fortune 100 company to use 10,000 pieces of software – including licenses, third-party software, hosted software, cloud software and other applications.

Fix those systemic issues before layering in AI, he said. Simply adding another piece of AI software as a “silver bullet” for that complexity is unlikely to work, he said.

Another issue is audit trails. Under traditional systems, if a decision is challenged, there’s typically a way to find out whether the employee followed standard operating procedures. For example, if a medical claim is denied, one could see if the employee denying the claim followed company policies. “That’s pretty defensible,” Krishnamoorthy said.

AI can make accountability more complicated because, unlike a traditional rules-based system, the internal process that produces a model’s output may not be fully interpretable. This is AI’s ‘black box’ problem.

Companies can constrain AI by giving it explicit rules that require it to document the evidence behind a decision, but that is not the same as being able to inspect the model’s internal reasoning.

Beware of ‘value leakage’

But whether traditional or AI-driven transformation, enterprises must beware of what he calls “value leakage.”

“When you embark on the program and you start running the program, invariably 20% to 25% of that portfolio leaks,” Krishnamoorthy explained. “What we essentially mean by that is you’re not able to execute that initiative or bring that lever to fruition, or even if you do so, you’re not able to extract the P&L value and the advantage that you contemplate.”

He identified three recurring causes.

The first is beginning with an unrealistic baseline. For example, a company might have $1 billion in total costs but only $700 million may consist of expenses it can realistically reduce. Calculating a savings target against the $1 billion can make it unrealistic from the start.

The second is inadequate governance and leadership accountability. If no single executive is clearly responsible, accountability can become blurred.

The third is when companies stack multiple performance targets on top of one another and effectively try to extract the same savings several times.

A business, for example, may already have an annual performance-improvement target. Regulatory pressure or another problem can create an additional savings requirement. Management then launches a broader cost-optimization program intended to generate capital for growth.

Let’s say a company spends $1 billion a year and sets the following goals:

• Its annual efficiency program calls for, say, $50 million in savings.
• Management promises another $40 million in cuts to deal with a regulatory or earnings problem.
• It launches a transformation program targeting another $100 million in savings to fund future growth.

On paper, those look like three separate initiatives totaling $190 million in savings. But the teams running them may all be looking to cut the same employees, vendors, software, facilities and other operating expenses.

“Now you’ve got three competing goals,” Krishnamoorthy said. “They’re going after the same P&L. They’re going after the same cost base.”

That unnecessary complexity, he said, increases the likelihood of value leakage.

These problems help explain why transformation success remains uneven. Deloitte found that 96% of surveyed organizations had completed, were undertaking or were planning a cost transformation, but only 14% were delivering the expected value against their biggest priority.

“Unless you fix your data, your platform, your people, your process, extracting value from AI is always challenging,” Krishnamoorthy said. “It’s not that AI is not efficient; we’re not giving it the ingredients to be effective.”