EY: How Companies Can Turn AI From a Cost Cutter Into a Growth Engine

AI has largely entered corporate America through the cost side of the ledger. But that is starting to change.

The 2026 EY-Parthenon Growth Survey found that 78% of corporate growth leaders believe AI will accelerate growth. But 63% said their companies primarily use AI for efficiency and productivity, while 49% strongly agreed their organizations know how best to use data and AI to enable growth.

“The real gap here from what we see in the survey results? It’s execution,” said Malinda Gentry, TMT and technology leader of EY-Parthenon, which is EY’s strategy and transactions consulting arm.

For companies trying to move beyond productivity gains, Gentry said the starting point is not choosing an AI model or compiling a long list of potential use cases. It is identifying a small number of business decisions where better judgment or faster execution could materially affect revenue.

“Those that grow and scale start with the business decision, not the tech,” she told The AI Innovator. “They’re clear about the problem.”

Start with the business problem

After identifying the business problem, determine how AI can support it.

That could mean deciding how to price a product, which customers should receive a particular offer, where a company should invest or which market it should enter.

The approach runs counter to the tendency to deploy AI first inside individual departments and look for benefits afterward.

She recommends that executives identify two or three major growth priorities and then align capital, talent, data and decision rights around them.

That alignment is important because many of the highest-value AI opportunities cut across traditional organizational boundaries.

A revenue process, for example, may touch sales, finance, technology, risk and operations. Gentry said companies pursuing growth are beginning to redesign their operating models so those functions work together rather than applying AI separately within each department.

“They’re looking at it across the process, not in silos,” she said.

Governance also has to be designed into the process from the beginning, including clear decision rights and accountability for results.

Legacy systems get in the way

That sounds simpler than it is.

A large multinational may have Salesforce managing customer relationships, SAP or Oracle supporting enterprise resource planning and numerous additional applications involved in individual business processes.

Gentry used lead-to-cash as an example. At a major company, the process could involve hundreds of employees and 10 to 15 different systems.

“That is the legacy problem,” she said. “I’m not going to tell you that is impossible to work around, but it is very hard.”

That complexity helps explain why corporate enthusiasm for AI has moved faster than execution.

Companies cannot necessarily redesign a revenue process simply by dropping an AI agent into one part of the workflow. Data must move between systems, agents must interact with existing applications and different departments must agree on how the redesigned process will operate.

Some companies are using acquisitions, spinoffs and other major corporate events as opportunities to rethink those processes.

Gentry described one large company using an acquisition to reimagine its finance function rather than simply combining two existing technology stacks. The organization is developing agentic processes above its existing systems, beginning in areas including financial planning and analysis, tax and the financial close.

Companies without such a triggering event are taking smaller pieces of large business processes and assigning cross-functional teams to redesign them separately from day-to-day operations.

“There is no easy button here,” Gentry said.

Put agents above existing systems

Few large companies are prepared to replace their core technology stacks simply because newer AI-native alternatives have appeared.

Instead, many are adding AI on top of what they already have.

Enterprise software providers are embedding AI and agentic functions into ERP, CRM, billing and other products, allowing customers to adopt new capabilities without immediately abandoning existing systems.

But Gentry cautioned that the integration work does not disappear.

“When those new features and functions come into your ERP or CRM, you inevitably still have work to do,” she said. “Because you’ve got data work to do or integration work to do.”

Agents create another layer.

Gentry used expense management as an example. An automated process might already route an expense report through a company’s systems. If an item requires review, an AI agent could question the employee rather than immediately handing the task to another worker.

But the agent still has to communicate with the system that executes the transaction.

That means businesses trying to create end-to-end agentic processes must connect the agent layer with the underlying applications rather than treating the agent as a standalone tool.

For now, Gentry does not expect large global companies to quickly abandon established systems.

“I do not think the Fortune 500 is ready to rip and replace,” she said.

Smaller companies may have more flexibility. Gentry cited an approximately $800 million-revenue client replacing an aging ERP system with an AI-native alternative.

She said newer AI-native ERP, CRM and billing vendors could eventually support increasingly large companies, although they are not yet equipped for all of the complexity faced by major global enterprises.

Use proprietary data to create differentiation

Once companies solve the execution problem, Gentry said another question becomes important: What does the business possess that competitors using the same AI models do not?

Her answer is often proprietary data.

“When we think about the impacts of AI and durability, a lot of that comes down to proprietary data and data that cannot be synthetically replicated,” she said.

Other sources of differentiation could include intellectual property, customer relationships, institutional knowledge and regulatory advantages.

The underlying AI model itself is less likely to provide a durable moat because competitors can increasingly access many of the same models.

Gentry said companies should therefore build AI architectures that allow them to switch models as the technology changes rather than becoming dependent on a single provider.

She cited advice she heard from the CEO of a large hyperscaler: “Use all, be dependent on none.”

The principle, she said, is to make the company’s proprietary assets durable while keeping the underlying AI technology replaceable.

Put business rules around AI decisions

Moving AI from productivity tasks into revenue decisions also raises the stakes.

Companies may be comfortable using AI to summarize documents or automate administrative work but less willing to rely on it when determining prices, allocating capital or evaluating an acquisition.

One approach Gentry discussed is neurosymbolic AI, which combines systems that learn patterns from data with systems that apply explicit rules, logic and relationships.

For example, AI analyzing an acquisition target could identify patterns in market growth, customer behavior and comparable transactions. A rule-based layer could then apply a company’s requirements, such as minimum investment returns or unacceptable risks.

The approach is designed to combine pattern recognition with structured reasoning.

Gentry said similar techniques could inform pricing, personalization, customer retention, next-best offers and investment decisions.

It could also help companies connect large amounts of unstructured information with structured knowledge about markets, customers and products.

The technology remains only one component, however. Gentry’s broader prescription is to combine AI analysis with the rules and institutional knowledge already inside the business rather than turning consequential decisions over to a model alone.

Build for adaptability

For executives, Gentry boils the growth strategy down to a few priorities.

First, identify the small number of growth decisions where better judgment and faster execution could change the company’s trajectory.

Second, align capital, talent, data and decision rights behind those priorities rather than scattering AI initiatives throughout the organization.

Third, build for adaptability.

“Sustainable advantage is going to come from the proprietary data, the customer relationships, the IP,” Gentry said.

That makes the question facing companies less about how much AI they deploy and more about where they deploy it.

Using AI to process the same work faster can reduce costs. Using it to improve pricing, identify customers, find new markets or make investment decisions requires companies to redesign how those decisions are made in the first place.

That execution challenge may determine which companies turn AI spending into new revenue – and which simply become more efficient at doing what they already do.