Santander and Novartis Test the Returns on AI Spending

Santander and Novartis say AI is beginning to lift revenue, reduce costs or improve profitability. For investors who have spent years hearing what the technology could do, companies pointing to gains in ordinary operations offer a more useful test. Their claims do not yet put a quantified AI contribution beside the spending that produced it.

That leaves a practical question for shareholders, employees and customers: when a company says AI helped its results, what changed on the income statement, and who paid for the change?

The two companies illustrate why there will be no single answer. A bank might use AI to handle routine work or support commercial decisions. A pharmaceutical company might find a different set of uses across its operations. Those are examples of where gains could arise, rather than descriptions of specific projects at Santander or Novartis. Each would need a different measure of success. Time saved on a task, revenue won and expenses removed from a budget all have different effects on profit.

What an AI Return Would Have to Include

Imagine a company deploys an AI tool that saves its staff 100 hours a week. That is an operational improvement. The financial return depends on what happens next. The company might serve more customers with the same team, avoid hiring, reduce overtime, or leave staffing and output unchanged while giving employees more room for other work. Only some of those outcomes will show up quickly as lower costs or higher sales.

The spending side has similar layers. A comparison has to include software subscriptions or computing costs, integration with existing systems, staff training, oversight and any additional work created when the tool makes mistakes. It also has to account for what the company would have spent without AI. Replacing an older service differs from adding a new expense line.

Revenue claims require another check. If sales rise after an AI deployment, prices, demand and sales staffing may have changed at the same time. A company can make a stronger case by comparing similar teams or processes, showing what happened before and after deployment, and explaining its method. Investors need not demand a laboratory experiment from every earnings call. They can ask management to identify the measure, the cost and the period over which the gain held.

A modest, repeatable improvement can be valuable. If a tool helps a bank process routine requests faster while preserving the quality of service, the gains may accumulate across a large operation. If it helps a business make better commercial decisions, the effect may appear in revenue rather than headcount. The strongest case for enterprise AI is a collection of such improvements, each tied to a process the company already pays for.

The Buyers and Sellers Face Different Bills

Investor pressure has another side. Fears surrounding an $828 billion AI capital-spending spree helped erase $1 trillion in market value. That figure concerns the scale of spending under scrutiny, rather than a bill assigned to Santander, Novartis or their individual AI projects.

The largest technology companies spend on infrastructure that they hope many customers will use. Enterprise customers buy or build tools because they expect improvements inside their own businesses. A successful AI deployment at a bank could strengthen the case for buying AI services. It does not, by itself, reveal whether a supplier charged enough to cover the infrastructure behind those services.

That split is easy to miss when every participant uses the word return. A customer can save money while its supplier struggles with computing costs. A supplier can grow revenue while a customer has yet to turn a pilot into a durable workflow. Earlier AI capex scrutiny centered on the cash needed to build infrastructure; these enterprise claims concern what customers can earn or save by using it. Both sides of the transaction have to work for the spending cycle to sustain itself.

Timing complicates the comparison. Infrastructure costs can arrive well before customers settle on dependable uses. Enterprises, meanwhile, can report an immediate task-level improvement while still paying to expand and supervise a deployment. An investor looking at either side needs to know whether today's result survives wider use and whether the next unit of spending produces a similar benefit.

Who Keeps the Efficiency Gain?

Employees experience an efficiency program through schedules and job design before investors see it in a margin. If AI allows a team to handle more work, management can reinvest the capacity, slow hiring or cut positions. Those choices have different consequences for workers and different financial effects for the company. A cost saving also looks different if maintaining service quality requires more human review than management anticipated.

Customers have a stake in the same decision. Faster service could make a product better. Lower operating costs could support lower prices, though companies may instead retain the savings. If AI changes how a service operates, speed and cost need to be considered alongside errors and the effort required to resolve them. An attractive margin improvement gives outsiders only part of that picture.

Founders selling tools into banks, drug companies and other large organizations face their own version of the return test. An enterprise deal can look impressive on a pitch deck. Its economics depend on what the vendor spends on computing, implementation, support and sales to win and keep that customer. A growing customer list can coexist with weak margins if each installation requires extensive custom work.

Funding terms shape how long a startup can pursue those deals. Suppose, purely as an illustration, an investor puts $10 million into an AI vendor for 25% of its shares with a one-times nonparticipating liquidation preference. If the company sells for $20 million, the investor would take the $10 million preference rather than convert its shares into a $5 million payout. The remaining $10 million would go to the other shareholders, including any employees who own shares. At a $100 million sale, conversion would pay the investor $25 million instead. Actual payouts depend on the full cap table and deal terms. A headline valuation alone cannot tell staff what their equity will be worth.

It explains why the commercial case for an AI product matters to people beyond its backers. A vendor under pressure to grow can pursue expensive contracts that flatter revenue while leaving little room for profit. A founder with a product that serves customers profitably has more options, including slower growth financed by customer payments rather than another funding round.

The next useful disclosures will connect deployments to costs, revenue or margins over more than one reporting period. Management can show whether a gain came from serving more customers, spending less to serve each one, or changing staffing. Investors can then compare those results with the continuing cost of the tools. Workers and customers can judge what changed for them. Santander and Novartis have put AI's early returns on the agenda; the next set of numbers will show how far those returns travel.