Companies Have Spent Billions on AI. Now Comes the Hard Part: Proving It Makes Money

Photo By: Towfiqu barbhuiya

Imagine a company has spent the past year deploying artificial intelligence across its marketing organization.

Campaigns are created faster. Teams analyze customer feedback in minutes instead of days. AI agents surface opportunities employees might previously have missed. Thousands of hours of work have been automated.

Then the CFO asks a different question: How much incremental profit did it create?

The answer is surprisingly difficult.

As enterprise AI moves from experimentation into operating budgets, companies are entering a more demanding stage of adoption. Executives no longer need only to demonstrate that employees are using AI or that a system can complete a task. Increasingly, they have to show that those capabilities create economic value.

That shift could change how companies measure AI altogether.

AI’s First ROI Story Was Productivity

The initial enterprise case for generative AI was relatively straightforward.

If software could summarize a document in five minutes instead of an hour, the time savings could be measured. Customer-service systems could track tickets resolved. Coding assistants could measure development cycles. Marketing teams could compare the time required to produce content before and after AI. Those remain legitimate benefits.

But enterprise AI is expanding beyond assisting employees with tasks. AI agents are increasingly being designed to participate in business processes, identify opportunities and recommend actions.

The closer those systems get to revenue-generating decisions, the less useful productivity alone becomes as the measure of success.

Recent research illustrates the gap. A 2026 survey by Boston Consulting Group and The Consumer Goods Forum found that more than half of participating CPG and retail companies did not formally measure the ROI of their AI investments. Roughly three-quarters of CPG respondents remained in pilot or exploration mode, and the most frequently cited barrier to scaling was that pilot economics did not translate into full-scale business results.

The problem isn’t necessarily that AI isn’t creating value.

Companies may still be struggling to determine exactly where that value came from.

Revenue Is Not the Same as Incremental Revenue

Consider an AI system that recommends a promotion.

The company follows the recommendation and sales rise 15%.

It would be tempting to credit the AI.

But sales could have increased without the promotion. Demand might already have been rising. A competitor could have experienced a stockout. Seasonality could have contributed to the increase. Customers may simply have moved purchases forward to take advantage of the discount.

Revenue can rise even while contribution margin deteriorates.

The financial question is therefore not simply what happened after an AI recommendation.

It is what happened because of it.

That is a counterfactual problem: what would the business outcome have been if the company had made a different decision, or done nothing?

Marketing departments have wrestled with attribution for years. As AI systems begin influencing pricing, promotions, inventory and capital allocation, AI itself may face an attribution problem.

AI ROI Is Becoming an Incrementality Problem

This creates a higher standard for systems that claim to improve commercial performance.

A writing assistant can reasonably demonstrate value through time saved.

A system recommending a multimillion-dollar change in marketing allocation faces a different burden. Executives need to understand not only whether revenue increased afterward, but whether the decision generated incremental revenue or profit relative to the alternative.

That requires connecting AI recommendations to measurable business outcomes.

The timing is notable. Gartner’s September analysis of 107 agentic AI deployments found that the largest opportunities for business value were emerging from specialized, domain-specific agents rather than broad general-purpose systems. Gartner predicts those specialized agents will account for 80% of tangible agentic AI ROI by 2028.

Specialization can make the ROI question more concrete.

A general assistant might be evaluated by how much work it accelerates. A pricing system can ultimately be evaluated against price realization, volume, revenue and margin. A promotion system can be judged against incrementality. An inventory system can be measured against working capital and availability.

The closer an AI system gets to a specific business outcome, the harder it becomes to hide behind usage metrics.

Designing AI Around the P&L

Kapnova approaches the problem from that direction. The company is an agentic revenue and profit optimization system built for consumer brands. Its agents identify potential commercial opportunities across company and market data, while quantitative methods evaluate what different actions are likely to do to revenue, gross profit and contribution margin.

CEO James Sun co-founded the company with CTO Shenbo Xu, whose background includes causal research at MIT and quantitative work at Point72 and Scale AI.

That causal foundation matters to the ROI question because measuring incrementality requires distinguishing an outcome that followed a decision from one that occurred because of it.

Not every business question requires causal inference. Forecasting, econometrics, optimization and simulation can answer different questions. But a system claiming to improve commercial performance eventually needs a mechanism for comparing the decision made with plausible alternatives. Otherwise, companies risk measuring activity instead of value.

AI’s Next Benchmark May Be Economic

The AI industry has spent years comparing models through benchmarks: reasoning performance, coding ability, latency, context windows and increasingly complex evaluations.

Businesses ultimately operate under a different scoreboard.

Worldwide AI spending is still accelerating. Gartner currently forecasts a 49.5% increase in 2026 alone. At the same time, its recent research increasingly emphasizes linking AI consumption and agentic systems to measurable business value.

That pressure is likely to grow as experimental AI budgets become permanent operating expenses.

For enterprise leaders, the question will no longer be how many AI agents were deployed, how many recommendations they generated or even how many employee hours they saved.

Eventually someone will ask which decisions actually changed the business outcome.

And whether that change was worth more than the AI cost.