The hidden costs distorting your AI ROI | MarTech

AI is delivering measurable gains for marketing, but proving what those gains are worth is surprisingly difficult. As spending accelerates, the gap between using AI and demonstrating its financial impact is becoming harder to ignore.

By 2029, AI will power more than 50% of all U.S. marketing activity, according to The CMO Survey. Furthermore, last year, AI helped sales productivity and customer satisfaction increase by 14.1% and 10.8%, respectively. And marketing overhead decreased by 14.6%, a substantial year-over-year improvement.

However, demonstrating ROI on AI investments remains a challenge. Only 9% of senior executives surveyed by Witness.AI said over 75% of their AI initiatives showed meaningful financial returns. Additionally, 68% said that AI programs were over budget at some point in the preceding 12 months.

The marketing picture is just as grim: Only 16% of respondents in a global survey of CMOs could confidently prove the results of AI investments. Nearly 70% of those surveyed could not measure results with much precision, and 21% had no infrastructure for consistent measurement. One of the biggest reasons for this is that AI’s impact spans so many operations.

How can marketers unlock and show value from AI investments?

Measure the change, not the AI

The hard part of getting value from AI is no longer giving models access to more data. It is determining what they need to know and whether the underlying data can be trusted.

Connecting more systems will expose AI to inconsistent definitions, duplicated records, stale information, and conflicting signals. An AI agent needs to know what a field means, when it was updated, which source takes precedence, and what actions the information permits. Data architecture, therefore, becomes part of AI architecture.

AI ROI gets slippery when organizations measure the activity AI performs rather than the change it produces. Generating more assets, producing them faster, or saving employee hours demonstrates operational improvement.

Measurement should start with the process AI changes, specifically, how it performed before deployment. The more removed a claimed benefit is from the activity AI directly affects, the harder it is to prove AI produced it.

ROI calculations also need to include integration, data preparation, governance, monitoring, training, and human review.

AI can move work instead of eliminating it

Generating hundreds of content variations quickly doesn’t save much time if employees then spend hours checking them for accuracy, brand compliance, duplication, and legal risk. The same goes for agents: They can cut research time while creating more work reviewing exceptions and monitoring their decisions. A 70% reduction in one task isn’t much of a productivity gain if the work simply moves somewhere else.

That’s why it makes more sense to measure the entire workflow, rather than the task AI performs.

The same principle applies when choosing AI tools. An impressive tool won’t deliver much value if employees constantly have to move data between systems, supply missing context, fix outputs, or wait for approvals. A less capable tool that fits smoothly into the workflow may ultimately save more time and money. How well a tool fits the way people actually work can matter more than how many features it offers.

AI ROI can disappear between departments

There’s another problem with calculating AI ROI: The costs and benefits often show up in different departments. Marketing may get the productivity gain while IT pays for infrastructure, engineering handles integration, legal and security take on governance, and other employees absorb additional review work.

That can make a marketing AI initiative look more profitable than it really is because some of the costs are on someone else’s budget.

A more accurate calculation looks beyond the marketing team’s numbers. Measure what changed across the entire workflow, count the costs wherever they occur, and only credit AI with financial results you can reasonably connect to the work it changed.