Why faster AI tasks do not always improve ROI

AI can increase sales productivity by 14.1% and still leave executives unable to prove whether their AI investments make money. The CMO Survey projects that AI will power more than 50% of all U.S. marketing activity by 2029. The measurement boundary explains how both can be true: a company can measure a faster AI-assisted task while excluding costs and extra work elsewhere in the workflow.

AI can improve the task and still leave ROI unproven

The CMO Survey reports three operating improvements associated with AI:

These figures answer a different question from whether the financial benefit attributable to an AI investment exceeds its full cost.

A Witness.AI survey found that only 9% of senior executives said more than 75% of their AI initiatives showed meaningful financial returns. Some 68% said AI programs had gone over budget at some point during the preceding 12 months. Witness.AI sells AI security and governance technology, giving it a commercial interest in enterprise concern about the management and economics of AI deployments.

A global survey of CMOs produced a similar measurement gap:

Together, these measurements show the distinction executives need to test. Evidence that AI changes an activity is different from evidence that the change produces an enterprise financial return.

Task speed and workflow productivity measure different things

Consider a hypothetical marketing team using generative AI to produce hundreds of content variations. Generation time may fall sharply, while employees still need to check output for factual, brand, duplication, or legal issues. Measuring generation time alone would omit some of the labor required to produce approved content. The relevant comparison is the change in total resources needed to complete the workflow.

The same rule applies to AI agents, systems designed to perform multi-step work and take actions with some autonomy. Executives should count supervision, exception handling, correction, escalation, and approval introduced by an agent alongside the labor it removes. A large reduction in research time can produce a smaller improvement in total cycle time when extra work appears elsewhere. The economic result depends on the net change across the process.

This changes the unit executives should measure. The boundary should run from the inputs employees provide through the final approved business output affected by the deployment. Compare labor, elapsed time, error correction, review, and other required resources before and after implementation. That comparison separates a faster task from a more productive workflow.

Causal distance creates another measurement challenge. AI may directly reduce the time needed to create a campaign asset, making production time a relatively immediate outcome. Revenue, retention, and customer satisfaction sit farther downstream and can reflect many business changes. Executives need stronger attribution evidence before assigning those outcomes to a specific AI deployment.

Enterprise costs can cross departmental budgets

Workflow boundaries are one part of the ROI model. Organizational boundaries create another. When an AI initiative requires infrastructure, integration, governance, security, training, or human review, the calculation should include those resources even when another function pays for them. The test is whether the deployment caused the cost.

A departmental calculation can become distorted when one function records a productivity gain while costs caused by the same initiative sit in other budgets. For a CMO, an AI project could meet a marketing efficiency target while producing weaker economics for the enterprise. A CTO or engineering leader could face the reverse accounting view when their function incurs implementation costs while another records the benefit. Both views become comparable when the calculation follows the same changed workflow.

This gives executives a simple boundary rule. Include integration, data preparation, governance, monitoring, training, and human review when the initiative requires them. Assigning a cost to engineering, security, legal, or marketing changes departmental accounting. The enterprise ROI calculation should still count the resources consumed to produce the measured benefit.

AI capability creates value through workflow fit

The same economic test applies to the systems an AI tool can access. When a deployment connects additional data sources, executives should assess any work needed to establish data quality, definitions, source precedence, and permissions. Those requirements can affect implementation costs and ongoing labor. Technical access creates economic value through the change it produces in the business process.

For an AI agent to use a field reliably, the deployment may need rules covering what the field means, when it was updated, which source takes precedence, and what actions the information permits. These are data architecture decisions: choices about how information is structured, defined, governed, and exchanged across systems. When a deployment requires this work, its cost belongs inside the investment case. This connects technical design directly to the ROI boundary.

Tool selection should compare changes across the full workflow. A product with more features can deliver weak savings if deployment requires enough data transfer, context entry, correction, or approval to offset those gains. A simpler system can deliver greater economic value when it removes more total labor or delay. The decision depends on measured net process improvement rather than feature count.

Measure the changed workflow

A stronger ROI model starts with a pre-deployment baseline. Measure the labor, process steps, elapsed time, and resources required to reach an acceptable outcome before the AI system changes the work. After deployment, use the same boundary and measure what disappeared, what remained, and what new work appeared. This creates a consistent basis for calculating the net change.

The post-deployment calculation should capture the resources the AI system adds to the process and the costs it causes across functions. This keeps the numerator and denominator at the same organizational level. Enterprise benefits are compared with enterprise resources consumed.

Executives should distinguish direct operational effects from downstream financial outcomes. A measured reduction in time for a defined activity can be compared directly with its pre-deployment baseline. Assigning higher revenue, retention, or customer satisfaction to AI requires evidence that separates the deployment’s effect from other changes affecting those outcomes. The ROI model should reflect that difference in attribution confidence.

Measurement can shape purchasing decisions before money is committed. Teams can estimate how each option changes total labor, elapsed time, required context, approvals, corrections, and cross-functional resources, then apply the same measures after deployment. Comparing actual results with the baseline makes workflow fit an economic variable in the investment decision rather than a technical characteristic considered in isolation.

Main highlights

• Measure AI ROI beyond task speed: Faster individual tasks do not prove enterprise returns. Measure the full workflow, including review, correction, supervision, and approval required to produce an acceptable output.
• Count costs across departmental boundaries: Include integration, data preparation, governance, security, training, monitoring, and human review when the AI initiative causes them, regardless of which function pays.
• Evaluate AI based on workflow fit: More features or data access do not automatically create more value. Compare tools by their net effect on labor, delays, corrections, approvals, and other resources across the process.
• Establish a consistent workflow baseline: Measure labor, elapsed time, process steps, and resources before and after deployment using the same boundary. Treat downstream outcomes such as revenue or retention separately when they require stronger attribution evidence.