Basis Finds AI Task Replacement Outpacing ROI Measurement

Basis’s October 19 survey release gives marketing managers a reason to examine what happens after a task moves to AI. Before changing workloads or renewing tools, include the employee’s review time in the result you measure.

Among 184 marketing and advertising professionals surveyed, 47.8% reported that their organizations had replaced human tasks with AI during the previous year. Only 6.5% reported evaluating AI investments against defined benchmarks. Those answers describe a gap between changing the work and establishing what the change achieved.

Katie McAdams is Basis’s chief marketing and commercial officer. Her assessment:

“Yet, more work is needed to justify the investment”

Managers need to count the work left after the software finishes. A report delivered earlier creates room in the week only after someone checks it and resolves the errors. Any evaluation that stops at generation time leaves that work uncounted.

Count the work that remains

In the public report summary, 75% of respondents felt AI had made them more efficient over the past year. That combines moderate and significant improvements. Perceived gains can identify workflows worth investigating and help explain why employees keep using a tool.

They also answer a different question from a measured return. Someone can feel faster while the company lacks a consistent way to compare total effort with the previous process. The next decision is to establish that comparison before more work gets assigned against the apparent saving.

Use an existing recurring task as the test. For a weekly marketing report, record the time spent preparing the finished version, including corrections. Keep the standard for acceptance consistent across runs. A quicker first draft that needs extra reconciliation belongs in that total, alongside any time the automation saves.

The benchmark also needs an owner. Agree who decides whether the report is usable, then record the corrections that person requests. Changing the acceptance standard halfway through would make the comparison harder to interpret.

There is a concrete earlier example of that remaining work. Anthropic’s July 8 account describes a report-checking instruction that checks each figure using a verified source. On campaigns, Annabel Custer retains approval before release. This is a vendor describing its own process, not independent confirmation of the new survey. It shows a review step a team can include when measuring a workflow.

Our coverage of Ahrefs’ content maintenance workflow examines another person retaining publication approval. Neither example establishes a saving for a different employer. Each makes the handoff visible enough to ask who performs it and how long it takes.

Keep the employment claim within the evidence

The survey’s task-replacement finding doesn’t establish jobs eliminated. An employee can hand a repeated task to software and take on other work. The release doesn’t quantify how those changes affected staffing, so it provides no basis for calculating positions a team can remove.

That distinction matters to the person asked to adopt the tool. Set expectations around the workflow being tested and the evidence that will determine whether it expands. An employee’s corrections are information about how the system performs. Include them in the evaluation, rather than treating the finished draft as proof that the task has disappeared.

There are limits to the broader comparison, too. The issuer describes summer responses from a small sample; the release says August. Its public pages don’t establish representative recruitment, and the full report requires a business-information form. The findings aren’t a census of marketing organizations. They also don’t reveal whether the people reporting task replacement were the same people reporting weak measurement.

Decide what earns a renewal

McAdams advocates unified media platforms. Her proposed route to accountability involves:

“building data foundations to show investment accountability”

Basis sells advertising automation, so its proposed remedy also supports its commercial pitch. The survey alone doesn’t establish which platform a team needs. A manager can begin with a record of an existing workflow before buying another measurement product.

Our earlier reporting on AI cost visibility examined the budget side of this problem. For the next AI budget review, bring one completed comparison: the accepted output and the total effort required to get it there. Renew or expand the workflow against that evidence, with the employee’s review work included in the decision.