More Marketing Organizations Use Agentic AI. Few Measure the Return on Their AI.

What happened

In 2026 Basis, a vendor of media automation software, published AI and the Future of Marketing: 2026 Edition, its annual survey of marketing and advertising professionals at agencies and brands. The 2026 sample is 184 people; the 2025 sample was 140, so year-on-year comparisons are between different groups.

The report documents fast movement in three places. Agents: 51.1% of respondents say their organization has created or begun to use agents for marketing or advertising work, against 20.0% in 2025. Among the 94 respondents whose organization uses agents, the most common jobs are reporting and analytics (72.3%), drafting campaign briefs (56.4%), content creation (45.7%), creative optimization (38.3%) and media planning (35.1%). Investment: 75.0% say their organization invested in technology to automate or streamline processes in the past 12 months (63.6% in 2025), and 78.8% plan to in the next 12. Self-reported gains: 33.2% say AI made them significantly more efficient, up from 25.7% in 2025 and 13.5% in 2024.

Then the report asks the question this piece is about: does your organization measure the return on its AI investments? 6.5% say yes, with defined metrics and benchmarks. 23.4% say yes, but only anecdotally. 18.5% say no but plan to, 27.7% say no, and 23.9% are not sure. Basis' own reading, on page 20, is blunt: organizations without defined benchmarks "cannot distinguish a genuinely effective deployment from an expensive one that merely feels productive."

What the evidence actually says

Observed fact. One survey, by a software vendor, of 184 professionals, fielded 16 July to 14 August 2026. The figures above are from pages 2, 12, 16, 20, 21, 30, 31 and 41 of the full report, which we read.

Reported claim. 51.1% of organizations use agents; 6.5% measure AI return with defined metrics; 70.1% do not measure it or do not know whether anyone does. These are respondents' descriptions of their own organizations, not audited measurements. The efficiency gains are self-assessments by the same people, not measured outcomes.

Inference (ours). The gap between 51.1% using agents and 6.5% measuring AI return is a gap between adopting a capability and being able to say whether it works. The survey does not link the two: it does not report how many organizations that use agents also measure their return. The 6.5% concerns all AI investments, not agents alone.

Unknown and limitations. The sample is small, and the report does not describe how respondents were recruited or weighted. The agent question asks whether the organization has «created or begun to use» agents; it does not measure how autonomous they are, so the 51.1% should be read as «has started», not «relies on». Basis sells media automation, which gives it an interest in the theme. Basis also suggests that weak first-party data, which only 13.0% describe as extensive and well-structured (17.9% in 2025), is probably holding agents below their potential; that is the vendor's interpretation, not something the survey tested.

Why this matters for marketing

Basis points out that marketing has spent two decades moving toward accountability for every dollar of media spend. The pattern in its own survey is different for AI: money and workflows move first, measurement later or not at all. That matters most for agents, because an agent changes who does a piece of work and often what that work looks like, and that can happen without a before-and-after.

Basis names three consequences of having no defined benchmark (page 20): an organization cannot tell an effective deployment from one that merely feels productive, cannot build the case to scale what works, and cannot say what to cut when budgets tighten. None of those is a technical problem. Each is a decision that needs a number nobody wrote down at the start.

The operating implication

1. Write the before-number first. For each agent or workflow in use, record the one number it should move and its value before it was switched on, taken from reports that predate it. If no such report exists, write «no baseline», start measuring now, and set a date to compare. The last few weeks of an agent already running are not a baseline.

2. Give each one an owner, a date and a rule. One named person checks the number on a set date. Decide in advance what happens next: if after the review period the number has not moved, narrow the agent's job or stop it.

3. Keep the evidence classes apart. Time saved reported by the people who use a tool is a self-assessment. A change in the number the workflow exists to move is a result. Reports should say which one they show.

A practical start takes about thirty minutes: take your three busiest agents or workflows, by volume handled or by cost, and write one line for each: what it does, the number it should move, that number before it started, who checks it, on what date, and what you will do if it has not moved.

The Olymuse lens

This sits under Judgment & Governance. In What Is Process Intelligence in Marketing? we argued that a gap only matters in relation to the decision it affects. An agent is a recurring decision about how work gets done, and the evidence behind that decision is a before-number, an owner and a date. Our point of view: an agent without a baseline can show that it is busy, not that it worked, so the measure comes before the next agent.

This is an editorial reading of external research. It is not a claim that Olymuse has measured agents' returns for any marketing team.

What we would watch next

First, whether Basis or others report, in the same sample, how many organizations that use agents also measure their return, which this survey does not show.

Second, independent studies that define «agent» tightly and report measured changes in a business metric, rather than self-assessed efficiency.

Third, whether the share of organizations measuring AI return with defined metrics moves from 6.5% in next year's edition.

---

Read next: What Is Process Intelligence in Marketing? · Olymuse is in founder-assisted private alpha. Apply for the Private Alpha →