Executive survey · Fielded August 3 to September 8, 2026. About

Chapter 03 / The unfinished task

Approved AI still stops before the job is done.

Having a chatbot seat tells a CMO nothing about whether an account brief can land in the CRM, whether a campaign can be set up where it will run, or whether the result will pass review.

A text answer leaves the sales development rep to paste it into the CRM, add the missing account facts and chase the approval. The team saves the drafting time and keeps every other step.

03.1

Marketing needs a say in what the company’s AI must deliver.

Being asked for input is different from agreeing which marketing jobs the company’s AI program has to deliver, and by when.

Read the finding and CMO decision ↗Sign in for the full finding
03.2

Test what your approved AI can actually finish.

An AI model can write an excellent brief and still leave the person to do every step that follows.

Read the finding and CMO decision ↗Sign in for the full finding
03.3

An internal brief should not wait in the same queue as a public claim.

Approval rules work better when they match the action being taken and what it costs to get it wrong.

Read the finding and CMO decision ↗Sign in for the full finding
03.4

A corrected draft should make the next draft better.

Write down why a correction was needed so the next request starts from it.

Read the finding and CMO decision ↗Sign in for the full finding

The numbers behind this chapter

First, how leaders rate this area overall. Then four specific ways of working, each asked separately.

How leaders rate this area overall · Q1

The one-line self-assessment, asked before the detailed questions.

13%Starting. No consistent approach yet.
38%Experimenting. Individual teams are trying their own approaches.
28%Repeatable. Everyday work follows an agreed approach.
18%Established. One shared approach, improved regularly.
2%Don’t know
581 people were asked; blank answers are left out. This is how leaders rated their own company overall. The four specific ways of working below were asked separately. Nobody checked either answer against the actual work.

What is actually in place · Q11

Four things your approved AI can do today, each rated on its own.

Marketing teams can use AI models that have passed tests on their actual tasks.

Not happening 12%A few one-off trials 47%Used in several recurring jobs 28%The normal way of working, reviewed regularly 11%Don’t know 2%

AI can draw on current account, product and market facts it is allowed to see.

Not happening 26%A few one-off trials 32%Used in several recurring jobs 28%The normal way of working, reviewed regularly 10%Don’t know 4%

AI can act inside the work systems, with permissions set for each task.

Not happening 27%A few one-off trials 38%Used in several recurring jobs 18%The normal way of working, reviewed regularly 11%Don’t know 7%

Quality checks, a record of what the AI did, and a named person for exceptions are built into the workflow.

Not happening 24%A few one-off trials 28%Used in several recurring jobs 30%The normal way of working, reviewed regularly 11%Don’t know 7%
581 people were asked; each row counts only the people who answered that row. Each row adds to 100% before rounding. “Used in several recurring jobs” and “the normal way of working” are shown as separate answers, not added together. Answer wording is shortened here; the questionnaire and explorer carry the full text. Explore every response ↗

Put the findings to work

Ask your company’s AI lead to show one marketing task finished end to end with a normal employee login, then agree with legal which steps AI may take on its own for internal, public and spending work.

Read what to do next ↗
Next chapter / 04

Marketing’s AI lead cannot redesign everyone else’s job.