What it is

Capgemini Research Institute with frog, the consultancy’s design arm. The report ends with a catalogue of Capgemini’s marketing services.

The survey ran in mid-2025 with directors and above who oversee marketing strategy, 27% of them CMOs, alongside 29 executive interviews. Its funding finding is the one finding 03.1 draws on: 39% say generative and agentic AI marketing initiatives are funded by marketing, 55% say by IT. CMOs involved in critical decisions fell from 70% in 2023 to 55%.

On AI results, 72% now use generative AI in marketing, up from 37% in 2023, but only 7% strongly agree it has boosted their effectiveness, and realized gains in content quality, cost and time run at 8% to 11% against expectations of 14%. Agentic AI is early: an estimated 27% are testing it, and 43% cite unclear ownership as a barrier. Nearly 64% call an overcrowded stack a key barrier, and only 23% say sales and marketing share common targets.

What marketing teams should take from it

  • Who pays and who sets priorities are different questions. The 55% IT-funding figure is the reason finding 03.1 asks marketing to put its priorities and its definition of done in writing with the company’s AI lead.
  • Realized gains of 8% to 11% are a sober baseline to set against vendor promises before promising a return to the CFO.
  • The 43% who cannot say who owns agentic initiatives, and the 23% with shared sales targets, are two ownership gaps a CMO can close by decision rather than by technology.
  • The sample is large companies only, so mid-market teams should read it as direction rather than benchmark.

Where to start

Bring two of its numbers to the planning conversation with your company’s AI lead. 55% of marketing AI is paid for by IT, and 43% cannot say who owns agentic work. Then put a name against each of your three priority tasks, as finding 03.1 sets out.

Use the 23% shared-target figure to open a conversation with the head of sales, starting from the account brief both teams need.

Keep in mind

A consultancy report whose recommendations map to its own services. Year-on-year comparisons use different samples, and several headline figures count only the strongest answer on a five-point scale.

Where the report uses it