AI changes the economics of marketing services. Production and versioning get cheaper; ideas, strategy, craft and orchestration keep their value. Your agency model should reflect that.
Where things stand
- The ANA reported in May 2023 that 82% of its members have an in-house agency, up from 78% in 2018 and 42% in 2008.
- Klarna said in 2024 that AI helped it cut spending on external marketing suppliers (translation, production, CRM and social agencies) by 25%.
- Mondelez built its generative AI content tool together with an agency group (Publicis Groupe) and an IT firm (Accenture) — a sign that the answer is often a new kind of partnership, not "in-house vs agency".
Sort the work, then decide the owner
| Work type | Often best owned by |
|---|---|
| Brand strategy, big ideas, hero craft | Agency partner or senior in-house creative |
| Always-on content, versioning, localisation | In-house team with AI tools |
| Product imagery and catalogue content | In-house or specialist, from a product source of truth |
| Media buying and optimisation | In-house, agency or platform automation — with independent measurement |
| AI platforms and workflows | In-house marketing ops with IT |
Contract terms that matter now
- Disclosure of AI use in deliverables.
- Approved tools and data limits (no client data in unapproved tools).
- IP warranties: rights to AI-assisted output; no infringing material.
- Pricing: how AI efficiency changes fees (output- or outcome-based rather than hours).
- Brand guardrails: agencies follow your voice guide, claims rules and disclosure policy.
Avoid
Moving production in-house without the ops capacity to run it; or paying agency hours for work AI now does in minutes.