Bring the work that creates knowledge and advantage in-house
Full internalisation is neither realistic nor desirable. The useful question is to distinguish what produces a cumulative advantage for the company from what can be mobilized outside.
The strategic asset is not just the data: it is the loop that transforms each marketing action into a new owner learning.
Internalisation becomes selective
The observed movement is not a shift from the outside to the inside. It is hybrid, and it plays arbitration. Of the marketing leaders interviewed in the AAR and ISBA study, 44% build media capabilities internally, while 41% consider in parallel removing or re-externalizing certain internal structures.
AAR and ISBA · Evolution of the marketing operating model · 2025 · Marketing executives interviewed — View source
In terms of budgets, Gartner notes that 39% of the CMOs surveyed plan to cut their agency budgets and 22% say that the generative AI already allows them to reduce their dependency on certain creative or strategic tasks.
McKinsey · Modern marketing: what it is, what it isn't · 2025 · CMOs questioned — concerns certain tasks, not the whole agency relationship — View source
The conclusion to be drawn from this is not that the agencies disappear, but that The border between internal and external is being redesigned. Agencies, specialists, suppliers, platforms and external models continue to intervene. What the company seeks to master is the layer that produces a cumulative advantage.
What must remain under control
Three families of active people are distinguished by their effect over time.
Proprietary context
Customer data, brand knowledge, products, audiences.
Proprietary methods
Rules, workflows, decisions, validations.
Proprietary learning
Results, tests, performances, lessons learned.
It is these assets that must enrich the next marketing cycle. The rest — computing power, models, point execution, leading edge expertise — can legitimately come from outside.
The learning loop is the central asset
The real active owner is not just the data. It's the loop that turns every marketing action into reusable learning.
- 01Data
- 02Understand
- 03Decision
- 04Create
- 05Activate
- 06Measure
- 07Learn
Each cycle enriches the context of the following: the loop closes and leaves.
If this loop belongs to several actors without common memory, the learning is fragmented. Each campaign recreates part of the knowledge already produced elsewhere, and the company pays several times the same knowledge without ever capitalizing it.
Gains exist, integration remains difficult
Internal agencies report real gains. Among those in charge of internal agencies interviewed by WFA and MediaSense, the internalisation of high volume content is associated with 40% faster production, 33% efficiency gains and 93% intention to increase AI investments.
Reuters · WPP cuts out agency help as brands create their own ads with AI · 2025 · heads of internal agencies surveyed — declaratory figures — View source
The same corpus shows the other side: only 12% of the internal agencies interviewed consider AI to be fully integrated, and 36% consider integration into business systems still difficult.
WFA · How AI is changing in-house agencies · 2025 · internal agencies surveyed — View source
Retrieving the content and data is therefore also tantamount to resuming the operational complexity that accompanies them. Without a system to absorb it, internalization moves the load without reducing it.
To regain control means to govern usages
This mastery has a counterpart: you need to know who accesses what, with what rights and for what use. The use of AI tools outside the validated perimeter — the shadow AI — moving business data to uncontrolled services; access rules, confidentiality, GDPR compliance and intellectual property issues become parameters of enforcement, not just legal matters.
NIST · AI Risk Management Framework — Generative AI Profile · 2024 · risk management framework, institutional reference — View source
A precise vocabulary is required here. To talk about "retaining ownership of its data" is imprecise when external suppliers intervene: the exact legal regime depends on contracts. maintain the necessary control, governance and rights over its data, rules, workflows and results.
Neither fully outsourced nor fully in-house
The right model is therefore neither total outsourcing nor full internalisation. It consists of keeping in the company what produces cumulative learning, while mobilising agencies, specialists and suppliers when they bring additional expertise or capacity.
It remains to be seen how these internal and external resources, models, data and tools can work together. This is the subject of the last article.
- 01Production and validation
- 02Reliability of models
- 03Consistency and brand risk
- 04Internalisation and proprietary knowledge
- 05Orchestration and governance