P&G
Pantene reported more assets, faster production and lower cost in one example. Useful evidence, not a universal benchmark.
Research paper · September 2026
A practical guide for leaders shaping marketing, content and creative operations.
01
AI changes the economics of making content. The harder business question is whether additional capacity improves customer experience and profitable growth.
Faster generation creates little value when work waits for approval, misses a local cultural cue, uses an outdated product claim or never reaches an audience. The creative supply chain should operate as a business capability with an accountable owner, explicit service standards and a measurable cost to serve.
02
Announcements, demonstrations and operational use are different evidence states. The public record does not yet establish independently verified, enterprise-wide autonomous marketing.
Pantene reported more assets, faster production and lower cost in one example. Useful evidence, not a universal benchmark.
Lay's used AI-assisted localisation and reported improved ad recall in Benelux. The wider connected vision remains broader than this case.
A six-week North America test connected workflow, creation, review and asset distribution.
After a seven-market pilot, the company described wider adoption led mainly through agency users. Human publication authority remains.
A creator and paid-media integration brought evaluation into the workflow, with invite-only testing stated in the announcement.
CreAItech links internal content capability with external models and AI-enabled consumer journeys.
The 2026 CMO Survey reports mean generative-AI use of 22.42 percent of marketing activity time among 187 respondents to that question. It is self-reported US data, not the percentage of brands that have adopted AI successfully.
03
Operating model and architecture
A connected workflow needs stable identifiers for briefs and asset versions. It needs relationships among product, market, audience, rights, approvals, activation and performance. Without them, the organisation cannot explain which decision produced an outcome.
Central standards should cover identity, rights, lineage, approved facts and minimum quality. Local teams should retain authority over language, culture and market claims. Exceptions need an owner and expiry date.
Own the context, standards and decision rights. Buy generation capability when it is efficient, and preserve the ability to change providers.
04
An agent can prepare an adaptation and route an exception. Authority to publish, change a budget or contact a customer requires a separate decision.
Expose evidence, interpretation and missing inputs.
Generate against a deliberate hypothesis and approved context.
Test claims, rights, brand fit and destination requirements.
Attach authority to the exact version and intended use.
Return context-specific observations to the next brief.
05
Models and internal capability
A model may be strong at video motion and still be unsuitable for confidential inputs or a particular commercial use. Evaluation must use representative briefs and assets, including difficult languages, packaging, people, claims and negative cases.
Each approved model needs an owner, permitted tasks, deployment conditions, known limitations, a review date and a retirement path. Eligibility comes before price optimisation.
06
Brands can own business context and governance while agencies use the environment for ideas, craft and scaled production. Ownership of the platform and location of the work can diverge.
07
Earned content and participation
Brand-owned content, paid creator work, earned participation and synthetic creator-style content require different treatment. A voluntary customer video can become paid distribution after the brand obtains permission. Reporting should retain both facts.
AI can assist discovery, language support, moderation and rights administration. It should not invent customer experience or infer unrestricted reuse from public availability.
08
Brief to approved, including waiting and rework
Approved assets activated for a defined need
Fully allocated cost per accepted output
Compliance and distinctiveness measured separately
Repeat workflow completion among eligible users
Incidents, near misses and failed controls
Incremental contribution with relevant controls
09
Select one repeatable workflow. Map real work, rights, costs, queues and quality failures. Agree the evidence required for the next decision.
Run representative and difficult cases. Include rejected approvals, expired rights and model failure. Measure repair as well as success.
Scale only if output quality, controls, user behaviour and economics justify a production pilot. Assign service ownership first.
The durable advantage lies in the quality of an organisation's decisions and its ability to learn from real work.
Original publication by Mayur ChaudharyThe research window is 9 January to 9 September 2026. Company and vendor accounts establish stated direction and reported activity, not independent verification. The full paper contains the methodology, limitations, detailed recommendations and source list.