Why you still need your agency

As artificial intelligence fundamentally alters marketing economics, the traditional relationship between brands and media agencies is undergoing a strategic recalibration.

91% of marketing organisations now report using AI (Jasper). Meanwhile AI stack absorbs work that used to sit with the agency, marketing leaders are reconsidering what they buy and from whom.

Almost every major study points to the same three pressures caused by this shift:

• Proving marketing ROI.
• Getting data and AI to work across channels.
• Scaling AI without losing brand, quality, or compliance.

Those pressures do not just reshape internal decisions. They change why and how brands work with agencies.

And they explain why every major agency holding company has spent the last twelve months rewriting its own model. Each has either changed its CEO, restructured its legal entity, or announced a multi-hundred-million labour programme. Omnicom and IPG closed the largest agency merger ever. WPP swapped Mark Read for Microsoft's Cindy Rose, and her Elevate28 strategy puts the holding-company label up for retirement. Publicis signed a $2.2bn agreement for LiveRamp. Dentsu replaced its CEO and announced cutting 8% of international staff. Havas committed close to €1bn to AI.

Meanwhile the AI stack itself is absorbing pieces of the agency role. Salesforce launched Agentforce Marketing at Connections with Marketing Goals Agent, Content Agent and Slack/MCP campaign management. (Salesforce) Adobe released Brand Intelligence and the agentic content supply chain in GenStudio, with support for ChatGPT Ads.(Adobe) And Coca-Cola presented Fuel Light 360 with Bain as build partner.

What this means for brands, in my view, is that the agency relationship is being rewritten.

To remain viable partners, agencies must adapt to expectations around ROI, transparency, and governance.

I can see four shifts already taking shape:

Shift 1. From campaign execution to AI orchestration.

The first shift is about what the agency actually does for its retainer. CMOs are making governance central to how they scope work and measure partner performance. The biggest barrier to scaling AI in 2026 is legal, compliance, and brand review, named by 27% of marketers as the top reason they are not scaling AI faster (Jasper). Brands now expect partners to document verification protocols, brand-safety rules, and escalation paths for autonomous decisions. The unfamiliar but obvious new contract clause: which prompts, datasets, and channels are off-limits, and who reviews an agent's output before it goes live. The real opportunity for agencies here, I believe, is to design "content factories" with governance built into the AI pipeline.

Shift 2. From retainers for labour to retainers for efficiency gains.

The commercial model is being rewritten alongside the work. Among the small group of European marketing leaders who have reached gen-AI maturity (around 6%), efficiency gains come in above 20% (McKinsey). The budget data in 2026 forecasts reveals the pressure: 55% of marketers worry about budget cuts, and 46% say personnel and headcount are on the cut list, versus just 15% for MarTech and tools (Supermetrics). When something has to give, agency labour fees are first on the list. New contracts are being written with pass-through clauses, where efficiency gains from AI flow back to the client as lower fees or higher iteration counts. The old retainer paid for "hours spent". The new one pays for the "gain those hours produce".

Shift 3. From volume KPIs to value KPIs.

Measurement is moving away from execution metrics toward business outcomes. Among marketers who can measure AI ROI, 60% report at least 2x returns (Jasper). On the other hand, only 41% of marketers say they can measure the ROI of their AI investments today (Jasper). New contracts increasingly require agencies to demonstrate AI proficiency as a condition of partnership rather than a pitch-deck talking point.

Shift 4. From done-for-you to done-with-you.

The work agencies bring is moving from finished asset delivery toward capability transfer. 97% of marketers say access to AI tools factors into their job decisions, and 75% say it is critical when considering a role (Jasper). Talent goes where it can keep building. Agencies that want to stay in the room should run playbooks, training, and joint workflows with internal teams.

A four-step playbook to redesign how you work with agencies

Those four shifts raise an obvious question for marketing leaders: what do you actually do about them this quarter? My short answer is a four-step approach.

Step 1. Decide what should never be outsourced to AI and agencies.

The work that truly differentiates brands stays human. In my conversations, I draw a hard line around three things. Brand point of view and taste, because 61% of marketers say both matter more than ever when humans and AI work together (HubSpot). The value creation agenda, meaning where to grow, which customers to serve, what the brand stands for. And the AI risk appetite itself: which work you are willing to hand to AI, and which you are not.

Step 2. Match agency types to AI-era capabilities.

Most marketing leaders still hire against legacy boxes: creative, media, performance, social. If you look at the 2026 data, the hardest problems cluster into four other pillars:

• Systemic performance and ROI proof: 40% of marketers say proving ROI across channels is one of their toughest challenges (Supermetrics).
• Data and AI readiness: 37% struggle with the lack of integration between analytics and activation tools, and 52% say external teams define data strategy and measurement (Supermetrics).
• AI-driven content scale: 86.4% of marketing teams now use AI in at least a few areas of marketing, mostly content and media, but speed and quality at scale remain elusive (HubSpot).
• Governance and risk: legal, compliance and brand review are the top barriers to scaling AI.

Make each pillar's owner explicit in scopes and governance documents.

Step 3. Give your agency permission to own the AI operating model within their scope.

Speed is the gap most often missed: 29% of organisations still take 3 to 5 months to launch a multi-asset campaign, and only 12% can do it in days or hours (Jasper). In that clutter, brands end up with four providers all "doing AI" in disconnected workflows, and no one owns the connective tissue. Ask your agency to own three concrete things: how AI shows up in your workflows, how AI shows up in your metrics, how AI shows up in your governance. I would call this centralising responsibility.

Step 4. Change what you ask your agencies to prove.

Most organisations still measure AI in hours saved and vendor cost reductions. If that is what you ask for in the briefing, that is what you will get in the report. Three asks I would put in the next briefing session instead.

• Show us how AI-assisted campaigns changed time-to-market and time-to-learning versus our baseline.
• Show us where AI helped us stop spending: which channels, audiences or assets we now know are low-value.
• Show us how you used AI to inform decisions, not just to produce assets.

The next two years

The holding companies that solve the operating-model question will hold the strategic seat. The rest will be reclassified as media resellers. Either way, the work the agency does is moving up the value chain or out of the room.

The same question lives on the CMO side. A brand that keeps briefing for production volume in 2027 will be paying for an agency model that no longer makes economic sense.

👉 Share with one marketer still briefing for production volume.

← All insights