Full analysis

Decision Council: Coca-Cola's Data-First Agency Review

Step 1 — Frame

The implication: Coca-Cola is running a global media/data/tech review (ex-North America, ex-Japan/Korea) where the deciding criterion is no longer "who buys media well" but "whose data architecture lets us fuse our first-party data with bottler and partner data." That makes this a referendum on two opposing models for handling customer data: Publicis's centralize-and-own stack (Epsilon, Lotame, LiveRamp, CoreAI) versus WPP's connect-without-moving approach (InfoSum's clean-room-style data collaboration, where two parties match data without either copying the other's).

For an ad-tech operator reading this, the real question is: which data model is winning the trust of the world's biggest advertisers — and what does that mean for where ad dollars, and data leverage, flow next?

• Reversibility: Type 1 for the holdcos (multi-year account, validates or undercuts a billion-dollar acquisition strategy). Type 2-ish for Coke (they can re-review).
• What's actually being decided: Not "which agency." It's "should an advertiser hand its first-party data into an agency-owned stack, or keep it in its own house and connect on demand." That choice ripples across the whole industry.
• Forcing function: Decision expected fall 2025; coincides with Publicis closing its LiveRamp deal.

Step 2 — The Council

The Market Analyst This is a proxy war for two acquisition bets. Publicis spent years and billions buying its way to an owned data stack — Epsilon, then the pending LiveRamp deal (the connective tissue that links one company's customer records to another's). A Coke win is the marquee proof point that strategy needed. WPP, growing slowly and under pressure, bet lighter via InfoSum partnership rather than ownership. For investors watching the holdcos, this is a live test of "buy the data assets" vs. "partner for them." In plain terms: two ad giants are betting their stock stories on opposite answers to the same question, and Coke is the judge.

The Skeptic The load-bearing assumption is that data architecture actually decides this. It rarely does. Pitches are won on relationships, price, and which holdco has the CFO's ear — and Murphy's "new intelligence" quote may be a CFO repeating a vendor's slide back to him. Second: "owned stack" vs. "data collaboration" is a false binary. Publicis runs clean rooms too; WPP can centralize when it wants. The real differentiator is execution and trust, not philosophy. Watch whether this "data-first" framing is genuine evaluation criteria or post-hoc justification for a decision driven by economics. To a non-specialist: the stated reason for the contest may not be the real reason.

The Operator Forget the philosophy slides — what happens when someone tries to match Coke's data with a bottler's on a Tuesday? Coke's distribution is a franchise model: hundreds of independent bottlers, each with their own systems, contracts, and legal teams. Whichever model wins, the hard part isn't the matching tech — it's getting 200 bottlers to agree to share consumer engagement data under one governance framework. That's where this breaks at 90 days. The model that lowers the legal and IT friction for each new data partner wins in practice, regardless of which is more elegant on paper. To a non-specialist: the demo is easy; getting hundreds of partners to actually plug in is the nightmare.

The Customer / End User (the advertiser) Coke is telling the whole CMO/CFO world something: advertisers now want to keep their first-party data as an owned asset and decide case-by-case how to share it. That's the through-line in Murphy's quote — "it's proprietary, it's ours." Big advertisers increasingly distrust handing raw data into anyone's centralized stack, agency or platform. This favors models that let the advertiser stay in control. For ad-tech vendors, the lesson: the buyer wants leverage and portability, not lock-in. Anyone whose pitch requires the client to surrender data ownership is fighting the current.

The CFO The real cost isn't the media fee — it's switching costs and lock-in. If Coke pours its data into a holdco-owned stack, it gets convenience now and a dependency later: leaving means rebuilding. The collaboration model costs more in coordination but preserves exit options. For Coke's own CFO, framing this as "intelligence creation" is also a hedge — data partnerships justify spend that pure media buying can't. For ad-tech operators: the value is migrating from the media transaction to the data layer, and margin will follow. Whoever owns the matching layer owns the renewal.

Step 3 — The Tensions

1. Does architecture actually decide this (Analyst) or is it a cover story (Skeptic)? The Analyst treats the data-model contest as the real fight; the Skeptic thinks relationships and price decide it and the data talk is theater. If the Skeptic is right, the "validation" narrative for either holdco's acquisition strategy is hollow.

2. Centralize vs. collaborate — and who benefits from each (Customer vs. CFO vs. Operator). The Customer and CFO say advertisers want control and exit options, favoring the connect-don't-move model. But the Operator notes the centralized model may have lower per-partner friction once it's built. Control and convenience pull opposite directions.

3. What gets validated. A Publicis win says "advertisers will pour data into an agency-owned stack" — bullish for the buy-the-assets thesis and the LiveRamp deal. A WPP win says "advertisers keep data in-house and connect selectively" — bullish for clean-room/collaboration vendors and bad for the lock-in playbook. The whole ecosystem reads the result differently.

Step 4 — Synthesis

What this hinges on: Whether large advertisers, going forward, are willing to centralize their first-party data inside an agency's owned stack, or whether they insist on keeping it in-house and connecting it on demand. Murphy's "it's ours" language tips toward the second — but Coke could still pick Publicis's owned stack if the execution and price are better. The stated philosophy and the eventual choice may not match, which is exactly the Skeptic's warning.

Where the council leans: Toward the data-collaboration / keep-control direction as the structural trend, even if Publicis wins this specific account. The advertiser's stated preference (own the data, connect selectively) is the durable signal. The winner of one pitch matters less to operators than the fact that the deciding criterion has openly shifted from media-buying skill to data architecture. That's the headline for the industry.

What this means for ad-tech operators — by stakeholder:

• Identity & clean-room vendors (LiveRamp, InfoSum, ID5, InfoSum-likes): Both models need a matching layer. You win either way — but the collaboration model is where advertiser sentiment is heading, so "connect without copying" is the stronger pitch.
• Holdcos / agencies: The "we own the data stack" pitch is now openly contestable at the highest level. If owned-stack loses here, expect more advertisers to demand portability clauses.
• DSPs/SSPs and publishers: As advertisers fuse first-party data directly with partners, more value moves into private data deals and away from the open auction. Build for direct, data-rich connections.
• Measurement vendors: "New intelligence" means advertisers want closed-loop outcomes from their own data. Whoever proves lift from matched data wins budget.

What to verify before drawing conclusions:

1. Is data architecture a real scorecard item or post-hoc framing? Watch the RFP criteria, not the CFO soundbite.
2. How does the bottler-data problem get solved? The model that lowers per-partner friction is the practical winner — and reveals which approach actually scales.
3. Does the LiveRamp close change Publicis's pitch mid-flight?

My view: The durable story isn't Publicis vs. WPP — it's that the world's most iconic advertiser just told the industry that data architecture, not media buying, is now the thing it's shopping for. Every operator whose value sits in the media transaction should treat that as a warning, and everyone in the matching/identity/clean-room layer should treat it as a tailwind. Discount the "which holdco won" narrative; bank the shift in the buying criterion.

What did we miss? Is there a persona we should add for this specific decision? A General Counsel lens might earn its place here — the centralize-vs-collaborate choice is fundamentally a privacy-liability question, and which model survives regulatory scrutiny (state privacy laws, EU rules) across all those bottler relationships could quietly decide the whole thing.