You cannot acquire accountability.

Publicis is acquiring LiveRamp for $2.2 billion, and the marketing industry has already decided what the story is. The story, apparently, is neutrality. For nearly two decades LiveRamp held the one position almost nobody else in the ecosystem could occupy: not a media owner, not a publisher, not a retailer, not an agency. Just the infrastructure everyone used and nobody owned. Switzerland. The rails. The plumbing. And now a holding company owns the plumbing, so the conversation everyone is having is whether neutral still means neutral.

It is a fair conversation. It is also the small version of the story.

Martha Peterson has written the sharpest piece I have read on the deal, and she reaches the part that actually matters. But she reaches it last, in the final section, which is exactly where most readers stop paying attention. So I want to take her closing point and make it the opening one. Because for anyone running commercial strategy in a regulated industry, it is not a closing thought. It is the whole thing.

What actually got sold

Let me deal with the neutrality argument quickly, because it is true and it is already over.

I spent a large part of my career inside the holding-company machine: building one of its businesses, then running a substantial part of another. “The rails belong to no one” was never a fact about the infrastructure. It was a position the infrastructure took, and it held that position for exactly as long as neutrality was the most profitable thing to sell. The moment owning the rails was worth more than renting out neutrality, the rails were going to be bought. $2.2 billion is simply the number at which the pretence stopped being worth maintaining.

This is not cynicism. It is how the category has always worked, and pretending otherwise is how buyers get surprised. Brands will respond the way they always do. They will route around it: Snowflake, Databricks, the hyperscaler clean rooms, InfoSum's decentralised model, the verticalised healthcare data environments. The plumbing will reshuffle, governance will move a little closer to internal IT, and within a few years there will be a new piece of infrastructure that everyone uses and nobody owns, right up until somebody does.

That entire cycle is a distraction. Not unimportant: your data residency and procurement teams will rightly spend the next two quarters on it. But it is a distraction from the line in Peterson's piece that actually changes what your business should be doing.

The line that matters

Here it is, in her words: stop owning pipes, start owning decisions.

Her argument runs like this. Identity is becoming infrastructure. Infrastructure commoditises. The competitive advantage was never really in connecting the data. It was in what you decided to do with it once connected. The organisations that win the next decade will not be the ones that own the identity graph. They will be the ones that own the decision engine: the system that ingests first-party data, media exposure, transactions, claims, search behaviour and market signals, and then makes real-time calls about budget allocation, creative rotation, audience prioritisation, channel sequencing, frequency and incrementality.

She is right. I want to be precise about that. The diagnosis is correct, and it is the most useful thing in the piece. Value is migrating from the connection layer to the decision layer, and it is not migrating back.

But there is a step missing between her diagnosis and what a pharma leader should actually do about it. That missing step is the entire reason I am writing this rather than simply resharing her post.

A decision is not an optimisation

Look again at the list of things a decision engine does: allocate budget, rotate creative, prioritise an audience, sequence a channel, manage frequency, suppress an exposure.

In consumer marketing, those are optimisation problems. You are trading efficiency against reach, cost against conversion, and the worst realistic outcome of a bad call is a wasted spend and an awkward quarterly review. The system can be aggressive, because the cost of it being wrong is money.

Now move the identical list into pharma. The decision engine prioritises one healthcare professional segment over another. It sequences a therapy-area message in a particular order. It suppresses exposure against a de-identified patient cohort because the model has decided that cohort is unlikely to convert. Each of those is structurally the same operation as its consumer equivalent. None of them is the same kind of act.

Suppressing exposure to a cohort in consumer marketing is a frequency-cap decision. Suppressing exposure to a de-identified patient cohort in pharma is a decision about who does and does not encounter information adjacent to their health. It is made by a model, at a scale and speed no human reviewed, on a logic no one in the commercial team could reconstruct on demand. That is not an optimisation. It is a governed act wearing the costume of an optimisation, and the costume is the dangerous part, because it lets the organisation file the decision as a performance metric instead of treating it as something a named person has to own.

This is the gap the consumer playbook gets to ignore and pharma does not. And it is precisely the gap I have been writing about from the clinical side.

The same argument, arriving from the other direction

I argued in The Named Owner that every consequential AI decision in a regulated setting needs a specific human being attached to it. Not a committee, not a policy. A name. That piece came at it from the clinical and coverage side: AI deciding what gets reviewed, what gets approved, what a patient is or is not eligible for. The regulatory world is already moving there. Iowa's Patients First Act writes the principle into statute: an AI-influenced coverage decision must have a human reviewer. It will not be the last law to do so.

What the Publicis-LiveRamp deal exposes is that the same argument is now arriving from the commercial side, and most pharma organisations have not noticed, because it is wearing a marketing-technology costume rather than a clinical one. The clinical function has at least started asking who owns the AI decision. The commercial function is buying decision engines on the strength of a consumer-marketing thesis and has, for the most part, not asked the question at all.

The evidence that this gap is structural rather than hypothetical is not subtle. 78% of health systems plan to deploy clinical AI within 24 months. 23% have governance in place for it. And 78% of leaders concede they could not produce an auditable explanation of an AI-driven decision within 30 days if a regulator asked for one. That is not a clinical-AI problem. It is an organisational-design problem, and a decision engine bolted onto the commercial function inherits every part of it without inheriting any of the scrutiny the clinical side has started to apply.

Why the commercial side is the more exposed one

There is an asymmetry here worth naming directly, because it is counter-intuitive and it is where the real risk sits.

The instinct in pharma is that clinical AI is the high-stakes domain and commercial AI is the low-stakes one. Patients on one side, media plans on the other. That instinct is wrong, or at least dangerously incomplete. Clinical AI is high-stakes and the organisation knows it is high-stakes, which is exactly why it is now wrapped in review boards, regulatory attention and an emerging body of law. The scrutiny is catching up to the risk.

Commercial AI is consequential and the organisation has decided it is not. It runs on consumer-marketing assumptions, under marketing budgets, reviewed by people whose performance is measured on efficiency rather than on accountability. The risk is real and the scrutiny is absent. That gap, between how consequential the decisions are and how seriously the organisation treats them, is the single most exposed surface in a modern pharma commercial function, and the Publicis-LiveRamp deal has just made it cheaper and faster to expand.

You cannot acquire accountability

So here is where I part company with the “own the decision engine” conclusion. Not because it is wrong, but because it stops one move too early.

You can acquire a decision engine. You can build one, license one, or, as this very deal demonstrates, buy the company that makes one. The engine is a purchasable asset. That is precisely why it will commoditise, on the same curve and for the same reasons the identity graph is commoditising right now. Owning the engine is not the durable advantage the framing implies, because anything you can acquire, your competitor can acquire too, and usually within a quarter.

The thing that does not commoditise, the thing that cannot be acquired or licensed or bought in an all-stock deal, is accountability for what the engine decides.

A decision engine with a named owner is an asset. The same engine without one is not a cheaper version of the same asset. It is an unowned liability with a dashboard. It looks identical to the asset right up until the moment someone, a regulator or a journalist or a patient's lawyer, asks who decided this, and the only honest answer is that no one did, the system did, and no human can fully reconstruct why. At that moment the dashboard stops being a capability and becomes evidence.

This is the accountability turn, and it is the actual story of the Publicis-LiveRamp deal. Not who owns the rails. Who owns the decisions that will run on top of them once the rails have stopped mattering.

What this actually means for you

If you run commercial strategy in pharma, this deal should not send you into a clean-room procurement exercise, or at least not only that. It should send you toward a harder and more uncomfortable set of questions, and they are not IT's questions to answer. They are yours.

For every commercial decision your organisation has already automated, or is about to, there is one question that matters more than the technology underneath it: who is the named owner of that decision, and do they know they own it?

Then: could that person produce an auditable account of how the decision was reached, in language a regulator would accept, within 30 days? Not the data scientist who built the model. The business owner who is accountable for the outcome.

And then the hardest one: is that ownership designed into the system, or is it being assumed? Because assumed ownership is unowned ownership. A named owner who discovers they were the named owner only after something has gone wrong was never an owner. They were a fall guy the org chart had not got round to labelling.

If you cannot answer those three questions for every automated commercial decision in your business, you do not have a decision engine. You have an accountability vacuum that happens to produce media plans, and this deal has just made it cheaper and faster for that vacuum to scale.

The turn nobody is naming

The neutrality conversation will run for a few weeks and then exhaust itself, the way the third-party-cookie conversation did, and the walled-garden conversation before it. The infrastructure question always resolves the same way: it reshuffles, and the people who spent the cycle arguing about it move on to arguing about the next one.

The accountability question does not resolve that way, because it is not an infrastructure question. It is a question about whether, as decisions move from people to systems, your organisation can still produce the name of the human who owns each one. That question is not specific to this deal, or to marketing technology, or even to pharma. It is the through-line of this entire phase of AI adoption, and in pharma it is now arriving from two directions at once: the clinical side, which has started to ask it, and the commercial side, which has not.

Neutrality was the last decade's question.

Accountability is this one's.

The organisations that understand the difference will not be the ones that bought the best decision engine. They will be the ones that could always, instantly, and without rehearsal, tell you who owned the decision.

Next in this series: The Judgement Muscle. Why the organisations that delegate the most to AI are the ones that most need to keep human judgement in working order, and why that capability atrophies precisely when you stop noticing it is being used.