Clean rooms got sold to retail as the grown-up answer to signal loss. Cookies were dying, iOS was clamping down, regulators were circling. The pitch was clean: a neutral, privacy-safe room where you and your retail media partner match data without either side seeing the other’s raw records.
The pitch worked. According to the 2025 State of Retail Media report, 66% of organizations now use clean rooms in some capacity, driven by the need for privacy-safe collaboration. The market is projected to sprint from a few billion dollars to nearly nineteen by 2034.
Here is the part nobody puts on the slide. For a growing set of retail use cases, the clean room is where your first-party data goes to die.
The math nobody wants on the board deck
Start with the money. The average company spends around $879,000 on a data clean room, per a Funnel.io survey of implementors. That is before you have activated a single audience or answered a single question.
Now the output. 39% of organizations struggle to drive actionable insights from clean room data, according to the 2025 State of Retail Media report. Read that again. Two in five buyers cannot get a usable answer out of the thing they paid nearly a million dollars for.
Then the match rate, which is the whole game. In retail media, identity resolution across fragmented consumer touchpoints delivers match rates typically ranging from 35% to 65% depending on data quality, governance and contracting complexity. So you spend the money, wait for the legal, and then match a third to two-thirds of your customers on a good day.
And the clean room will not save you from yourself. A DCR does not clean your data’s quality; it only maintains its privacy. If your customer records are full of duplicates or inconsistent formats, the match rates in the clean room will be too low to provide any statistical value. Garbage in, privacy-safe garbage out.
Latency is the tell
The deeper problem is architectural, and it is why this matters more every quarter. Clean rooms were built for a batch world. Early data clean rooms followed a centralized ‘bunker’ model. All participants were required to copy data into a neutral third-party environment for analysis. While straightforward in concept, this approach introduced significant friction. Data movement increased latency and cost, complicated legal and compliance agreements, and forced organizations to give up direct control of sensitive data.
Even the modern federated versions carry the tax. Some clean room setups struggle with low-latency data processing, potentially hampering real-time ad optimization. In plain terms: real-time performance can lag, and ad activation may be limited.
That was survivable when the only job was quarterly closed-loop attribution. It is fatal for what comes next. Agentic systems decide in milliseconds. Next best offer. Dynamic price. Inventory promise. Live personalization on a logged-in session. An environment that returns privacy-enforced aggregates on a delay cannot sit inside that loop. You cannot run a real-time agent on a batch answer that landed yesterday and only covered half your file.
So the first-party data goes in rich, live, and yours. It comes back capped, aggregated, and late. That is the death I am describing.
When it is the right tool, and when it is theater
I am not anti clean room. I am anti using a measurement instrument as an activation engine. Know the difference.
Use one when you genuinely have multiple parties and cannot see each other’s raw data. A clean room is no longer a crisis-response purchase. It is a strategic capability that pays off for organisations with enough media spend and enough first-party data to make the matching worthwhile, and a premature one for everyone else. The threshold is real: only at $2M and above, with genuine multi-publisher or retail-media complexity, does the setup cost of a neutral platform start to look defensible.
Call it theater when you are joining data you already own. If you are joining two datasets that you already own, a standard data warehouse or CDP is faster and more flexible. You don’t need a neutral room when you are the only party involved. Buying one to fix data hygiene you never addressed is theater too.
Here is the Cost of Doing Nothing. Every quarter you funnel first-party data through a batch clean room instead of a live decisioning layer, your agents run on stale, capped, aggregated signal while the competition’s run on the real asset. The gap does not stay flat. It compounds.
So separate the two jobs. Measurement across parties: clean room, above scale, with clean inputs. Real-time decisions on data you own: a live layer your agents can actually reach. The retailers who win the next cycle will stop treating one tool as both. They will keep the clean room for the ledger and give the agents the live wire. First-party data is too expensive to send somewhere to die.