What's Your First-Party Data Worth Now?

(Was this newsletter forwarded to you? Sign up here.)

The third-party cookie was supposed to be dead by now. It isn't.

In July 2024, Google reversed course on killing it. By April 2025 it had dropped even the consent prompt, and by that October it wound down most of the Privacy Sandbox replacement APIs it had spent four years building. Cookies still load in Chrome by default. The thing the entire industry braced for never arrived.

Now the irony. We spent those four years building something to replace the cookie. And that something, your first-party data, is the asset actually at risk, and almost nobody saw the threat coming from where it did.

• During the cookie panic, marketers stood up customer data platforms en masse. 67% of organizations onboarded a CDP, but only 17% reported high utilization, and average use of a CDP's capabilities slid to 47% from 55% (Gartner, 2024).

• The CDP category is consolidating. The independents got absorbed, and the survivors are rebranding as "agentic" decisioning platforms.

• Meanwhile, the buyer is starting to send a proxy. When an agent transacts on the Agentic Commerce Protocol, you stay the merchant of record, but the protocol passes you only a name, an email, and a phone number. Not the browsing. Not the intent. Not the deliberation.

The threat was never the cookie

The threat to your customer data was never the cookie. It's been intermediation all along, and we spent the cookie years pointing the fire extinguisher at the wrong room.

Think about what first-party data actually is for most brands. Strip away the CDP marketing and it's mostly observed behavior: what someone looked at, searched, clicked, abandoned, came back for. That observation is only valuable because you were the one watching. Put an agent between the buyer and your store and you are no longer the one watching. The agent ran the search, read the reviews, compared the options, and weighed the buyer's stated preference against the field. It saw all of it. You see a fulfilled order with a name attached.

Richard Crone calls the agent's vantage "the power position in agentic commerce," because large language models "see intent signals, buyer reasoning and infinite shopper-deliberation histories before checkout." The agent isn't taking your sale. It's taking your view of the customer, which was the actual asset.

This is already happening. Generative-AI referral traffic to US retail sites surged 693% year over year over the 2025 holidays. The deliberation is moving upstream into the model whether or not the final checkout happens on your site.

The Interception Test

Run every data asset you have through one question: could an agent have intercepted this signal upstream? The answer sorts your data into three tiers, and the tiers are moving in different directions.

Tier one is intercepted data. Observed behavior, browsing, search, intent. The agent now sees this first, and increasingly only the agent sees it. This is the bulk of what most stacks call "first-party data," and it depreciates a little every quarter the agents grow.

Tier two is transactional data. The order itself. You still get it, because you fulfill it, but agentic checkout strips it to the minimum: who and where, not why or what next. It's real, and it's thin.

Tier three is volunteered data. What a customer chooses to tell you directly: preferences, a profile they built, a subscription they manage, a community they post in. Forrester named this zero-party data years ago. An agent can't intercept it, because it isn't observed, it's given, and it's given specifically to you. This is the only tier that appreciates as the agents spread.

Most CDP budgets were built to industrialize tier one. The collection got cheaper and the asset got more fragile at the same time.

But agent checkout already flopped

The fair pushback: agent checkout already stumbled. OpenAI pulled back Instant Checkout in March 2026 after fewer than thirty merchants went live, and answer engines went back to referring shoppers to your site. So maybe the signal stays with you after all.

Read it the other way. The cookie just taught us that a data asset can be devalued by a platform decision you don't control, on a timeline you don't set. Whether checkout happens in the chat or on your site is a detail. The interception happens at discovery, and that's already live. Betting your customer relationship on the agent staying dumb at the last step is not a plan. Bain's read is blunter: agents could "turn direct-to-consumer brands into indirect ones," becoming the marketplaces that "control transactions, compress margins, and capture data."

Spend on what the agent can't take

Audit your data with the interception test and get an honest number. What share of what you call first-party data is really tier one, observed behavior an agent will soon see before you do? For most teams the honest answer is uncomfortable, and you can't reallocate against a number you've never run.

Shift spend from collection to volunteering. Every dollar going into capturing more behavioral exhaust is buying a depreciating asset. The same dollar spent giving customers a reason to tell you something directly, a profile worth maintaining, a membership worth having, a community worth posting in, buys the one tier the agent can't touch.

Stop calling the CDP your moat in front of the CEO. The moat was never the database. It was the relationship that fed it. When the agent intercepts the behavior, the brands left standing are the ones a customer chose to stay close to on purpose. That's the same argument that holds when AI makes everything else cheap, and it's where the real defensibility lives.

The cookie's death was a fire drill. We evacuated the building, stood in the parking lot for four years, and built a new wing while we waited. The fire started somewhere else.