Technology

Data: an asset to be negotiated

9/23/2026

Post by

Nicolás Rasmunsen

Every month, someone on your team puts together the same report by hand: they gather numbers from two or three vendors, cross-reference them in a spreadsheet, and present them as if they were their own analysis. They aren't. It's the summary someone else decided to show you, calculated using a definition someone else decided to use. If that vendor changes its methodology tomorrow, your historical data breaks, and nobody asked you.

That's the symptom of a deeper shift. For twenty years, buying market data meant buying an analysis: a report produced once and sold to everyone who paid for the license. That model is losing ground. What the most sophisticated companies are asking for now is event-level data, in their own hands, with an analytical model designed for their business and no one else's. Value has shifted from the analysis to the data, and very few companies have adjusted the way they buy information accordingly.

‍

If you don't own the metric, you don't get a vote

The main problem with the packaged model is dependency. If a third party sets the definition of your metric, any methodological revision reaches you without warning and without any chance to object. That's exactly what just happened to the entire TV measurement industry in the United States: one of the leading audience measurement companies confirmed seven methodological changes that take effect at the end of the month and will collectively alter the estimates on which the season's advertising budgets are negotiated. Not a single brand buying on those numbers had a say in the decision.

The case is about TV, but the mechanism repeats itself in any industry where a company reports on a number it doesn't control: retail on consumer panels, channel on partner data, marketing on agency benchmarks. The question worth asking isn't whether the vendor is good. It's: what happens the day they change their criteria?

YouTube showed this twice in less than two years, in opposite directions. In August 2026, it began counting a public view from the very first frame, without the minimum number of seconds it previously required (YouTube, 2026): numbers rose overnight with no additional real audience behind them, and anyone comparing against their own historical data could no longer do so. A year earlier, the opposite happened: the platform started filtering out impressions it had been counting as views, and several channels saw their public view counts drop by half overnight, while revenue and real engagement didn't move. Whether numbers go up or down, the mechanism is the same, and itisn't exclusive to YouTube: Meta, Google, and the rest of the closed platformsdecide behind closed doors what counts as a click, a lead, or a view, with nopossibility of external audit. In the specific case of YouTube and Google Ads,the advertising industry's own log-file transparency registry gave them the lowestrating, precisely for not sharing that level of detail.

‍

Buyers stopped asking for reports and started asking for log-level data

The logical response to that dependency is to ask for raw data instead of the finished report, and it's already a consolidated trend among the largest buyers of digital media. Several of the most important programmatic buying platforms have launched event-level data products—every impression, with its full detail—designed for advertisers to build their own analysis instead of relyingon the vendor's standard report.

Today, the obstacle is contractual. An industry survey on access to log-level data found that, of 67 advertisers who showed initial interest in obtaining their own records, only 21 ended up getting them (AdExchanger, 2023). The most cited reason: most had no contractual right to access datagenerated by their own investment. Sometimes the clause is in the contract withthe agency; sometimes the vendor simply doesn't have a product to deliver it.

That's the point worth underlining: paying to generate data and having no right to see itisn't a technical problem. It's the first thing to check before assuming that"the data is yours" because you paid for it.

Advertiser selecting first-party data from digital marketing metrics

The architecture that made this possible

None of this would be viable at a reasonable cost if the underlying infrastructure hadn't changed. Three shifts explain it:

1. Data stopped moving to the vendor. The debate over whether customer data lives in the customer's own warehouse or in avendor's proprietary store has been settled, and the warehouse won. Identity,storage, and activation have been separated into pieces that get assembledrather than bought as a closed package.

2. Data is delivered as a product, with an owner. The most advanced organizations treat their data the same way they treat software:with an assigned owner, an explicit schema, and defined service levels.

3. The semantic layer is where differentiation lives. The problem is almost never a lack of metrics; it's that the same metric is defined differently in each department. Revenue has one formula in finance, another in sales, and a third in the marketing dashboard.

That third point explains the hidden cost of buying a packaged dataset: you're also buying someone else's taxonomy. A consumer study frequently cited in the branding industry illustrates this well: a beer brand and a well-known energy drink would have hit a ceiling had they accepted the obvious definition of their category—Belgian-style white beer, premium energy drink. Instead, they understood they were competing on much broader ground—a consumer might swap a beer for a margarita, or an energy drink for a coffee or an energy bar—and that insight allowed them to grow their market penetration in the United States to over 5% and 14% of the adult population, respectively (Bain& Company, via Koji, 2026). No category hierarchy in a syndicateddataset was going to show that: margaritas don't live in the beer category, norcoffee in the energy drink category. Seeing it required asking people what wasreally on their minds when they made a choice, and that's a question onlycustom research can answer.

‍

Why this became urgent now

All of the above was true five years ago, and nobody was rushing. What changed the pace is AI, which turned a data quality problem into a visible business bottleneck: a generic data model produces generic results, and that stops being competitive the day your competitors train on proprietary, governed data. Recent McKinsey research on AI agent adoption in companies shows the same pattern over and over: nearly two-thirds have already experimented with agents, but fewer than one in ten have scaled them to real use, and eight out of ten points to data limitations as the main obstacle (McKinsey,via Covasant, 2026).

‍

What this model doesn't solve

It's worth saying just as frankly, because building custom isn't free either. Three real costs:

• A 100% proprietary model is a sample of one: there'snothing to compare it against, and no number a third party is obligated toaccept. That's why syndicated data isn't going away; it remains the market'stransaction standard.

• Having raw data without the capacity to model itcreates a false sense of control. The industry itself acknowledges it:event-level data exists, but exploiting it requires technical resources mostcompanies still don't have.

• The contractual right to data doesn't come includedwhen you pay for media. It has to be negotiated, and reviewed before signing,not after.

The right modelisn't necessarily replacing everything syndicated with something proprietary.In the end, it's about knowing what's worth buying packaged—what needs to becomparable—and what to build custom—what needs to be exclusive.

‍

Three questions to ask your vendors this week

Do I have acontractual right to the event-level data generated by my own investment? Whodefines the metrics I make decisions on? Is the analysis I'm buying also beingsold to my competitors?

‍

At Portinos, we've been helping companies turn scattered data into decisions of their own, with models and dashboards designed for their business.

‍