IP addresses and postal/email addresses are a match made in … not heaven.
According to a new report, IP-to-email matches are accurate about 16% of the time on average, while IP-to-postal matches are accurate only 13% of the time.
Not great news for advertisers, who rely on these linkages to connect billions in digital ad spend with real consumers and measure the results.
Data validation provider Truthset conducted the study on behalf of the Coalition for Innovative Media Measurement (CIMM) and Go Addressable, an addressable TV trade org founded by AMC Networks, Comcast Advertising, DirecTV Advertising, DISH Media and Spectrum Reach.
“If we don’t figure this out,” said Scott McKinley, CEO and founder of Truthset, “the walled gardens, which don’t have this problem, will just take the audiences and all the money that comes with them.”
‘Sold a pup’?
An IP address is a unique numerical identifier assigned to a device’s connection on the internet.
Advertisers use IP addresses for targeting and measurement by associating online behaviors and ad delivery with specific households or business locations.
In connected TV advertising, IP addresses are especially important because they link devices within the same household so marketers can target and measure ads across multiple devices sharing the same IP.
Too bad they’re so unreliable.
“We have a profound anxiety that marketers are, in a sense, being sold a bit of a pup,” said Jon Watts, managing director of CIMM.
AdExchanger Daily
Get our editors’ roundup delivered to your inbox every weekday.
Daily Roundup
To “sell someone a pup” is a UK idiom that means to swindle or trick someone by selling them something worthless or of far less value than promised. (The more you know.)
“There’s this promise in the industry of completely accurate, deterministic data that matches sophisticated and vast advertising targets and attributes onto individuals and households,” Watts said. “This leads marketers to believe that when they buy against a particular target, they’re reaching that target – and, unsurprisingly, the truth is much more complicated than that.”
Bad connections
Truthset analyzed records from six major data vendors, including 1 billion email addresses, nearly 250 million IP addresses and 164 million postal addresses.
To assess the accuracy of each vendor’s data, Truthset matched it to verified email and postal records it got from two internet service providers and one large multichannel video programming distributor over a 90-day period between December 2024 and February of this year.
Truthset’s head of data science, Kathryn Barnitt, and her team applied Bayesian statistical techniques and machine learning to estimate the accuracy of each match.
They looked for patterns in what made a data match likely to be correct or incorrect, such as how often a link appeared across sources, how recent it was and whether it was seen in multiple states or ZIP codes.
Beyond the low match rates, they observed little overlap between vendors – more than 90% of linkages were unique to one provider – a sign that each vendor is using its own methods and criteria to make connections.
On top of that, they found that providers use inconsistent definitions and formats to timestamp their data, which makes it hard to determine how fresh or reliable it is.
There was also geographic variation. States with higher population density have slightly higher IP-to-postal linkage accuracy, while states with population centers on borders, like Rhode Island, Massachusetts and Connecticut, have lower accuracy.
And that’s not all.
Providers often assign too many IP addresses per household and overuse stale data to pad their data sets so they can offer broader reach and more matches to marketers, despite making the data less accurate. And despite increasing real-world use of IPv6, which is the most recent version of the internet protocol, IPv6s were noticeably underrepresented.
Upton Sinclair knows
Taken together, these findings paint a dire picture.
But the headline here shouldn’t be “IP address match rates are terrible!” The takeaway is more nuanced than that, Truthset’s McKinley said.
“Matching is a flexible, uncertain and black-box process whereby the data ecosystem attempts to match one thing to another,” he said, “The problem isn’t matching; it’s that advertisers and agencies ask for maximum match rates, because that’s how they deliver scale.”
When scale and price “are the only functional KPIs” on the buy side, he said, advertisers aren’t incentivized to care about accuracy.
Even brands that are vocal about fixing supply-chain problems aren’t immune to this mentality.
“You might have an executive shouting about it from the rooftops threatening to withhold $100 million this quarter or whatever it is, and then right below them is someone whose job is to turn a little wheel on a machine,” Watts said. “And then, from that moment on, they’re compensated not to ask questions, because every time they do, they might expose a problem that doesn’t have a fix.”
Basically, Upton Sinclair knew what he was talking about when he said, “It is difficult to get a man to understand something when his salary depends on his not understanding it.”
An existential problem
What’s the solution here?
Standards, education and awareness.
“This is actually a business problem more than it’s a data or tech problem,” McKinley said. “The industry can be really bad about leaving their knives at the door, but they need to collaborate if they want a chance at combating the walled gardens.”
A rising tide and all that.
“It’s obvious to me as a data dork that this can’t be solved by just one company,” Barnitt said. “It’s going to take an industrywide push toward standardization.”
At the same time, buyers need to educate themselves and stop taking match rates at face value.
For example, there was a wide variation in accuracy between providers, Barnitt said. Although match rates are low on average, some partners are better than others. The most accurate provider was 4.5x better than the least.
Meanwhile data providers should be able to answer questions like: Are you collecting IPv6s and do you include them in the data you sell? How do you time-stamp your data? How often do you refresh it? What’s the lookback window for establishing linkages?
But getting this right isn’t just a technical exercise.
“If we as the collective open internet don’t figure this out, all of the power will go to the tech giants who don’t have any of these problems,” McKinley said. “It’s existential.”
