Truthset put a number on something a lot of us have suspected for years: roughly 40% of every dollar spent in the open CTV programmatic market is wasted on inaccurate identity data. That works out to about $7.4 billion in lost U.S. CTV spend in 2026 alone. IP to postal linkages, the backbone of most CTV household targeting, are accurate only 13% of the time. IP to email sits at 16%. Even email to postal, which most of us treat as a reasonably solid match, is only accurate about half the time.
Adstra responded to this with ID Connection Strength, a five dimension scoring model that rates the reliability of a HEM, a UID2, a MAID, a CTV ID, or an IP before activation. It’s a useful diagnostic. Buyers finally get a visible risk score instead of buying blind. But a score tells you how bad a match probably is. It doesn’t make the match correct. The industry has spent the last several months building better instruments to measure a problem that BDEX built its graph to avoid in the first place.
The Root Cause Is the Household IP, Not the Score
Every identity signal in CTV eventually has to resolve down to a household, and the household is represented by an IP address. That’s the choke point. If the IP to household mapping is wrong, every identifier stitched to it inherits the error, no matter how clean the CTV ID or HEM looked upstream. This is why probabilistic graphs keep failing in public. They infer household from co-occurrence and recency, then hope the inference holds. Most of the time, per Truthset, it doesn’t.
BDEX approaches this from the other direction. Instead of inferring a household IP from behavioral proximity, our identity graph deterministically maps and validates CTV IDs, MAIDs, and HEMs directly against verified household IP addresses. Each identifier in the graph is tied back to a real, confirmed household connection rather than a statistical guess about which devices probably share a roof. When a CTV ID shows up in a bid request, it’s not being matched against a modeled household. It’s being checked against one that’s already been validated.
Why Mapping Three Identifier Types to One Household Matters
CTV IDs, MAIDs, and HEMs each carry different signal strength and different failure modes. A CTV ID tells you about a device sitting on a specific connection. A MAID tells you about a mobile device that may or may not be in that same household at a given moment. A HEM tells you a real person opted into a real email relationship somewhere. Individually, none of them is sufficient to make a confident household call. That’s exactly why the industry’s probabilistic models blend them loosely and accept the error rate as a cost of doing business.
Validating all three against the same household IP, rather than against each other, changes the shape of the problem. Instead of asking “do these two device signals probably belong to the same person,” the question becomes “does this identifier resolve to a household we’ve already confirmed.” That’s a verification check, not an inference. It’s the difference between a probabilistic hop and a deterministic one, and it’s why a validated graph doesn’t need a connection strength score in the same way an unverified one does. The confidence is built into how the identifier got into the graph, not bolted on afterward as a rating.
This also closes a lot of the room fraudulent and synthetic IDs use to operate. Spoofed device signals and fabricated identifiers tend to fail quietly when they’re checked against a validated household IP, because there’s no real household connection behind them to confirm. Filtering happens as a byproduct of validation, not as a separate cleanup step run after the damage is done to a campaign report.
What This Looks Like in Practice
For a buyer, the practical difference shows up before activation, not after a post campaign audit. A validated household IP means a CTV ID reaching that address has already cleared a real world check, not a probability threshold. It means cross device identity, connecting a CTV ID to the MAID and HEM of the actual people in that household, is built on the same verified foundation rather than three separate probabilistic guesses stacked on top of each other. And it means custom audience building and retargeting off that graph start from households that are confirmed to exist and be reachable, not households a model believes are likely.
The $7.4 billion figure is a fair estimate of what unverified identity resolution costs the open CTV market this year. It’s less a story about CTV specifically and more a story about what happens when an entire supply chain builds on top of inference instead of validation. The fix was never going to be a better score on a bad match. It’s mapping every identifier back to a household IP that’s actually been confirmed, and that’s the graph BDEX has been building the whole time.
