- Published
- Mar 23, 2026
- Updated
- Nov 1, 2026
- Reading time
- 13 min read
Household Identity in Streaming: Measurement Without the Cookie Crutch
Measure linear and streaming overlap with consistent audience units, explicit denominators and a clear separation between new reach and incremental purchases.

In this guide (15 sections)
To find out whether linear TV and streaming reach different audiences, ask for a deduplicated exposure report with the same population, dates and identity unit. To find out whether the campaign caused additional purchases, define a separate outcome study. An audience can be new to your TV campaign while already knowing your brand; a first-time buyer can also have purchased without the ad.
This distinction sits behind a real advertiser question about linear and streaming overlap: traffic increased, but converters already knew the brand. The useful next step is to separate audience coverage, customer status and causal sales effects before choosing a measurement tool.
Why CTV Outgrew the Cookie Mental Model#
Third-party cookies never defined CTV the way they defined desktop display. In streaming, measurement leans on device identifiers (where platforms expose them), network-level signals such as IP, publisher and platform first-party data, and modeled or privacy-enhanced joins—including clean-room style collaborations—rather than a portable cross-site cookie.
That shift matters for reach, frequency, and outcomes: the “user” you count on CTV is often a device, a household inference, or a logged-in account—not a stable anonymous web profile.
Define the unit of analysis in writing before you debate numbers. “Household reach” and “device impressions” are not interchangeable—they are different questions with different error budgets.
Define the household before deduplicating exposure
Cross-screen reach depends on population, dates and identity unit. Discuss which exposure evidence is available for your campaign.
Discuss video measurementA short inquiry to scope relevant options; no campaign access is needed.
What “Household” Means When the Screen Is Shared#
On a living-room TV, co-viewing is normal: multiple people, one glass. A single impression may correspond to zero, one, or several attentive viewers. An impression does not establish the number of people present without a supported audience measurement method.
Household in practice is frequently inferred: IP stability, router or ISP characteristics, proprietary graphs, and platform IDs—not a single universal household registry buyers can audit end-to-end.
Three definitions teams confuse
| Term | What it often means in practice | What it does NOT guarantee |
|---|---|---|
| Device impression | An ad event tied to a device or app instance | A person; attention |
| Household exposure | A modeled or IP-clustered unit | Demographic truth; individual reach |
| Person-level reach | Requires a disclosed person-level measurement or estimation method | Proof that exposure caused a purchase |
Campaign reviews go sideways when finance uses household, sales uses device, and product uses modeled reach—all called “reach.”
Identity Signals: CTV vs Mobile Web#
| Dimension | CTV / OTT (typical) | Mobile web (typical) |
|---|---|---|
| Persistent bridge | Platform or app IDs where exposed; limited browser storage in embedded webviews | Third-party cookies (declining), first-party cookies |
| IP / network | High importance for geo, frequency, household-style models | Often combined with cookies or mobile IDs |
| Login / subscription | Strong when authenticated; weak for anonymous linear-style use | Strong on logged-in sites |
| Shared use | One screen, many viewers common | Personal device use more common |
| Measurement surface | Often SSAI, SDK, or server-side events | Pixels, tags, server-side conversions |
| Stability of graph | Tied to OEM and platform policy changes | Tied to browser and OS privacy changes |
The table is directional: every publisher and app differs. The point is to stop importing web assumptions into CTV planning without mapping the actual signal path.
An “Identity Ladder”: From Strong to Fragile#
When evaluating any identity or graph product, ask what rung you are standing on:
- Deterministic — Same stable ID observed across contexts (rare at scale in CTV; more common where login is universal).
- Publisher or platform scoped — Strong within one environment; weak across walled gardens.
- Probabilistic household — Statistical linkage with known false positive/negative rates—often opaque to buyers.
- Geo and network heuristics — Useful for coarse targeting; risky for precise frequency claims without disclosure.
Ask the vendor to document which identity method supports each report, including its inputs, validation and known gaps.
IP-Based Graphs: Strengths and Failure Modes#
IP address is widely used for geography and rough household clustering. It is not a perfect stand-in for identity:
- Rotation — Residential IPs change; mobile gateways and CGNAT collapse many users behind one address.
- VPNs and privacy tools — Shift apparent location and break simple geo rules.
- IPv6 vs IPv4 — Different density and stability characteristics by market.
- Shared Wi‑Fi — Coffee shops, dorms, and offices create false households if you over-cluster.
Frequency capping “per household” on IP alone can overstate or understate exposure relative to the intended household rule. Vendors compensate with probabilistic models and publisher-side caps, each with tradeoffs in accuracy and auditability.
When IP is “good enough”
- Coarse country or region checks, with the location method and its limitations disclosed.
- Rough pacing signals when you disclose uncertainty.
- Anomaly detection (e.g., impossible travel) when combined with other signals.
When IP is not enough
- Person-level competitive separation.
- Precise incremental reach without modeled panels.
- Legal jurisdictions where IP is treated as personal data—process must match counsel, not this article.
Clean Rooms and Privacy-Enhanced Collaboration#
Clean rooms (and similar controlled environments) are often pitched as ways to match advertiser and publisher data for reach, frequency, and outcomes without pooling raw rows in one database. They can work well when:
- Use cases and queries are tightly scoped
- Legal and technical controls align (purpose limitation, access logs, minimum thresholds)
- Definitions match: what counts as an impression, a completed view, or a conversion
They do not magically fix fragmentation: each platform’s graph and rules still constrain what can be joined.
When clean rooms are overkill
- Operational reporting that only needs aggregated delivery metrics.
- Early-stage tests where you have not aligned definitions across partners.
When they earn their cost
- Outcome and lift studies where cross-party joins are required under contract.
- Audited measurement programs where query logs and access controls matter.
Why Cross-Device Deduplication Is Still Hard#
- Different screens rarely share one durable, consent-aligned key at scale.
- Co-viewing breaks simple “one impression = one person” math.
- Platform policies change which identifiers exist and how long they persist.
- Walled gardens limit how independently measured events join across partners.
- Probabilistic matching introduces false positives and false negatives.
- Legal limits may forbid joins that are technically possible.
- Latency and definition drift between partners break naive ID reconciliation.
For the separate question of repeated exposure, use the cross-device frequency audit.
Measuring linear and streaming overlap without mixing outcomes#
Start with three separate questions. They require different evidence, even when the same campaign funds all three.
| Decision | Evidence to request | What it does not establish |
|---|---|---|
| Did streaming add households beyond linear? | Jointly measured reach, overlap and streaming-exclusive reach in a common eligible population | Whether those households were new customers |
| Did we acquire first-time customers? | Deduplicated customer/order records with a stated history window and a separate unknown-status category | Whether advertising caused those purchases |
| Did TV cause additional purchases? | An outcome study with a credible counterfactual, uncertainty and the intervention defined in advance | That all attributed purchases were incremental |
A buyer who purchased previously but was missing from the available customer history can be mislabeled new. Record the history window, account/household distinction, returns policy and unmatched share. “New to our dataset” is a narrower claim than “first-ever customer.”
Establish one exposure comparison
Request a report brief before requesting a percentage:
- Same country, target population, campaign window and timezone.
- Same unit: households compared with households, or persons compared with persons.
- A stated qualifying exposure rule and treatment of co-viewing.
- The linear and streaming sources actually included, plus excluded inventory.
- The linkage method, its validation, unmatched share and projection basis.
- The overlap numerator and each denominator, including any uncertainty.
The MRC cross-media video standards, sections 3.1–3.2 and 4.3.2, address consistent universes, coverage disclosure and empirically supported deduplication. IP alone is insufficient for audience-based unique estimates. These requirements explain why a matching date range alone cannot make a household estimate comparable with a person estimate.
When the report covers only linkable households, label that population. Do not silently treat unlinked observations as additional unique homes, or assume that they have the same overlap as the measured portion. Ask how any projection to the intended population was validated.
A hypothetical overlap calculation
Suppose one joint measurement process identifies the following sets of reached households in the same eligible population during a hypothetical four-week campaign. These are invented teaching figures, not AdProof customer results or an industry benchmark.
| Quantity | Reached households |
|---|---|
| Linear TV, including homes also reached by streaming | 1,000,000 |
| Streaming, including homes also reached by linear | 600,000 |
| Reached by both | 200,000 |
| Linear only | 800,000 |
| Streaming only | 400,000 |
| Deduplicated combined reach | 1,400,000 |
The set arithmetic is:
Combined reach = linear + streaming − both
= 1,000,000 + 600,000 − 200,000 = 1,400,000
Additional streaming reach beyond linear = streaming − both
= 400,000 households
The overlap is 20% of linear reach, 33.3% of streaming reach, or 14.3% of combined reach, rounded to one decimal where needed. Each percentage describes the same 200,000 homes with a different denominator. None is a universal expected overlap rate.
The 400,000 streaming-only homes are additional measured exposure relative to linear within this window. They are not automatically people who had never heard of the brand, first-time buyers or incremental purchases. The same report also cannot tell you how much reach would remain after changing the media plan.
Use this arithmetic to understand compatible sets. Independently modeled platform outputs may use different populations, projections and estimation procedures; forcing their totals into this table can create false precision. Obtain a documented joint estimate or retain the coverage gap.
For a platform-specific example, DV360's overlap documentation distinguishes pairwise overlap from exclusive and duplicate reach across its supported reporting dimensions. That is not evidence that the report includes an advertiser's external linear buy. Google also records a January 13, 2026 method update that may affect reruns and manual comparisons. Preserve the report's extraction date and methodology when comparing versions.
Move from audience coverage to purchase effects
Branded-search growth and traffic immediately after an airing can help identify periods to investigate. A timestamp match alone cannot rule out promotions, other media, returning customers or ordinary demand changes. Likewise, a brand study should be interpreted against the particular response it measured; its label does not turn it into a purchase experiment.
Google's Conversion Lift overview illustrates the causal question through treatment and control groups. Access is restricted to eligible accounts; it is not a promise of a ready-made test for every TV buy. Its geographic-study guidance also discusses feasibility and contamination when people are exposed in one region and convert in another.
Before commissioning a TV outcome study, agree on:
- Intervention: the whole linear-plus-streaming bundle, or streaming added to a linear baseline? These estimate different effects.
- Outcome: all completed purchases or first-time-customer purchases, using a fixed definition and deduplication rule.
- Comparison design: treatment assignment, control suitability, pre-period checks and how other marketing changes will be handled.
- Sensitivity: the smallest effect worth detecting, historical variation, feasible sample size, study duration and conversion delay.
- Readout: effect estimate, uncertainty, contamination checks and the decision for an inconclusive result.
Simply choosing two similar-looking regions and comparing their final totals does not complete that design. If experimentation is impractical, an observational model needs explicit assumptions. Google's Meridian methodology explains why good statistical fit alone does not validate causal effects.
For a noisy regional experiment, audit detectable lift, power and uncertainty before using a small point estimate as a fixed sales-lift assumption.
Keep the final review in separate rows: observed delivery, deduplicated reach, customer composition and estimated causal effect. An inconclusive sales result does not erase measured reach, and a large reach number does not resolve an inconclusive sales test.
Building a Measurement Contract (Template)#
Before you sign a multimarket deal, align on written answers:
- Unit of analysis — Device, household, person, or modeled reach—name it.
- Impression definition — Served, rendered, viewable, audibility if audio—per MRC or your disclosed standard.
- Deduplication window — 1-day, 7-day, campaign—per channel.
- Co-viewing policy — Ignored, adjusted with factors, or modeled—explicit.
- Known gaps — What the vendor cannot see (e.g., certain SSAI paths, certain app webviews).
This is not legal advice—it is operational hygiene that prevents QBR fights.
Privacy: Themes, Not Legal Advice#
Before linking audience data, have the responsible privacy and legal teams confirm the permitted purpose, data flows, retention and user-choice requirements for the relevant market. Aggregation alone should not be treated as approval to perform a proposed join.
Operational themes measurement teams should discuss with counsel:
- Purpose limitation — Collect what you need for stated analytics, not “everything for later.”
- Aggregation and thresholds — Reduce re-identification risk in outputs.
- Retention — Align raw event storage with audit and business needs.
- User rights — Opt-out, access, and correction where applicable by role and jurisdiction.
APAC and Indian OTT: Structural Realities#
Markets like India and broader APAC often combine:
- Fragmented app ecosystems and many regional services—harder to benchmark uniformly than a handful of national broadcasters elsewhere.
- Mixed ad-supported and subscription models and uneven ad-tech maturity—so event quality and metadata vary by partner.
- Mobile-first audiences moving between phone and TV—weaker cross-device graphs when identifiers are not comparable across environments.
- Network diversity (fixed broadband, mobile tethering, shared Wi‑Fi) affecting IP stability and geo accuracy.
- Multilingual and regional campaigns—reach and frequency stories need clear locale and partner definitions.
- Evolving regulatory expectations—multinational campaigns must reconcile EU, U.S. state, and local norms in parallel.
Practical value: Run pilot measurement with local partners before you export a global playbook. Definitions that work in one country may fail silently elsewhere.
KPIs: What Each Signal Can Support (Honest Matrix)#
| Goal | Often needs | Often insufficient alone |
|---|---|---|
| Delivery QA | Server + client signals, SSAI path map | IP only |
| Geo compliance | IP + declared geo + policy | IP only in VPN-heavy cohorts |
| Household frequency management | Graph + caps + disclosure | Single-device caps |
| Outcomes | Lift design, clean room or panel | Last-touch on CTV alone |
Measurement Maturity: A Practical Ladder#
- Level 1 — Device counts — Consistent definitions, basic reconciliation.
- Level 2 — Quality filters — IVT, viewability or audibility rules disclosed.
- Level 3 — Cross-environment — Household or modeled reach with documented uncertainty.
- Level 4 — Outcomes — Designed experiments and audited where required.
Document the evidence available at each level before using it for the next decision.
An Operational Checklist (Non-Legal)#
- Document which identifiers and inferences power each report.
- Prefer aggregated reporting where raw joins are not essential.
- Align definitions (impression, completion, household) in contracts with partners.
- Plan for identifier churn and policy changes—your stack will not stay static.
- Pair modeled reach with delivery verification so you are not optimizing purely on opaque scores.
- Revisit the contract when SSAI, app, or identity vendors change—measurement is not static.
Before evaluating household or purchase claims, check what the CTV measurement integration can substantiate.
Bring the evidence into a campaign review#
Use the cross-device frequency audit to examine exposure distributions, and the SSAI verification guide to identify which delivery signals a streaming integration can actually supply.
Our anonymized streaming-publisher video case shows a scoped gross-to-net impression reconciliation across three campaigns. Its count-once rules concern impression events; they do not establish household overlap, viewer attention or purchase lift.
Bring your campaign dates, placement scope, delivery reports and counting definitions to an AdProof measurement review. Confirm which delivery evidence the integration can support, reconcile the available counts, and document which audience or outcome questions require another measurement method.
Define the household before deduplicating exposure
Cross-screen reach depends on population, dates and identity unit. Discuss which exposure evidence is available for your campaign.
Discuss video measurementA short inquiry to scope relevant options; no campaign access is needed.
