Understand why the identity graph breaks
The paper contrasts household-based linear television with individual mobile identities and the ambiguous shared nature of CTV. The same viewer can generate a set-top-box exposure, a mobile device or login record, and a CTV identifier. Those systems were not designed as one identity layer, so their reach figures can describe overlapping people while appearing independent.
Start every analysis by naming the measurement unit: person, profile, device, account, household, impression, or modeled relationship. Record how the identifier is created and how long it persists. A platform-reported logged-in user is not automatically comparable with a television household or an advertising device ID. Clarity at this stage prevents false precision later.
Account for network and telco-login mismatches
A home CTV can use broadband while a mobile device uses a separate cellular network, producing different and changing network signals. IP matching alone cannot reliably prove that both devices belong to one person. Shared networks also create the opposite risk: several people may appear connected because they use the same household connection.
The paper adds a specifically important India scenario. A telco bundle user may authenticate on mobile with a phone number but use an email or platform identity on CTV. Unless the services provide a consented link, one person can look like separate users. That affects frequency, attribution, engagement, and even subscriber economics. The limitation should be built into planning, not buried in a methodology note.
| Measurement problem | What may be confused | Planning response | Evidence standard |
|---|---|---|---|
| Household versus person | Shared TV exposure and individual viewing | Report both units where relevant | Do not infer the viewer without support |
| Cross-device IDs | Mobile, browser, app, and CTV records | Use overlap ranges | Label deterministic and probabilistic links |
| Network mismatch | Home broadband and mobile data | Avoid IP-only person matching | Require corroborating consented signals |
| Telco login split | Phone identity and email identity | Test sensitivity to duplicate users | Use a verified linking path |
| Walled gardens | Different definitions and attribution windows | Create a common reporting dictionary | Reconcile before comparison |
Build a privacy-safe measurement contract
The paper notes that India's privacy environment, including the DPDP Act, raises the standard for linking and secondary use of data. Measurement design should specify purpose, data source, consent basis, retention, access, and allowed use before a campaign begins. More matching is not automatically better if the relationship cannot be explained or governed.
Create a shared data dictionary across agencies, platforms, and analytics partners. Define impressions, views, completed views, households, users, conversions, attribution windows, and invalid traffic treatment. Record which results are directly observed and which are modeled. A smaller set of comparable, governed metrics is more useful than a broad dashboard of incompatible precision.
Use a layered measurement architecture
Use platform data to manage delivery inside each environment, but do not ask it alone to prove cross-platform causality. Add reach and overlap analysis with uncertainty ranges, controlled or quasi-controlled tests for incremental effects, and marketing mix modeling for broader contribution over time. Each method answers a different question and carries different assumptions.
Design experiments around decisions. Hold out matched geographies or periods where feasible, vary one meaningful media component, and state the expected effect before seeing results. For tentpole campaigns, plan around event-driven demand that could otherwise be mistaken for advertising impact. Feed experimental findings back into reach and contribution models instead of treating a test as a one-time case study.
Report uncertainty as a decision input
A measurement report should show the observed result, plausible range, major assumptions, and decision implication. If unique reach depends on an overlap estimate, show what changes under lower and higher duplication. If a conversion result depends on an attribution window, report the sensitivity. Decision-makers can work with uncertainty when its commercial impact is visible.
Set escalation rules for conclusions. Descriptive platform delivery may support optimization; causal claims require stronger evidence; person-level identity claims require verified linkage and governance. This hierarchy protects the organization from turning a modeled convenience into a fact. It also directs investment toward the missing evidence that would most change the decision.
Decision implication
Perfect deterministic cross-platform attribution is not the current operating reality described by the paper. A credible India measurement plan names its units, respects consent, separates observed and modeled links, and triangulates delivery with reach ranges, experiments, and mix analysis. Honest uncertainty is not a weakness; it is part of responsible approval.
Quantify cross-platform reach and contribution without hiding identity uncertainty
Use a PyxiVisio decision-intelligence model to connect assumptions, delivery, economics, risk and approval conditions.