1. The Core Taxonomy: Device vs. User vs. Cross-Device
In programmatic advertising, confusing hardware targeting with audience identity is one of the most common architectural mistakes:
- Device Targeting (Hardware/OS Level): Filtering bid requests by explicit physical attributes passed in the OpenRTB
deviceobject (e.g.device.make = "Apple",device.model = "iPhone",device.osv = "17.4",device.connectiontype = 6[Ethernet]). There is zero concept of a person or persistent history. - User Targeting (Single-Profile Level): Targeting based on behavioral segments, past purchase history, or third-party demographic data associated with a single client identifier (e.g., cookie ID or MAID).
- Cross-Device Targeting (Identity Graph Level): Connecting multiple distinct physical devices (a Smart TV, a work laptop, an iPhone, and a personal tablet) into a single cohesive profile or household graph unit.
2. Hardware, OS, and Network Signals in the Bidstream
How does an ad server or DSP know what device is requesting the ad? In modern browsers and apps, this occurs through structured headers and OpenRTB parameters:
// User-Agent Client Hints (Modern replacement for frozen User-Agent string):
Sec-CH-UA: "Chromium";v="124", "Google Chrome";v="124"
Sec-CH-UA-Mobile: ?1
Sec-CH-UA-Platform: "Android"
Sec-CH-UA-Platform-Version: "14.0.0"
Sec-CH-UA-Model: "Pixel 8 Pro"3. Inside the Device Graph: Deterministic vs. Probabilistic
To target a consumer across devices without third-party cookies, identity vendors (LiveRamp, TransUnion/Neustar, Experian) construct Identity Graphs:
| Dimension | Deterministic Matching | Probabilistic Matching |
|---|---|---|
| Foundation | Authenticated logins (hashed emails SHA-256 / phone numbers). | Statistical clustering of IP addresses, location co-presence, and timing. |
| Precision | Very High (95%+ match confidence). | Moderate (70%–85% confidence; prone to office/public Wi-Fi noise). |
| Scale | Limited to logged-in user environments (e.g. streaming apps, news sites). | Broad scale across the open web and mobile ecosystem. |
4. The CTV-to-Mobile Sequential Storytelling Architecture
The most profitable application of cross-device targeting is Living Room CTV to Personal Mobile retargeting:
[Living Room Connected TV (CTV/Roku)]
│
▼ 1. Unskippable 30-second automotive video ad plays on Hulu / Max
[DSP Impression Logger]
│ Logs Household IP: 74.125.19.42 + Timestamp + Campaign ID
▼ 2. Device Graph Lookup
[Household Identity Resolution Engine]
│ Resolves: IP 74.125.19.42 contains:
│ - Device 1: Roku Ultra (CTV)
│ - Device 2: iPhone 15 (Mobile IDFA)
│ - Device 3: MacBook Air (Browser Cookie/UID2)
▼ 3. Dynamic Audience Injection
[Mobile Retargeting Segment Populated]
│
▼ 4. User opens news or social app on iPhone 15 minutes later
[Mobile Display Ad Rendered with "Book a Test Drive" Direct CTA]
5. Signal Loss & Modern Privacy Safeguards
Cross-device tracking faces strict regulatory and architectural constraints:
- Apple Private Relay & iCloud Privacy: Masks Safari IP addresses via dual-hop proxies, preventing residential IP clustering.
- Dynamic IPv6 Addressing: Mobile carriers rotate IPv6 prefixes dynamically, reducing the persistence of IP-based graphs to mere hours.
- Consent Enforcement (GDPR / CPRA): If an EU or California user opts out of sale/share, the device graph must drop all associated node edges immediately.
6. Solution Engineer RCA: "CTV Ads Run, but Mobile Pool is Empty"
Incident Investigation
Symptom: Brand spent $40,000 on CTV ads. The downstream mobile retargeting campaign delivers 0 impressions.
Solution Engineer Resolution:
- Audited CTV bidstream logs. Discovered the SSP was transmitting truncated IP addresses (e.g.
74.125.19.0with the last octet zeroed out) to comply with legacy German privacy defaults. - The device graph required a full IPv4 address to map to a single residential household; because the last octet was zeroed, the graph engine rejected 100% of impressions as non-resolvable.
- Solution: Transitioned targeting to deterministic
user.ext.eids(UID2 tokens) shared between the CTV publisher app and mobile publishers.
7. Real-World Interview Scenarios
Scenario: "What are the primary failure modes of IP-based household targeting?"
Answer: 1) Commercial / Corporate Networks (treating an entire corporate headquarters with 5,000 employees as one household). 2) Carrier CGNAT (Carrier-Grade NAT, where thousands of mobile subscribers share a small pool of public IP addresses). 3) Dynamic IP re-assignment by ISPs. Robust solutions require frequency capping at the device identifier level in addition to IP matching.
Audit Privacy Strings & Consent Gates
Ensure cross-device targeting complies with IAB TCF v2.2 Purpose 1 & Purpose 3 consent requirements in the live Consent Decoder.