Target by carrier & network.

1. Define the eligible opportunity

For 3g and wi-fi traffic targeting, write the measurement unit before choosing inventory or creative. The unit for this page is a connection-context impression or visit linked to source, device, GEO and accepted outcome. That definition prevents impressions, clicks, visits, installs and accepted business outcomes from being mixed into one ambiguous conversion total. State the inclusion rule, the disqualifying conditions and the time at which the event becomes final.

Record the targeting hypothesis in one sentence: the selected signal should improve the probability of the primary outcome compared with a broader baseline. Keep the hypothesis narrow enough to falsify. When several signals are bundled together, create separate ad groups or campaign cells so each major assumption can be evaluated without guessing which input caused the result.

2. Separate targeting from observation

The main planning dimensions are connection type, carrier, device, operating system, browser, GEO, source, page weight and conversion path. Decide which dimensions actively restrict delivery and which remain reporting fields. Observation can preserve learning and reach while the team measures whether a segment deserves a stricter targeting rule. Exclusions must be documented with the same care as inclusions because an exclusion can remove profitable demand just as easily as a target can add relevance.

Build a small taxonomy for campaign, source, placement, creative, audience or device rule and destination. Preserve those identifiers through redirects, analytics, conversion tracking and the final business system. A targeting report that stops at the ad platform cannot prove lead acceptance, subscription retention, approved revenue or another business-defined result.

3. Design the controlled test

Use one stable destination, one primary event, one attribution window and one loss ceiling for the first comparison. Hold the offer and core creative promise constant while testing the targeting dimension. Set a minimum observation period that covers normal weekday, device and conversion-delay variation. Do not declare a winner after a single cheap day or one unusually strong placement.

A practical test contains a broader control cell and one or more targeted cells. Budget should be large enough to observe the useful event but small enough that a failed hypothesis remains affordable. If volume is thin, widen only one restriction at a time. Document every change so later improvements are not incorrectly attributed to the original targeting choice.

4. Protect experience continuity

The creative, audience or device promise must continue on the destination. A visitor should immediately recognize why the page, app or offer is relevant to the context that produced the click. Validate loading speed, form usability, deep links, browser or app compatibility, language, location availability and the path to the primary action. Targeting cannot rescue a slow, misleading or technically broken destination.

Review the journey on representative devices and environments rather than only in a desktop preview. For mobile or app contexts, test keyboard behavior, orientation, consent flows and return navigation. For desktop contexts, use the available screen space without creating dense or inaccessible layouts. The measurement plan should record technical failures separately from user rejection.

5. Evaluate quality, not nominal price

A cheap 3g and wi-fi traffic targeting campaign is useful only when the lower media price survives quality reconciliation. Compare valid delivery, engaged visits, useful actions, accepted conversions, refunds or reversals, and complete acquisition cost. Segment size and click-through rate are diagnostics, not proof of profit. Mature the data before comparing cells whose conversion or approval delays differ.

The most dangerous shortcut is treating a connection label as a proxy for intent or human quality. Prevent it with source-level monitoring, clear frequency rules, invalid-activity review and a stop condition defined before launch. When the platform reports modeled or estimated results, label them separately from directly observed first-party events so decision makers understand the evidence quality.

6. Scale without losing the explanation

The operational role of this page is to test network context while holding offer, source and destination measurement constant. Scale only after the targeted cell repeats across enough time, sources and creatives. Increase one material dimension per step, such as budget, GEO, audience size, placement count or creative volume. Keep the prior stable state available so the team can roll back quickly if quality deteriorates.

During scaling, watch marginal rather than blended performance. A campaign can retain an attractive overall average while each new unit of spend becomes unprofitable. Re-check exclusions, frequency, source concentration and destination performance after every expansion. Stop or reduce spend when the mature marginal result falls below the written threshold.

7. Privacy, consent and data boundaries

Use only targeting and measurement signals that are permitted for the platform, destination, jurisdiction and user relationship. Record whether a signal is first-party, contextual, platform-estimated or derived from device or location information. Respect consent and opt-out states, minimize retained data and avoid promising user-level precision where the available evidence is aggregate or modeled.

Remarketing, app and operating-system environments can impose additional identifier and authorization limits. Build the campaign so it still produces useful aggregate evidence when a user-level identifier is absent. Missing attribution should not automatically be treated as zero value, but modeled value should not be presented as directly observed fact.

8. Decision and rollback rule

The final decision is whether the connection segment creates incremental accepted value after technical and source-quality controls. Define the acceptable range before traffic starts. A scale decision should require the primary accepted event, a complete cost calculation and enough repetition to reject an obvious one-day anomaly. Secondary metrics explain why performance changed, but they do not replace the primary business threshold.

The rollback package should contain the previous budget, targeting rules, exclusions, creative set, landing-page version and tracking configuration. Pause the affected expansion first, preserve logs and diagnose whether the loss came from audience dilution, source mix, creative fatigue, destination failure or measurement drift. Reopen only after the cause and the validation test are documented.

Launch checklist

1. Name the primary accepted event and its maturity window.
2. Document the targeting rule, observation fields and exclusions.
3. Confirm source, placement, device, audience and destination identifiers.
4. Validate consent, privacy, location and operating-system constraints.
5. Test the creative-to-destination journey in representative environments.
6. Set budget, loss ceiling, stop rule and rollback state before launch.
7. Reconcile platform delivery with analytics and business-system outcomes.
8. Scale one material variable only after the result repeats.

Primary documentation

• IAB Tech Lab: OpenRTB 2.6 specification
• Google Ads: About device targeting
• Google Ads: Target ads to geographic locations
• Google Ads: Invalid traffic
• IAB/MRC: Viewable impression guidelines
• Google Analytics: Traffic acquisition report