Paid Social:
Reach, Targeting, Costs, and Incremental Results

A paid-social report can show that an ad was delivered, that someone was eligible to receive it, and that a conversion received credit—without showing that the advertising caused the conversion. Reach, targeting, attribution, and incrementality belong to four different evidence layers. Only the last one asks what changed relative to a counterfactual.

Each layer produces a different record: reach audits delivery, targeting compares populations, attribution reconciles credited paths, and incrementality evaluates a counterfactual. The combined report joins those records without turning them into one claim.

Paid social purchases delivery under platform rules#

Advertisers provide an objective, audience inputs, creative, destination, budget, bid or cost controls, and conversion signals. The platform determines delivery through its current auction, eligibility, placement, pacing, and optimization systems. The exact system differs by platform and changes over time.

This makes paid social distinct from organic social activity. Payment buys an opportunity for distribution under the platform’s rules. It does not buy verified attention, persuasion, qualified pipeline, or incremental revenue.

Campaign reports can expose delivery, engagement, and conversion-related metrics. LinkedIn’s Campaign Manager documentation is one platform example. Its definitions should be read as LinkedIn measurement rules, not universal advertising facts.

Reach is delivery evidence#

Reach generally counts people or members to whom ads were delivered under a platform’s identity model; impressions count deliveries. Because identity resolution and privacy controls vary, reach is not a census of unique humans across every device and platform.

Reach can answer whether the system delivered at the intended scale and frequency. It cannot tell whether recipients noticed, understood, believed, or acted because of the ad.

Reach needs the platform’s exact definition and export date beside it. Cross-system comparisons otherwise risk treating different identity and counting rules as identical.

Targeting defines eligibility, not certainty#

Targeting can use geography, organization, role, interests, behavior, lists, modeled audiences, exclusions, and platform optimization signals. The selected fields constrain an eligible or prioritized population. They do not guarantee that every delivered person matches the buyer definition, sees the creative, or has authority and need.

Three populations must remain separate:

1. Intended audience: the people or accounts described in the campaign plan.
2. Platform-eligible audience: identities matching the platform’s available fields and policies.
3. Delivered audience: recipients selected through auction and optimization during the campaign.

Differences among the three can come from stale profile data, modeled attributes, list-match rates, exclusions, inventory, bidding, optimization, and budget. Evaluate targeting with aggregate fit evidence and privacy-safe downstream outcomes, not by claiming precise knowledge of individuals.

Targeting narrows eligibility. It does not prove identity, attention, need, or consent to a sales interaction.

Attribution allocates credit under a model#

LinkedIn documents attribution through click and view windows. Google Analytics defines attribution more generally as assigning conversion credit to touchpoints using rules or algorithms. The same conversion can therefore receive different credit across a platform and an analytics system.

Common reasons reports disagree include:

• different click-through and view-through windows;
• different identity resolution across devices and browsers;
• event timestamps, time zones, and processing delays;
• conversion definitions and deduplication rules;
• modeled or consent-limited data;
• platform self-attribution versus cross-channel allocation; and
• incorrect or inconsistent UTM and third-party tracking configuration.

LinkedIn’s third-party tracking guidance addresses configuration and tagging within its ecosystem. It cannot make two systems identical when their fundamental counting and identity rules differ.

Platform and analytics attribution depend on models, windows, tagging, identity, and conversion definitions. LinkedIn Ads attribution and Marketing Solutions tracking practices show that their outputs allocate observed credit and can legitimately differ without either number being a causal estimate, as GA4 also explains.

Reconcile reports by writing a measurement contract: event, source of truth, window, timezone, identity, deduplication, model, currency, and revision date. Then label each number by system rather than selecting the larger result.

Incrementality asks what changed because of ads#

Incrementality is a causal question. It compares the observed outcome under advertising with an estimate of what would have occurred without that advertising. The no-ad outcome is not directly observed for the same person at the same time, so the design must construct a defensible counterfactual.

Randomized user-level holdouts can do this when the platform, population, consent, scale, and business constraints permit. Randomized paired-geo experiments can compare treated and control geographies when individual randomization is unavailable and geographic assumptions are credible. Google Research documents methods for incremental return on ad spend using randomized paired geographies and for repeated measurement across test periods.

These methods have conditions:

• enough eligible units and outcome volume to distinguish signal from noise;
• stable assignment and limited contamination between treatment and control;
• no concurrent intervention that differs systematically by group;
• a measurement horizon covering the relevant sales response; and
• an analysis plan that reports uncertainty and implementation failures.

B2B SaaS can be especially constrained by small account counts, long sales cycles, sales intervention, and cross-region buying groups. When a credible experiment is not feasible, use cautious observational language. “Attributed” is accurate; “incremental” is not.

Randomized geo experiments can estimate advertising incrementality under explicit design and inference conditions. The Google multiple-test-period study reports that the research also shows that effectiveness can vary over time, so one estimate is not automatically permanent or transferable.

Join the four records without collapsing their claims#

Build a paid-social decision record with separate columns:

If only delivery and attribution are available, make a decision appropriate to those layers: repair delivery, improve measurement, or form a bounded hypothesis. Do not manufacture an incremental ROAS by relabeling platform revenue.

No universal paid-social CPM, frequency, attribution window, ROAS, or lift benchmark was verified. Markets, objectives, platforms, offers, margins, and experimental power differ.

Use common questions to locate the missing evidence#

Delivery missed the plan. Where should diagnosis start?#

Compare the intended, platform-eligible, and delivered audiences, then inspect reach under the platform’s identity rules. Targeting explains eligibility inputs; reach records delivery, not buyer fit or attention.

What may the team call a credited conversion?#

Call it attributed under the named model and window. “Incremental” requires a defensible no-ad counterfactual.

Which system should win when LinkedIn and analytics disagree?#

Neither total wins by being larger. Reconcile their windows, models, identities, timestamps, event definitions, and deduplication, then label each result by system.

When does a randomized result still need qualification?#

Randomization strengthens the counterfactual, but power, contamination, noncompliance, outcome quality, and horizon still limit the estimate. Scope it to the tested population, period, implementation, and assumptions.

Escalate a paid-social claim only as far as its evidence permits: delivery records support delivery decisions, population comparisons diagnose targeting, and named models allocate attributed credit. Reserve “incremental” and finance-approved contribution for an effect estimate with a defensible counterfactual, stated uncertainty, and decision-matched scope.

Frequently asked questions#