# Worked example: headline test for a fictional B2B software company

**Illustrative data only.** Northstar Metrics sells analytics software to operations teams. These names, IDs, dates, spend, and outcomes are invented to show how to fill the brief. They are not a live LinkedIn test or performance claim.

## Complete plan, before native setup

Northstar asks whether a benefit-led headline produces cheaper website visits than a feature-led headline. The **Ad** variable changes only the headline. A says “See reporting gaps before renewal.” B says “One dashboard for renewal reporting.” The same single-image creative, body copy, CTA, landing page, demo offer, US operations-leader audience, exclusions, placements, website-visits objective, bid strategy, tracking event, and conversion settings apply to both. One ad belongs in each ad set. The selected LinkedIn test metric is cost per click; it tests traffic efficiency, not lead quality.

Planned budget: USD $2,400 total lifetime, USD $1,200 per ad set. Planned creation: 2026-11-02; start: 2026-11-03; end: 2026-11-24; native review: 2026-12-02; CRM cutoff: 2026-12-15, all America/New_York. These are planning values, not a platform minimum or accepted setup. Owner Maya will confirm the account's current displayed requirements, tracking, and Brand Lift overlap, then create and launch the test in Campaign Manager. Decision owner is Lee.

CRM rule: a qualified lead is a distinct person at a target account who requests a demo and passes sales acceptance within 21 days of an attributed click. Deduplicate by normalized work email, then by CRM contact ID. Attribute to the last eligible LinkedIn click within seven days before the form; one person belongs to one variant cohort. Join with the actual ad-set/variant ID, retain missing IDs as unknown, and wait through December 15. The business guardrail is cost per qualified lead at or below USD $300 and no worse than A by more than 10%. Lee will not expand B merely because it buys cheaper clicks.

**Plan review:** This is complete enough for Maya to attempt native setup. It is not evidence that the account accepted the test, that it ran, or that the promised CRM join worked.

## Readout 1: native inconclusive

Suppose Campaign Manager shows test `EX-AB-101`, variant IDs `EX-A-11` and `EX-B-12`, and cost per click of USD $7.20 for A and USD $6.90 for B. The report category is **Directional winner found / Inconclusive**, captured 2026-12-02 from fictional export `EX-NATIVE-101`. B is directionally cheaper, but LinkedIn has not reported a clear winner. Lee records “native inconclusive,” checks whether the test can be extended under current platform options, and waits for CRM lag before making a business choice. The exact fictional numbers do not create a significance threshold.

## Readout 2: cheaper clicks, worse qualified outcomes

In a separate hypothetical completed run of the same plan, LinkedIn reports **Winner found: B** on the selected cost-per-click metric. The fictional test is `EX-AB-102`, with variant IDs `EX-A-21` and `EX-B-22`. The fictional native export `EX-NATIVE-102`, captured 2026-12-02, shows A: USD $1,200 spend, 100 clicks, USD $12 per click; B: USD $1,200, 150 clicks, USD $8 per click. The stated native category, not the arithmetic alone, supports a native traffic-cost winner, B.

The fictional CRM snapshot `EX-CRM-102`, captured after the 2026-12-15 cutoff, joins `EX-A-21` and `EX-B-22` into distinct attributed cohorts with the same seven-day attribution and 21-day qualification windows. After work-email/contact-ID deduplication, A has 12 matched demo requests and 6 sales-accepted qualified people; B has 15 matched demo requests and 3 qualified people. Actual delivery was November 3–24, 2026, with no edits, pauses, or extensions. The CRM snapshot was captured December 16, 2026 at 09:00 America/New_York. Both cohorts include eligible clicks from November 3–24, with the same seven-day click-to-form attribution and 21-day click-to-qualification window. A has 6 disqualified people; B has 12. Unknown qualification, unmatched variant IDs, unresolved joins, and removed duplicates are each zero in both cohorts. These are explicit invented evidence values for this completed exercise. The variant-ID joins are reconciled. Cost per matched qualified person is A: USD $200; B: USD $400. B fails the USD $300 guardrail and is worse than A by more than 10%. Lee does not scale B on the click result; Maya checks message-to-offer fit, then proposes a new single-variable test. These values illustrate a decision, not a general claim that B2B headline tests behave this way.

## Readout 3: mismatched CRM cohorts

Suppose an export labels A by ad-set ID for the full run but labels B by campaign name for only its final week. A uses a seven-day click window; B includes 30-day view-through leads. Even if B shows more qualified leads, these are not comparable variant cohorts. Mark the CRM comparison **held**, request a new export using both actual variant IDs, the same dates, one attribution rule, one qualification rule, and deduplication, then set a new review owner/date. Do not fill missing outcomes with zero or use the native click winner as a substitute.