LinkedIn A/B Test Planning for Rare Qualified Leads

Check feasibility in the A/B Test Sample Size & Duration Planner. Supply your own qualification and eligibility rules; the tool does not retrieve LinkedIn leads or configure a native ad test.

Make qualification the outcome rather than a flattering proxy

A professional campaign may collect clicks, resource downloads, accepted connections, and meeting requests before producing a genuinely qualified lead. Those events answer different questions. If the decision concerns qualified opportunities, planning around the much more common click rate can produce an impressive sample that is irrelevant to the desired outcome. Begin with the qualification rule and the population that can reasonably satisfy it. The planner needs a single binary result per independent eligible person. It does not assign commercial value to engagement or infer that a person has purchasing authority because a profile appears relevant to the campaign.

Consider a fictional industrial energy-audit consultancy offering an assessment brief. Two destination invitations explain the same assessment: one leads with reducing avoidable energy use, while the other leads with preparing a site review. New eligible arrivals from the specified LinkedIn campaign are assigned externally to one invitation. A success is a request that meets a prewritten site-responsibility, facility-type, and assessment-scope rubric within 14 days. Both arms use the same qualification process and service offer. This is a comparison of invitation framing among destination arrivals, not a claim that one LinkedIn targeting strategy, job title, or advertising format outperforms another.

Keep the rubric separate from sales enthusiasm. A polite request for general information may not qualify, while a concise technical request may meet every criterion. Decide who reviews requests and which evidence is sufficient before launch; where practical, avoid letting reviewers’ knowledge of the invitation influence the decision. The calculator records only the planned outcome description, not the qualification evidence or reviewer judgments. If the team cannot apply the rule reliably within the observation window, the first task is to improve the measurement process. A sample-size output cannot compensate for a definition that changes from one adviser or campaign week to another.

Keep the people-based denominator separate from account demand

An eligible person in this illustration is a newly enrolled campaign arrival who meets the predefined professional and service-region requirements and has not already entered the experiment. Set the criteria before exposure to the invitation, and preserve assignment across return visits. Excluding people after they see a version because their response seems unpromising can bias the comparison. The denominator must include eligible assigned people who never submit a request as well as those who do. A list containing only completed forms would measure qualification among submitters, which is a different rate from qualified requests among all enrolled destination visitors.

Professional buying decisions often involve colleagues. Two people from the same facility may exchange the assessment brief or coordinate a request, so person-level independence is not automatic. If the organization must be assigned as a unit, the experiment becomes clustered and this model is not sufficient. Do not enter the number of companies as though it were the number of independent people while retaining a person-based baseline. An account-level qualification question needs an account-level design and method. State the unit clearly enough that the commercial team understands whether the eventual conclusion concerns people, facilities, or customer organizations.

The example assumes a defensible independent-person design is available externally; it does not claim that ordinary LinkedIn delivery establishes it. Native campaign experiments have their own documented setup and metrics, and the sample estimates here are not a reconstruction of those systems. Selecting LinkedIn changes the planning context, not the power formula or the assignment mechanism. The declaration checkbox similarly records the user’s intended conditions rather than verifying nonoverlap, deduplication, or company relationships. Resolve shared-decision and identity risks with the experiment owner before presenting a person-level estimate as an achievable measurement plan.

Choose a rare-event baseline and an economically meaningful lift

Suppose the consultancy supplies a 1 percent baseline qualified-request rate from a comparable mature destination cohort. Confirm that the historical denominator included all eligible assigned arrivals, not just requesters, and that the same 14-day qualification rule was used. A broad lead-form completion rate would overstate the baseline for this narrower outcome. Note the cohort dates, campaign promise, and source of the qualification decisions. The calculator treats the supplied rate as a planning assumption; it does not quantify uncertainty in that historical estimate or verify that the coming campaign will recruit a similar professional audience.

For the worked scenario, the smallest improvement the team chooses to power for is 50 percent relative lift: a rise from 1 percent to 1.5 percent. That is only half a percentage point in absolute terms, or five additional qualified requests per thousand eligible people at the assumed rates. A large relative percentage can therefore describe a small absolute contrast. The commercial owner should decide whether that gain would justify the new invitation and the effort of measuring it. The effect input is neither a sales forecast nor a promise that changing a few words can produce a 50 percent improvement.

A relative effect of 20 percent would target 1.2 percent, not 21 percent. A percentage-point effect of 1 would target 2 percent, doubling the baseline rather than adding one percent relatively. Read the derived target whenever changing modes because the numeric entry retains its value while its meaning changes. Choose confidence and power with the decision owner before results exist; this illustration uses 95 percent confidence and 80 percent power. The method plans a two-sided equality comparison under an assumed upward alternative. It does not establish equivalence, noninferiority, or the probability that one invitation will be commercially best.

Work through the energy-audit sample and 118-day horizon

Enter baseline 1 percent, relative effect 50, confidence 95 percent, power 80 percent, and an equal A/B split. The actual core returns 7,750 people in A and 7,750 in B, totaling 15,500 independent eligible participants. Achieved power under the approximation is about 80.0022 percent after rounding. The expected success counts are 77.5 for A and 116.25 for B at the assumed rates. Fractional expectations are mathematical averages across hypothetical repetitions, not partial people or promised observed leads. Do not convert them into a sales commitment, and do not substitute the number of requests required for the participant denominator.

At 150 combined eligible people per day, each arm receives an assumed 75 daily. Set minimum enrollment to 21 days, maturity to 14 days, and start date to 2026-11-16. Sample accumulation requires 104 days, so the 21-day minimum does not shorten enrollment. The calculated enrollment end is 2027-02-27, and the review date is 2027-03-13. Total duration is 118 days, consisting of 104 enrollment days plus 14 maturity days. That result is a warning about feasibility for this particular question, not a recommendation to run a campaign unchanged for nearly four months regardless of commercial circumstances.

The schedule assumes eligible recruitment every calendar day, including weekends and holiday periods. If 150 is a working-day estimate or a launch-week peak, the duration is too optimistic for the intended traffic pattern. It also assumes that the qualification definition and outcome collection remain workable across the entire horizon. A change in service scope or facility eligibility midway through enrollment can undermine comparability even if the numeric sample target is eventually reached. Discuss those risks before approving the run. A longer timeline creates more opportunity for operational drift, which the calculator’s calendar arithmetic does not model or correct.

Distinguish attribution windows from qualification maturity

LinkedIn conversion attribution describes when and how a platform credits an outcome after an ad interaction. Qualification maturity describes how long an enrolled participant has to satisfy this experiment’s outcome rule. These are not interchangeable clocks. A platform-attributed lead count can reflect its selected attribution settings without matching the experiment’s assigned-person cohort or 14-day definition. Keep those reporting concepts separate in the brief and reconcile the actual data source before analysis. This planner’s maturity input adds time after enrollment; it does not configure attribution, identify the credited campaign, or establish that every reported conversion belongs to the randomized destination comparison.

For the energy-audit example, a person enrolled on the final day needs the same 14-day opportunity as someone enrolled on the first day. Reviewing all requests immediately when enrollment closes would leave recent participants less time to qualify. The added maturity period is intended to accommodate that final window, but simply waiting does not prove that all outcomes were captured. Qualification records can still be missing, delayed, or inconsistent. Set an external procedure for closing each participant’s window and checking completeness. Avoid assigning no merely because an adviser has not yet reviewed a request that arrived within the allowed period.

The planner accepts a bounded maturity allowance rather than a model of long sales cycles. If the meaningful outcome requires months of procurement review beyond what this planning tool represents, choose an appropriate specialist design or a genuinely useful earlier outcome with a new baseline. Do not rename a resource download qualified lead to obtain an earlier review. A shorter proxy can be informative when its role is explicit, but it answers a different question. Record the handoff between qualification, meeting scheduling, opportunity creation, and revenue so later teams do not infer that a qualified-request result proves downstream sales impact.

Use smaller and larger effects to expose the decision trade-off

At half the entered effect, the target becomes 1.25 percent, a 25 percent relative improvement and a quarter-point absolute increase. The core requires 27,937 people per arm, totaling 55,874. With the same 150 eligible people daily, enrollment takes 373 days and the full horizon takes 387 days. A November 16, 2026 start gives a planned review date of December 7, 2027. This is not a claim that the campaign will remain stable for that period. It shows that sensitivity to a smaller rare-event difference can be far beyond the available recruitment and decision window.

At double the original effect, the target is 2 percent, or a 100 percent relative lift. Required enrollment falls to 2,319 people per arm, 4,638 total. The same traffic supports 31 enrollment days and 45 days overall, with a planned review on 2026-12-30. That schedule is easier to imagine, but it tests a much larger alternative. The consultancy cannot choose it for convenience and still claim the design was sensitive to the original half-point gain. Ask whether doubling qualified requests is the smallest effect that matters, not whether it is the effect that produces the most appealing calendar.

If none of the meaningful effects fit the available horizon, pausing the confirmatory test may be the best planning decision. The team can improve the qualification process, run qualitative interviews, or use an explicitly exploratory comparison without claiming that it resolves the powered question. Increasing eligible recruitment can help only if the additional participants fit the same design and outcome definition. Adding many low-fit visitors may change the baseline rather than solve the shortage. Lowering confidence or power also changes the error trade-off; it should not be presented as obtaining the same evidential standard with fewer people.

Read approximation warnings without treating them as certification

The planner uses a normal approximation with pooled variance under equality and unpooled variance under the proposed alternative. It is not an exact rare-event calculation. It checks expected successes and failures in each arm and warns when the minimum expected cell is below ten. The energy-audit fixture’s smallest expected cell is 77.5, so that particular warning is absent. Absence of the warning does not verify the baseline, independent assignment, stable qualification, or absence of bias. Those design conditions can fail at any sample size. Treat the count as an approximate planning target and obtain specialist review when the decision or data structure requires it.

The sample is rounded upward and approximate power is evaluated at the resulting arm sizes. This helps the returned integer plan meet the requested power under its model, but it is not a guarantee of the globally smallest possible integer allocation. The model also has a per-arm cap; an unavailable capped scenario is not a zero-person requirement or permission to use the cap as an adequate target. Sensitivity rows with invalid target rates or excessive requirements may be unavailable independently of a usable main plan. Read each scenario’s status and warnings instead of assuming that every row represents an executable experiment.

During a long run, outcome peeking can be especially tempting because the team wants an earlier commercial answer. This fixed-horizon plan does not provide sequential error control, repeated-look adjustments, or corrections for testing many audience slices. Monitor data integrity and implementation health, but keep the inferential decision at the agreed horizon. Once mature independent counts are available, the separate A/B Test Results Analyzer can review supplied binary results where appropriate. Neither tool can retroactively make a selected favorable subgroup or a stopped-when-ahead run match the originally planned design.

Give the commercial owner a reproducible feasibility brief

A useful handoff states the eligibility population, exact qualification rule, baseline source, useful effect, allocation, recruitment assumption, and full review horizon. It also identifies who owns assignment, qualification, and outcome completeness. Save the editable project JSON before revising assumptions, and rebuild after changes so no old participant target remains attached to a new brief. The report CSV is a planning report, not a CRM import or an observation-data template. No contact records are required to calculate the sample. Keep private prospect details and adviser assessments in the systems authorized to hold them rather than adding them to the plan’s free-text notes.

Context fields can still disclose sensitive commercial intentions even without a prospect list. Copy, CSV, and PDF may include the supplied title, outcome, eligibility, and notes; the full PDF also contains the project JSON and report CSV attachments. Review those artifacts before sharing them with a client, agency, or sales team. The tool does not offer an automatically anonymous summary or strip confidential planning assumptions. Use neutral campaign labels when appropriate and keep a deliberate local save for reopening. The calculator does not connect to Campaign Manager, inspect attributed conversions, or continue collecting qualified outcomes after the browser session.

SMTasker can be evaluated separately for the LinkedIn account-workspace operations it supports, including optional Skye assistance and permission-controlled MCP access in configured workflows. Those are operational capabilities, not lead-qualification evidence, CRM synchronization, experiment randomization, or statistical certification. Keep the adviser responsible for the service-fit decision and the experiment owner responsible for the assigned-person comparison. The strongest result of planning may be a clear decision not to test the rare outcome yet. Account administration can still improve around that decision without pretending it removes the recruitment or maturity constraints the calculation has made visible.

Questions about LinkedIn A/B Test Planning

Continue with another planning guide

Plan your two-arm binary-outcome test in the A/B Test Sample Size & Duration Planner, or choose another guide below. Find more practical resources in all free tools.