Why a healthy LinkedIn CPL can still produce no pipeline

A cost per lead is a cost per form

The report is green. The cost per lead is inside what was agreed. The pipeline review shows nothing. Both facts can be true at once, and they usually are for the same reason: a cost per lead is a cost per form, not a cost per buyer. Four documented things explain most of the gap, and none of them needs an invented statistic.

Reason one: the measurement itself is softer than the report looks

The Insight Tag is browser-based and cookie-based

LinkedIn describes its own tag that way. The Insight Tag is “a snippet of JavaScript code”, “lightweight”, added to your website to track conversions, retarget visitors and report demographics, and it relies on “LinkedIn cookies, which enable us to match your website visitors to their respective LinkedIn member accounts”. A cookie-matched visitor is a real signal and a partial one, and the report does not say which. Source: LinkedIn, Insight Tag.

LinkedIn publishes its own reliability ratings, and people rarely read them

LinkedIn’s conversion-tracking guide lists three data sources and rates the reliability of each:

Insight Tag — “Browser-based, requires JavaScript code added to pages on your website” — reliability MEDIUM, “Cookie-based, can be subject to browser limitations”.

Conversions API — “Server-to-server (LinkedIn)” — reliability HIGH, “Doesn’t rely on cookies, direct connection with servers”.

CSV file upload — “Manual data input and upload into Campaign Manager” — reliability MEDIUM, “Relies on manual data input”.

Most reports are built entirely on the source LinkedIn itself rates MEDIUM. Source: LinkedIn Help, Get started with LinkedIn Conversion Tracking.

Running both does not double-count

The question every client asks before agreeing to server-side tracking is whether sending the same event twice inflates the numbers. LinkedIn’s answer: “We deduplicate information sent through Conversions API and Insight Tag, which means we identify if an event has been sent through both and only count it once.” And on the data itself: “Email addresses are hashed using SHA-256.” Source: LinkedIn Help, LinkedIn Conversions API.

The attribution window is a setting, not a fact

Conversion windows are configurable. LinkedIn’s help page states the default as a “90-day click, 90-day view window, which is the default in Campaign Manager”, with standard conversions settable to 1, 7, 30 or 90 days and longer lookbacks available for certain conversion categories sent through the Conversions API or a CSV. Both kinds of conversion are recorded: a view-through conversion “records when a member sees a LinkedIn Ad” and a post-click conversion “records when a member clicks on a LinkedIn Ad”. A 90-day view window means a form filled in December can be credited to an impression from September. Change the window and the same quarter reports a different number. Source: LinkedIn Help, LinkedIn conversion window.

So is the attribution model

“When a member converts after interacting with multiple ad sets, you can choose to credit all ad sets or just the most recent one.” Two accounts with identical spend and identical outcomes can report very different conversion counts on this setting alone. Source: LinkedIn Help, Get started with LinkedIn Conversion Tracking.

Reason two: the audience is wider than the one you built

Audience expansion is on by default

“Audience expansion is automatically enabled on ad sets where it’s available.” It shows your ads to members with professional attributes similar to your target audience, and the reporting “will include activity from both your target audience and expansion audience”, while the forecast “won’t include member accounts from audience expansion”. For a cost per lead that means leads can arrive from outside the ICP while the report shows one blended number. We attach no waste percentage to that; the figures circulating for it have no methodology. Source: LinkedIn Help, Use audience expansion in your ad set.

The Audience Network delivers your ads off LinkedIn

The LinkedIn Audience Network “allows you to extend the reach of your applicable ad sets by delivering your ads beyond the LinkedIn feed to members on third-party apps and sites, if their settings allow”, with “the same targeting parameters, bid type, and budget”. The brand-safety controls exist and are worth naming: publisher-level IAB category exclusions provided by Pixalate, custom allow or block lists, and imported DoubleVerify profiles. A lead from a third-party app is still a lead in the report. It is not the same lead. Source: LinkedIn Help, LinkedIn Audience Network.

Reason three: nobody told the platform what the lead was worth

Offline conversion import, and what it costs in patience

If the platform only ever hears about the form fill, it optimises to the form fill. The fix is to send the qualified stages back, and the manual route is a CSV upload with rules: the data must be formatted identically to LinkedIn’s template, the currency “must be the same as the ad account currency”, LinkedIn recommends updating “daily or weekly for the best results”, and “it can take up to 24 hours for the data to be available”. It is patient work, and it is the work that makes the report mean something. Source: LinkedIn Help, Manually upload a conversion data CSV file.

The Revenue Attribution Report, and its real prerequisites

LinkedIn describes it as “a report that measures the influence your LinkedIn marketing has on revenue” — revenue won, pipeline, win rates, days to close and deal sizes. The prerequisites are real: a Business Manager account connected to your CRM (Salesforce, Dynamics 365 and HubSpot are the three LinkedIn lists), admin access to Business Manager, an authorization protocol that connects the CRM “without sharing login credentials directly with LinkedIn”, and “up to 72 hours for data to appear”. The report offers impression-based and engagement-based attribution models over a lookback window you choose, so the model is a decision here too. We name only the CRM connection we have verified inside the account in front of us, and we do not repeat LinkedIn’s own uplift figures for the report, because they are vendor-reported, from self-selected customers, with no published methodology. Sources: LinkedIn, Revenue Attribution Report and LinkedIn Help, Use your Revenue Attribution Report in Business Manager.

Reason four: the decision happens before the click

One named, dated report, used carefully. 6sense’s 2025 Buyer Experience Report, published in November 2025, found that the buying journey “shifted from a 70/30 to 60/40 split between research and seller engagement” — buyers are talking to sellers earlier than the year before, which cuts against the familiar line that most of the journey is over before anyone talks to you, and is a better reason to be honest about the number. From the same report: “94% of buying groups ranked preferred vendors before first contact”, and “they ultimately purchased from that preliminary favorite 77% of the time.” That pair is the actual argument for advertising on this platform: being on the shortlist before the form exists. It needs no invented statistic. Source: 6sense, 2025 Buyer Experience Report announcement.

What we do about it

Three things. A required “How did you hear about us?” question on every form, read alongside platform data and never instead of it — with its limits stated in the same breath. HockeyStack’s study of self-reported attribution found roughly a fifth of answers invalid or unusable, evidence of memory gaps well beyond the few respondents who admitted them, and concluded that self-reported attribution “is neither correlative or causal” and is “not a sufficient data point for serious decision-making” on its own. Offline conversion import, so the platform optimises to the stage that matters rather than the form fill. And a report that puts the CRM view next to the platform view rather than choosing one. Source: HockeyStack, Self-Reported Attribution Report 2024.

What gets reported instead of a CPL

SQLs, pipeline created and CAC payback, with MQL named as a leading indicator and never as an outcome. The case against MQL as an outcome is not ours alone. As Mark Stouse put it in MarTech, “The traditional methods of counting MQLs — tracking form fills, content downloads and webinar registrations — have become increasingly detached from revenue generation.” Source: MarTech, Why the MQL model is failing B2B marketing.

Then the timing, honestly. On a long B2B cycle a pipeline signal lands months after the spend, so the first weeks are read on leading indicators, the first quarter on SQLs, and pipeline and CAC payback after that. An initial term of 90 days is not long enough to be judged on pipeline alone, and any agency selling it that way is selling a metric it cannot deliver inside the contract. The whole method is written down in How we measure.