Why this LinkedIn campaign had 38% invalid traffic

This is a field note from a six-week LinkedIn Sponsored Content pilot we ran alongside a B2B SaaS customer in Q1. The account was mid-market, ABM-shaped, ~$42K in spend across two campaigns, targeting a named list of ~4,200 accounts. The dashboard said the campaign was working. The receipts said 38% of the clicks were never going to convert — and hadn't been converting for six weeks.

What the platform said

LinkedIn Campaign Manager reported the numbers every paid social manager wants to see on a Monday: CTR at 0.71% (above LinkedIn's B2B SaaS benchmark), CPC at $12.40, a healthy tilt toward the named account list, and a respectable share of clicks from Director+ titles. Invalid Traffic, per LinkedIn's own filter, was reported at 2.1%. Nothing on the surface looked wrong. The customer's internal weekly report went out on Fridays with a green status.

What the receipts said

We were running click-level verification against the same campaign — every click stamped with a receipt (timestamp, ASN, device fingerprint, session behavior, journey continuity to the destination). When we reconciled the receipts against LinkedIn's reported click volume for the six-week window, three things fell out:

• 19.4% of clicks came from datacenter and hosting ASNs that LinkedIn's own filter had already passed. Not obvious bots — residential-proxy-fronted datacenter traffic, which the platform-side filter isn't tuned to catch at scale.
• 11.2% of clicks had zero on-site dwell — landing page loaded, no scroll, no interaction, session terminated inside 800ms. Human-shaped enough to pass a filter, useless for pipeline.
• 7.8% of clicks were valid humans clicking from the wrong context — mostly LinkedIn's Audience Network placements on third-party mobile apps, where the click was almost certainly a mis-tap on a full-screen creative. Real people, no intent, no memory of clicking.

Add them up: 38.4% of paid clicks over six weeks were invalid, low-quality, or accidental. LinkedIn's own filter had flagged 2.1% of that.

Why the platform filter missed it

Platforms grade their own homework. LinkedIn's IVT filter is optimized to protect the auction — to keep obvious click-farms and known bad actors out of the marketplace. It is not optimized to tell a specific advertiser whether a specific click is worth the $12.40 they just paid for it. Those are different jobs.

The three failure modes above all sit in that gap:

• Residential-proxy datacenter traffic is a supply-side problem the platform has commercial reasons to under-report.
• Zero-dwell human-shaped clicks are indistinguishable from a real user with a fast bounce — the platform can't see the destination page, so it can't judge intent.
• Audience Network mis-taps are, technically, valid clicks. The user did tap the pixel. The platform's job ends there.

None of this is a scandal. It's the structural reason platform-reported click counts and independently-verified click counts never match, and never will. The advertiser is the one absorbing the delta.

How the customer found out

They didn't. We did, in week seven, when the receipts had accumulated enough signal to isolate the pattern by campaign, placement, and ASN. The customer's weekly report had gone out green every Friday for six weeks. Their SDR team had been chasing the resulting form fills — the ones that made it that far — and closing at roughly a third of the rate the previous quarter's list had closed at. Nobody had connected the two things, because nothing in the platform dashboard suggested they should.

The recovery cost

Rough math on the six-week window:

• Total spend: $42,180
• Verified invalid / low-intent spend at 38.4%: $16,197
• SDR hours spent chasing form fills sourced from the low-intent segment (estimated from the customer's own CRM tags): ~112 hours
• Pipeline attributed to the campaign in-quarter that traced back to verified-clean clicks: 87% of the total, from 61.6% of the spend. Pipeline attributed to the invalid segment: near zero.

The customer didn't lose $16,197 — they lost $16,197 and the SDR capacity that got burned chasing the fills it produced and six weeks of an optimization loop that was optimizing against a corrupted signal. That's the compounding part. You don't just pay for the bad clicks. You pay for every downstream decision made in the belief that the clicks were good.

What changed after

Two things, both small, both dull:

1. The customer added the Audience Network exclusion at the campaign level. That alone removed most of the mis-tap volume within 48 hours and dropped reported CTR by ~0.09 percentage points, which is what an honest CTR looks like on that account.
2. They started reconciling the weekly platform report against the receipt-level verification before the report went out on Friday, instead of after. The green status became a green status the CFO could actually sign off on.

The point

The platform wasn't lying. It was doing its job, which is not the same job the advertiser needs done. 38% invalid on a well-targeted, well-creative'd, LinkedIn-benchmark-beating B2B SaaS campaign is not an outlier — it's what you find when someone independent is finally counting. The uncomfortable part isn't the number. It's that it ran for six weeks before anyone checked.