Chasing mid-market companies cost $130,468 a customer. Small companies on LinkedIn cost $35,288.
The jump starts above 500 employees.
In 2025, winning a customer cost $130,468 when the target was a 501–1,000-employee company, across 19 advertisers, and $80,109 at 1,001–5,000, across 23, against $58,887 across all 127 advertisers, where every ad dollar brought back 0.56x: 56 cents of closed-won revenue we can trace back to an ad, inside 12 months. One size group paid for itself, 51–200-employee companies on LinkedIn, 1.16x at $35,288 a customer across 27 advertisers, 3.7 times cheaper than the mid-market peak. Two rows are missing on purpose: for 51–200 and 201–500 across all channels we could not publish a cost per customer, because one advertiser held too much of the result, withheld (why).
Split every campaign that mixes company sizes this week, separate campaigns, separate budgets, separate bids. In 2025 a customer cost $35,288 on LinkedIn aimed at 51–200-employee companies (27 advertisers), $80,109 at 1,001–5,000 (23 advertisers) and $130,468 at 501–1,000 (19 advertisers). Then watch the weekly cost-per-customer line band by band and find where it bends. Two rows you will not find here: 51–200 and 201–500 across all channels, withheld (why) because one advertiser held too much of the result.
2025 · 19–31 advertisers per band · B2B · only groups with at least $50K spent · counted over 12 months · Methodology · What we withheld, and why
Executive summary
- The money: a customer cost $35,288 in the one group that paid for itself (51–200-employee companies on LinkedIn, 1.16x back, 27 advertisers) and $130,468 at 501–1,000 employees (19 advertisers), 3.7 times as much, against $58,887 and 0.56x across all 127 advertisers.
- The budget call: set the budget and the forecast per company-size band this quarter; one blended target hides the jump. Ask the small-company band to pay for itself and the mid-market to earn coverage, same budget, two scoreboards.
- The caveat: this is what happened in 2025, not a controlled test. 19 to 31 advertisers per band, and the two smallest all-channel bands are withheld (why) because one advertiser held too much of the result. It prices what a customer cost to win, not what that customer is worth over time.
Playbook
- The play: split every campaign that mixes company sizes, separate campaigns, separate budgets, separate bids.
- The setup: give 501–1,000-employee companies their own campaign and their own budget line, and measure what a customer costs there on its own.
- The measurement: cost per customer per band, weekly, with the deal count printed beside it. A number built on two deals is not a number yet. Where the line bends toward the $130,468 band, rotate audiences instead of scaling.
Where does the cost of winning a customer peak?
At 501-1,000 employees, where a customer cost $130,468 in 2025 across 19 advertisers, before easing to $80,109 at 1,001-5,000 across 23. The small-company end is where this report gets quieter: for 51-200 and 201-500 across all channels the cost per customer is withheld (why), because once the one over-weighted advertiser came out, a single advertiser held more than 40% of each band’s closed-won deals. What we can still publish at the small end is one channel inside it, 51-200-employee companies on LinkedIn: $35,288 a customer, 1.16x back on every dollar, and 19.5% of decided deals won, across 27 advertisers.
Read outward from that group and the shape is still a cliff: 3.7 times from the only band that paid for itself up to the mid-market peak, with all 127 advertisers sitting at $58,887 and 0.56x in between. The price of a lead, meanwhile, barely moves across the same range. What the cliff looks like is the mid-market buying committee, priced, and it is invisible to any report that stops at the lead.
Do cheap leads mean cheap customers?
Not here. Leads from 51-200-employee companies cost $241, more than the $204 that 1,001-5,000 leads cost, and yet the cheapest customer we published anywhere in this report comes from that same small band on LinkedIn, at $35,288 against $80,109 for the enterprise band. The lead price says the two are near-identical. The customer price says one costs roughly 2.3 times the other. A report that stops at the lead ranks them the wrong way round.
What should you change?
Write a cost-per-customer ceiling per company-size band into the annual plan: one blended target averaged a $35,288 group with a $130,468 one, across 27 and 19 advertisers. Past the cliff, “+20% more budget” is off the table, guardrails, not growth promises, at 501–1,000 employees.
Judge the 501–1,000 band on influenced pipeline and coverage for one quarter, and ask only the small-company band to pay for itself. Asking for payback is the wrong scoreboard where a customer costs $130,468; it is the right one where a customer costs $35,288 and brings back 1.16x.
“Skip the mid-market.” Nothing here prices what mid-market advertising influences. Only what it costs to close a deal you can trace. See the influenced lens in the explorer, and read it as a signal, not a result. Chasing an immediate return is the wrong tool for that band; it is not evidence against the band.
Split every campaign that mixes company sizes into one campaign per band, with its own budget and bid, this week: on one blended budget the $35,288 small-company group eats the spend while the $130,468 band hides inside the average. If you cannot split the report by band, that is your leading finding.
Watch the weekly cost-per-customer line, band by band, for a full quarter. At the bend, rotate placements and audiences instead of scaling the winning ad set. The cliff is where scaling the winner stops working.
“Bid down until the mid-market costs what small companies cost.” $130,468, across 19 advertisers, prices a buying committee, not a bidding mistake. Underbidding buys silence, not cheaper customers.
What are the exact numbers?
Every company-size row we can publish, plus the two we cannot. Across all 127 advertisers with revenue data ($29.4M of lead-generation spend, 154,000 leads matched to a CRM), every ad dollar brought back 0.56x at $58,887 a customer, with 21.2% of decided deals won.
| Group | Advertisers | Cost per customer | Traceable payback | Cost per lead |
|---|---|---|---|---|
| 51-200 employees × LinkedIn | 27 | $35,288 | 1.16x | — |
| 51-200 employees, all channels | 31 | withheld | withheld | $241 |
| 201-500 employees | 17 | withheld | withheld | — |
| 501-1,000 employees | 19 | $130,468 | — | — |
| 1,001-5,000 employees | 23 | $80,109 | — | $204 |
| All sizes, after the one over-weighted advertiser was removed | 127 | $58,887 | 0.56x | — |
Traceable payback = closed-won revenue we can trace back to an ad, divided by ad spend, inside 12 months. A dash means the number sits in the interactive explorer rather than on this page, or that the band did not have enough advertisers behind it to publish a payback figure of its own. Withheld means we did not publish it: on 2026-08-17 the 51-200 and 201-500 all-channel cost-per-customer numbers failed the rule that no one advertiser may hold more than 40% of a result. The reason, and what it cost us, are in the kill list. Cost per lead comes from $31.5M of lead-generation spend; cost per customer and payback from the $29.4M and 154,000 CRM-matched leads behind 127 advertisers, inside the $57.6M of 2025 spend we analyzed.
How do we know?
Every number above passed the same publication gates before it reached this page. In brief:
- How many advertisers are behind each number: 27 behind 51-200 × LinkedIn, 19 behind 501-1,000, 23 behind 1,001-5,000 and 127 behind the all-advertiser row, all above the 8-advertiser minimum we require before publishing a revenue number, and the count is printed beside every figure. The band above 5,000 employees is missing on purpose: too few advertisers, and one of them held too much of the result, so we refused to publish it.
- No one advertiser may swing a number, on either side: no single advertiser above 40% of a published group’s spend or above 40% of its closed-won deals. The deals half is what removed the 51-200 and 201-500 all-channel rows on 2026-08-17: once we excluded the advertiser whose CRM logs transactions rather than deals, one account held most of each band’s remaining wins. Withheld, not estimated, reasons in the kill list.
- Spend floor: at least $50,000 spent behind any published group.
- Scope and window: cost per customer and payback come from the $29.4M of lead-generation spend, 154,000 CRM-matched leads and 127 advertisers inside the $57.6M of 2025 spend we analyzed, followed through the CRM to closed-won revenue over the 12 months after the spend; cost per lead comes from $31.5M of lead-generation spend. We count revenue only where the ad-generated lead itself started the deal. That is traceable payback, not proof the ads caused the revenue, not profit, not lifetime value.
- What kind of evidence this is: every figure describes what happened in 2025, not a controlled test. It tells you what these advertisers experienced, not what is guaranteed to cause what.
- What else could explain this: who aims at each band. Advertisers targeting 501–1,000-employee companies may sell harder products at different prices, so the bands compare different sellers as well as different buyers. No single advertiser shapes a band, and every band has enough advertisers behind it to publish, but we did not control for which industries or deal sizes sat inside each one. The cliff is what we observed, not a proven cause.
Full rules in the Closed-Won Protocol, including the nine claims we refused to publish, with the exact reason each one was killed and the number it cost us.
Cite as: Metadata 2026 B2B Ad Spend Benchmark (153 advertisers, $57.6M, 2025), metadata.io/benchmark-report-2026
What do you do different tomorrow morning?
One named change per seat. The columns reorder to match your reading lens; both are always on the page.
Rebuild the board numbers
The one change: replace the single blended cost-per-customer target with one target per company-size band this quarter.
- Pull what a customer cost by target company size for the trailing two quarters and put it next to this page’s table. If you cannot split it, that is finding number one.
- The board story writes itself: “our blended number was averaging a $35,288 group with a $130,468 one, the blend was the fiction.” (27 and 19 advertisers)
- Move the 501-1,000 band’s goal off cost per customer and on to influenced pipeline and coverage, the explorer’s influenced lens is the honest scoreboard there.
Split everything by company size
The one change: break any campaign that mixes company sizes into separate campaigns with separate budgets and bids this week.
- Duplicate your current campaigns per band and never let small companies and the mid-market share a budget or a bid strategy. The cheap band will eat the spend and hide the expensive one.
- Watch what a customer costs in the 501-1,000 band on its own for a full quarter before judging it; blended reads are how the cliff stayed invisible.
- Do not rank bands on cost per lead: at $241 against $204 it calls them equal, while per customer the published numbers sit at $35,288 (51-200-employee companies on LinkedIn, 27 advertisers) and $80,109 (1,001-5,000, 23 advertisers).