97% Lower CPL: What Paid Media Did, and What Agents Did

The short answer

The 97% figure is what AfM's published case record states, but the old title attached it to the wrong cause. The reduction comes from paid media on LinkedIn Ads and then Meta Ads for an anonymised global technology platform, run by Jakub as a freelance consultant from Q4 2024 to Q3 2025, almost entirely before he founded AfM in August 2025. An AI agent was built in the same engagement, but its job was to research and file the leads once they had arrived. It did not buy media, choose audiences or set bids, so it cannot be credited with the lower cost per lead.

And most of the 97% is not a campaign getting better on one platform. It is the difference between one platform and another. The rest of this article shows why, using only the figures in the record.

What the record says

Every figure below is quoted from the case record 97% Lower Cost Per Lead Over About 11 Months, $110K+ Managed.

Paid media engagement: figures as published

Item	As stated in the case record
Client	Global Technology Platform (Anonymised), DACH division
Programme	Recruiting agencies, entrepreneurs and digital professionals into a creator network programme in Germany, Austria and Switzerland
Timeframe	November 2024 to September 2025, about 11 months (Q4 2024 to Q3 2025)
Starting point	Nothing to inherit: the engagement started from scratch in Q4 2024 on LinkedIn only
Baseline	$270+ cost per lead in that first quarter, LinkedIn only
Q2 2025	Budget cut by 54% with no creative refresh; cost per lead rose
Q3 2025	Meta launched as the main channel; 1,606 leads at $8.86 blended on $14,224 of spend
LinkedIn in Q3 2025	Still $201 a lead
Best segment	$7.12 cost per lead in Q3 2025
Totals	1,700+ leads; $110K+ ad spend managed

The headline arithmetic holds: $270 to $8.86 is a fall of 96.7%, which rounds to 97%. The Q3 figure also checks: $14,224 divided by 1,606 leads is $8.86.

Where the 97% actually came from

Split the change into two parts.

Like for like, on LinkedIn. The first quarter on LinkedIn cost about $270 a lead. The last quarter on LinkedIn cost $201. That is a fall of about 26% on the same platform over the engagement: real, and the kind of gain that audience work, lead form changes and creative rotation produce.

The platform switch. In Q3 2025 Meta became the main channel and the blended figure across both platforms was $8.86. A blend sits between its parts, so if LinkedIn was $201, Meta must have come in below $8.86, and most of that quarter's 1,606 leads must have come through Meta. The remaining distance, from about $200 to under $9, is the difference between two auctions reaching the same professional audience in this market.

So the honest description is: the first quarter on LinkedIn compared with the last quarter across Meta and LinkedIn. The record says this itself: "Most of the fall in cost came from the move to Meta." It was not a straight line either. When the budget was cut by 54% in Q2 2025 with no creative refresh, cost per lead went up before it came down.

What work was done each quarter

1. • Q4 2024, foundation. LinkedIn only, video-first creative, and a first read on which audiences responded.
2. • Q1 2025, lead forms. The record credits lead form optimisation with lowering cost per lead on LinkedIn.
3. • Q2 2025, constraint. A 54% budget cut with no new creative; cost per lead rose.
4. • Q3 2025, scale. Meta launched as the main channel, audiences were split into segments (agencies, freelancers, entrepreneurs, business owners, streamers), and budget was weighted across Germany, Austria and Switzerland by response rather than population. The best segment reached $7.12 a lead.

AI appears in this work as a production aid: the record says AI-assisted copy in German and English raised the number of creative variants that could be tested each cycle. The weekly decisions on budget, audiences and which variants to pause were made by the person running the account. Nothing in the record describes an autonomous agent running the campaigns.

Where the agent comes in

The companion record, Inbound Lead Processing and Enrichment Automation, covers the same engagement. When Meta launched in Q3 2025 and lead volume jumped, leads were being exported by hand, researched and typed into the CRM, and follow-up had slipped to "1-3+ Days".

The fix was a webhook that fires on each Meta form submission, a custom Relevance AI agent that researches the company, checks the contact and assesses programme fit, then a CRM write with standard field mapping, duplicate detection and a notification for the strongest leads. The record reports response time moving from "1-3+ Days" to "Same Day" and "1,600+" leads processed in the peak quarter, with leads that cannot be enriched confidently flagged for manual review rather than pushed through.

So the honest version of "autonomous agents for lead generation" here is narrower and more useful: an agent that removed a processing bottleneck, which would otherwise have left the cheaper leads waiting days for a reply.

What the record cannot tell you

• • Lead quality. The record holds no enrolment or lead-quality figures and says so. A Meta lead form and a LinkedIn lead form do not necessarily produce leads of the same quality, so a cheaper lead is not proven to be an equally good one.
• • Contribution of each change. Channel mix, segmentation and creative testing happened together. There was no holdout group left on the old approach, so their separate effects cannot be isolated, and market changes over the period cannot be ruled out.
• • Spend by quarter. Only two quarters' spend is itemised, so an engagement-wide average cost per lead cannot be rebuilt from the published figures.

None of this makes the result false. It tells you what the number can carry: a large, documented fall from an uneconomic first quarter, driven mainly by a platform decision, with the causes named by the person who did the work rather than isolated by experiment.

What you can take from it

• • Test a second auction before optimising the first to death. A 26% gain on LinkedIn was worth having; the move to Meta was worth far more. If one platform has never been viable for you, the cheapest test is often another platform, not another round of creative.
• • Report cost per lead by platform, not only blended. A blended figure falls when budget shifts to the cheaper channel even if neither improves. Without the $201 LinkedIn figure, this case would read as a miracle of optimisation.
• • Measure past the form. This record stops at the lead. Before you trust a cheaper lead, follow it to a meeting or a sale.
• • Fix processing before you scale acquisition. The automation record exists because cheaper leads created a backlog. If follow-up is already slow, a lower cost per lead will not become more customers until that is solved.

To test your own numbers this way, the B2B lead generation cost calculator separates cost per lead from cost per qualified lead and cost per opportunity, which is the comparison this case could not make. For how AfM runs this kind of work today as a managed service, see how AfM works.

FAQ

Was the 97% reduction achieved by an autonomous AI system?

No. The case record describes campaigns run with weekly human decisions, using AI to produce more creative variants. The AI agent in the companion record processed leads after they arrived and did not affect what each lead cost to acquire.

Was the $270 baseline inherited from a previous team?

No. The record says there was nothing to inherit and that the engagement started from scratch. The $270+ is the engagement's own first quarter, Q4 2024, on LinkedIn only.

Is $8.86 the average cost per lead across the whole engagement?

No. It is the blended figure for the final quarter, Q3 2025, across Meta and LinkedIn. Earlier quarters were far more expensive, and only two quarters' spend is itemised, so an engagement-wide average cannot be calculated from the record.

Was this an AfM client result?

It is work Jakub delivered as a freelance consultant, almost entirely before he founded AfM in August 2025. AfM publishes it as the founder's track record and labels it that way on the case page.

Would the same approach work for a UK B2B company?

The method transfers: test a second platform, segment audiences, rotate creative, report by platform and measure beyond the form. The size of the effect depends on how poor the starting point is and how differently the platforms price your audience, so an already efficient campaign should expect a much smaller change.