A falling CPL can actually hide deteriorating marketing performance — and this is one of the most important things we learned while running a meaningful-scale lead-generation account for a payroll software brand. Advize is an AI-powered performance marketing agency that tracks downstream business outcomes alongside platform metrics, because the dashboard that turns green while the business gets worse is a real and recurring failure mode in performance marketing. Here is exactly what happened, why it happened, and what we think it means for any team using CPL as a primary success metric.
What happened when our CPL dropped dramatically?
Our CPL dropped significantly after we introduced sale and discount-based creatives into the account. From inside Ads Manager, these looked like the strongest performers we had run. Cost per lead was down sharply. Volume was up. Budget was flowing toward them automatically because Meta correctly identified them as producing the cheapest leads in the account. The problem became visible only because the client gave us unusual access to their downstream funnel data — and what that data showed was that the cheap leads were disproportionately low quality. The leads were arriving. They were not converting into the outcomes the business cared about. The platform metric was improving. The business outcome was not.
Why did cheaper leads turn out to be lower quality?
We think several things happened simultaneously, though we cannot prove exact causality for each.
First, the sale messaging attracted people who were responding to the offer rather than to the product. Someone who clicks because they see a discount is not necessarily someone who needs payroll software.
Second, some creative executions may have been ambiguous. A face-scanning attendance device can visually resemble a consumer phone screen in certain static formats. Combined with aggressive promotional language, some users may have interpreted the advertisement as a consumer product offer rather than a B2B software solution.
Third — and most importantly — we had given Meta a reward function that said 'cheap lead = good.' Meta is extremely effective at fulfilling its objective. It found people who would submit a lead form cheaply. Those people were not necessarily the right people.
The mechanism behind all of this is the same: the platform optimizes whatever you reward it for. If you reward it for cheap leads, it will find cheap leads. It will not independently decide that those leads should also convert into customers.
How does Meta's optimization loop reinforce the wrong behaviour?
Once the sale creatives began generating cheap leads, Meta's algorithm started allocating more spend toward that type of traffic. This was not a mistake. It was the algorithm doing its job correctly.
We switched off the sale creatives and returned to historically better-performing formats. But this alone was not sufficient to fix the problem. Meta had been trained over several weeks that the profile of person who submits a cheap lead is the target audience. Simply changing the creatives while keeping the same optimization objective meant we were still sending Meta the same reward signal — a lead, regardless of quality.
This is the part that surprised us the most when we first worked through it. You cannot tell an optimization system that Metric A is success and then be surprised when it pursues Metric A aggressively, even at the expense of Metric B. The system is not capable of caring about Metric B unless you specifically reward it for Metric B.
What is the right optimization signal for a B2B lead-generation account?
We tested several conceptual approaches.
Option 1 — Optimize for the lead event. Fast feedback. High volume. Weak correlation with business value when the lead definition is broad.
Option 2 — Optimize for a final sales or demo event. Strong correlation with business value. Too infrequent and too delayed. Meta does not receive enough signal fast enough to learn effectively.
Option 3 — Optimize for a qualified or relevant lead event. Something between the two extremes.
The third approach worked best in our experience. A relevant lead — defined as a lead that passes a basic qualification filter, such as company size or use-case fit — provided a signal that was:
- closer to actual business value than a raw CPL
- frequent enough for Meta to learn from it
- fast enough to return to the platform without long delays
- less noisy than final sale outcomes
This produced what we now consider one of our strongest learnings in Meta advertising for B2B: the best optimization event is not necessarily the deepest conversion event. It is the deepest reliable business-value proxy that still gives the algorithm enough frequent and timely feedback to learn from.
What should marketers do when they suspect their CPL is hiding lead quality problems?
Three steps consistently help.
Step 1: Get downstream visibility before the problem becomes obvious. Ask the client for lead-to-demo, lead-to-qualified, or lead-to-sale data from the start of the engagement. If that data cannot be shared, the agency cannot distinguish between a good CPL and a misleading one.
Step 2: Audit creative for audience mismatch risk. Look at which specific ads are generating the cheapest leads and ask honestly whether those ads could be attracting the wrong audience. Look at visual ambiguity, offer-driven versus value-driven messaging, and tone.
Step 3: Consider whether the optimization event is actually a business proxy. If the event being optimized for is a generic form submission with no qualification, it is not a proxy for business value. Build a qualification step — even a simple one — and consider whether Meta can be rewarded for that event instead.
Conclusion
CPL is a useful signal. It is a poor north star. The account that generated cheap leads at scale was not performing better — it was performing more efficiently at the wrong objective. The lesson is not that CPL is useless. It is that CPL is only meaningful when paired with visibility into what happens to those leads after they are generated. Without that, an agency can optimise its dashboard while quietly degrading the client's business. We think about this every time we look at a lead-generation account now.
