The Problem with Standard “Conversion Optimization”
Most advertising platforms offer automated conversion optimization: the system learns which clicks turn into customers and shifts budget there. The problem begins with the definition. “Conversion” is a loose term, and a filled-out form, an incoming lead, and a button click are all counted as conversions, even if only a small fraction of them turns into a paying customer. The system receives a success signal every time a conversion is recorded, so it chases whatever is easy to get.
When optimization relies on raw conversions, the algorithm learns to bring in many cheap forms rather than good customers. It doesn’t distinguish between someone who wants to buy and someone who filled out a form by mistake or out of curiosity. The result: an impressively low cost-per-conversion in the report, and a marketing budget that fails to fill the sales pipeline.
Many managers discover the gap only when the sales team complains that the leads “lack quality,” while the ad report shows excellent results. Both sides are reading real data, just not the same metric.
Why Cost-Per-Registrant and Not Cost-Per-Conversion
Our methodology relies on a single metric: cost-per-registrant. A registrant is a lead that has passed a threshold check—such as verified details, an answer to a qualification question, or an initial call—not just any form filled out on the site. Once this becomes the success metric, every dollar in the campaign is measured against a business outcome, not just a marketing one. The business sets the threshold, not the advertising platform.
The difference between cost-per-conversion and cost-per-registrant can be two or three times, and sometimes more. A campaign showing an exceptionally low cost-per-conversion sometimes generates the weakest leads of all: they are “cheap” to acquire because they were never serious to begin with, and no one qualifies them before they are counted as a success.
The threshold definition itself varies from business to business, and that is the part that requires an understanding of the industry, not just access to an ad platform. For a service-based client, the threshold might be a completed phone call, while for an e-commerce client, it is a completed payment rather than an abandoned cart. Even two businesses in the same industry will define different thresholds because their sales cycles differ or because the lead quality they are willing to accept varies.
Where AI Enters the Picture
The models we run in production, primarily Gemini and Claude, do not replace the campaign manager’s judgment. They speed up data analysis. Instead of manually reviewing hundreds of rows of audience, ad, and placement performance, the system flags within minutes which ad sets are generating a high cost-per-registrant. The flag is a starting point for investigation, not a command to shut down an ad set.
This changes the pace. A weekly review discovers an issue after it has already wasted significant budget, whereas monitoring that identifies an anomaly within a day or two allows budget to be shifted before losses mount. In a campaign with a high daily budget, the difference between an immediate response and an end-of-week response can amount to thousands of shekels a month.
There are also patterns that are hard to catch with the naked eye: which combinations of audience, time of day, and device type produce the cheapest registrants, even when none of the individual factors look unusual on their own.
What an Advertising Cycle Looks Like for Us
Every advertising cycle is built around one question: how much did each registrant cost this week, and why? We check by campaign, by ad set, and by audience, comparing against the previous cycle to identify trends rather than a one-off snapshot. We ask this question even during weeks when results appear good.
- Collection: Lead data from all channels connects with advertising data, so cost-per-registrant is calculated rather than estimated based on assumptions.
- Analysis: Where cost-per-registrant exceeded the defined threshold, and why: fatigued audience, fading creative, or increased market competition.
- Action: Budget is shifted from expensive sets to cheaper ones before the automated system continues feeding them based on the wrong metric.
- Documentation: Every change is logged so the next cycle can pick up right where the previous one left off.
The cycle repeats every week, accumulating over time into a clear picture of what works on each channel, rather than a broad average that hides issues in specific ad sets. After a few months like this, seasonal patterns also emerge: which weeks of the year cost-per-registrant naturally rises in the industry, making it possible to plan budgets in advance instead of panicking over every increase as if it were an anomaly.
When a Human Decides, Not the Algorithm
AI is good at identifying patterns across volumes of data that a human could never review manually. It is less adept at decisions requiring business context: whether to invest in a new audience with no historical data, how to respond to market shifts, and when to pause a campaign whose numbers still look reasonable on paper.
That is why a human campaign manager approves every significant budget adjustment on our end. The AI surfaces the insight, the human makes the decision—factoring in things the system cannot see, such as a competitor’s promotion or a seasonal shift in demand. This also prevents the common pitfall of blindly optimizing against the wrong metric, the kind that produces a pretty report with no actual customers behind it.
A Methodology That Can’t Be Copied in a Single Click
A competitor can copy a campaign structure or creative concept in a day. It is far harder to replicate a process where every dollar is evaluated against actual cost-per-registrant in every single advertising cycle, with accumulated insight into what works specifically for each client. A new client does not receive the process built for someone else, but rather the same methodology run fresh on their own data. Ultimately, this determines whether an ad budget turns into customers or remains just a pretty number on a report that no one in management trusts.
This is how we manage Google and Meta campaigns for our clients: a single metric, a consistent cycle, and human oversight on every budget-altering decision.