Lots of leads but no sales: how to train the algorithm to find real customers, not filled-in forms

A familiar situation: advertising brings in a pile of inquiries, the cost per lead is laughably low, but there are almost no sales - the leads are "junk," they don't reply, they don't buy. Usually the ads or the sales team get the blame. But more often the reason runs deeper: you taught the algorithm to look for the wrong people. Let's break down why this happens and how to fix it - by teaching your advertising to focus on real customers rather than on the fact that a form was filled in.

Why cheap leads often turn out to be junk

Meta's and Google's algorithms optimize for the event you gave them. If your goal is "a lead" (a filled-in form), the system will look for people who happily fill in forms. And form-fillers and actual buyers are, unfortunately, not always the same people. As a result, you get cheap leads in large quantities but of low quality: the algorithm honestly did its job - the job was just set incorrectly.

Simply put, you said "bring me those who leave inquiries" - and it does. What you should have said was "bring me those who ultimately buy."

The key idea: train the algorithm on real customers

The solution is to show the platform what happens to a lead after the form: did they become qualified, did they reach a deal, how much money did they bring in. These events often happen offline - in the CRM, over the phone, in chat - which is why they're called offline conversions. When you return this data back to your ad account, the algorithm starts to understand which leads were "real" and learns to look for people similar to them.

This flips the logic: you optimize your advertising not for the number of inquiries but for the number of customers. And both major platforms provide official tools for this.

How it works in Meta

Meta has an integration of the Conversions API with your CRM for this. The scheme is two-way: leads from Facebook and Instagram forms flow into your CRM, and from there you send back to Meta events about what happened to them - "qualified lead," "deal." This lets you enable the "Conversion Leads" optimization goal, that is, to aim your advertising at leads that are more likely to turn into buyers.

For this to work, Meta recommends a certain volume and discipline: use Lead Ads forms, generate roughly 200+ leads a month, upload status data at least once a day, and account for conversions that happen within 28 days of the inquiry. Simply put, the platform needs enough examples to learn from.

How it works in Google

In Google the mechanism is similar - it's importing offline conversions. When someone clicks an ad, a special identifier (GCLID) is recorded, and when the lead later turns into a customer in your CRM, you return this identifier along with information about the deal. That's how Google understands which ads and keywords bring in real customers and optimizes bids for them.

A more accurate and modern option is enhanced conversions for leads: here, hashed (encrypted) data such as an email from the form is used for matching. According to Google, using first-party data together with the click identifier delivered on average 10% more tracked conversions compared to a standard import - meaning the picture becomes more accurate, and the algorithm has something to learn from.

What this gives your business

The effect is simple and tangible: over time you get fewer inquiries, but noticeably higher-quality ones, and the real cost of acquiring a customer (not a lead) drops. The advertising stops chasing "form lovers" and starts bringing in people similar to your actual buyers. For a business with a sales team, this is also a relief for the reps: fewer empty calls, more warm ones. In essence, you turn your CRM into a "teacher" for the algorithm.

What's needed for this to work

A few conditions, without which there's no magic.

• A CRM (or at least a system) that records the fate of every lead - who is qualified, who bought, and for how much.
• Volume. The algorithm needs examples: Meta's benchmark is 200+ leads a month. With very small volumes, learning goes slowly.
• Discipline and time. Status data must be uploaded regularly (at least once a day), and the algorithm itself needs two to three weeks to learn without you tweaking the settings.
• Clean data and attention to privacy. Data is transmitted in hashed form, but you're still working with customers' personal information - for a business in the EU, this is a matter of consent and correct processing that's worth keeping in order.

Where to start

Start simple: define what "a quality lead" and "a customer" mean for you, and make sure you record this in your CRM. Then set up sending these events back - via the Conversions API in Meta and importing offline conversions (preferably enhanced conversions for leads) in Google. After that, switch your campaign goal from "inquiries" to "quality leads/conversions" and give the system time to relearn. From there, the algorithm will increasingly find exactly those who buy.

The key points in brief

If you optimize advertising for filled-in forms, the algorithm will look for people who love filling in forms, not buyers. To get quality leads, you need to return to the platform data about what happened to the lead next - whether they became a customer. Meta does this through the Conversions API integration with a CRM, Google through importing offline conversions and enhanced conversions for leads. The result: fewer junk inquiries, a lower real cost per customer, and advertising that learns from your actual buyers. You need a CRM, data volume, consistency, and a bit of patience - but this is one of the strongest improvements a business can make in paid advertising.