Marketing That Works: How to Train Advertising Algorithms to Drive Higher Profits

Manual advertising campaigns no longer guarantee consistent results. Google Ads and Meta algorithms have become more complex, and gaps in analytics have grown bigger. In this environment, merely launching an ad is no longer enough; you need to optimize it for profitability.

In this article, I will explain how to leverage data to transform your advertising budget from an expense into an investment. To illustrate my points, I will provide examples and insights from my experience at Netpeak.

This article is based on a talk by Oleksandr Konivnenko, the Head of Web Analytics at Netpeak Ukraine.

Why traditional approaches no longer work

Automatic bidding strategies, such as Target CPA, Target ROAS, and Maximize Conversion Value, have become standard in digital advertising. According to Google, over 80% of advertisers use these algorithms. At Meta, the figure is just as significant, at around 60–70%.

These strategies are part of an automated bidding system (Smart Bidding), which uses machine learning to decide who to show ads to, when to show them, at what price, and how to optimize for conversion.

The system initially determines the best moments to show ads by taking into account thousands of signals in real time.

However, these strategies have blind spots that limit their effectiveness.

Google Ads and Meta algorithms rely on three main types of signals:

• Contextual parameters, such as user geolocation, device type, interface language, time of day, and day of the week.

• Audience characteristics, including uploaded customer lists, demographic traits, or behavioral traits.

• Conversions themselves, which are events that the system considers valuable and on which bid optimization is based.

The last point is where the main problem lies: the algorithm does not understand the importance of each conversion to your business. To the system, all events are the same.

According to my observations, in the B2B segment, only 10–30% of requests turn into real deals. 

The situation is no better in retail. Some purchases are returned, while others are made offline and do not enter the system.

There are also technical factors to consider: data loss due to blockers, browser updates, and unstable tracking.

As a result, the algorithm learns from incomplete or irrelevant data. It optimizes campaigns for the wrong users, actions, and value. In this mode, advertising is done blindly: budgets are spent, but campaign effectiveness declines or stagnates.

This requires a rethinking of the approach. What was once sufficient — launch an ad, set some goals, and just wait for results — is no longer enough. It's necessary to teach the system what constitutes a successful conversion for the business and to work with that data.

Which signals cause algorithms to work for profit?

In modern marketing, a high number of conversions alone does not provide a competitive advantage. To effectively bring in profitable customers, platforms need accurate, business-relevant signals. If you send events every day to Google Ads or Meta that do not reflect the real value of the user, the algorithm will optimize for these simpler, cheaper actions even though they do not generate revenue.

To change this, you need to send not just events but verified data from your CRM:

• Qualified leads

• Purchases including returns

• Repeat orders and actual offline sales

However, transferring such data is a technically complex process. Technical specifications must be created, integrations agreed upon, and APIs configured. This process often takes a long time, and developers need to be involved. Some teams may even become demotivated due to communication gaps if technical specialists do not understand how advertising works and marketers do not understand how CRM works.

An alternative solution is to integrate CRM with cloud storage, such as Google BigQuery, and store cleaned, confirmed events there. This data can then be easily transferred to Google Ads via a ready-made integration and to Meta via the Conversion API. As a result, the system will start working based on the quality of conversions rather than the quantity and generate real revenue, not just clicks.

Setting bids based on conversion value: moving from spending to investing

Although automatic bid optimization can generate profits, it can only do so if the algorithm understands how those profits are derived. Value-based bidding is an approach that allows you to convey not just the occurrence of an event (e.g., a purchase) to advertising systems, but also its real value to the business.

This could be any of the following:

• Margin, not total revenue.

• The predicted lifetime value (LTV) of a customer, calculated based on their behavior.

• The probability of conversion, calculated from a model based on historical data.

In lead generation or SaaS, for example, there is no set price per conversion. However, you can pass on the modeled value of each lead, depending on its parameters or stage in the funnel. This provides algorithms with a more accurate understanding of which users are more profitable and on whom to focus.

One way to implement value-based bidding is to use the server-side version of Google Tag Manager (server-side GTM). Unlike classic GTM, which runs in a browser and depends on blockers and front-end stability, server-side GTM moves tracking to a server or the cloud.

First, events are sent to your server, where they are securely processed and supplemented with necessary values, such as margin or RFM segment. Then, they are encrypted and transferred to Google Ads or Meta.