Auto- strategies.

What algorithms do in advertising systems

Machine learning is a field in which a program learns from data and makes predictions without a human setting the rules for it by hand. The algorithm works through large arrays of information, finds patterns and uses them to make decisions.

In advertising it looks simple: the system sees who among those shown an ad clicked, who submitted an enquiry, at what time, from which device, after which queries. Then it looks for similar people and raises the bid where the probability of a target action is higher.

A human does the same thing, only more slowly and on a smaller volume of data. A specialist revises bids once a day; the algorithm does it at every auction. That difference in speed is the main reason automated strategies usually beat manual management.

Let us look at exactly what automation affects, when it works and when it gets in the way, and what remains the human’s area of responsibility.

Three tasks automation takes on

Bid management. The system decides how much to pay for an impression to a specific person at a specific moment. This is the most tangible part: taking dozens of signals into account at once by hand is impossible.

Audience selection. The algorithm extends reach to people similar to those who have already completed a target action, and cuts off segments that bring no results.

Budget allocation. Between campaigns, placements and formats - money flows to where the return is higher, without daily intervention by a specialist.

Additionally, automation selects combinations of headlines and texts, determines the time of display and decides who should be shown the ad again. All in all, this takes routine off the specialist and leaves them the work that carries meaning: what to offer, to whom and on what terms.

How it works in Russian systems

In Yandex Direct, automated strategies select bids for a set goal: get the maximum number of clicks or conversions within a budget, hold a set cost per conversion, reach a target share of advertising spend. The system also selects relevant queries and combines ad elements.

In VK, algorithms handle audience selection and adjustment of delivery: ads go to people whose behaviour resembles that of those who have already responded to your ads.

In both cases the principle is the same: you set a goal and a constraint, and the system looks for a way to achieve it. The key word is goal: if it is set wrongly, the algorithm will diligently optimise the wrong thing. A system tuned for clicks will bring clicks, not enquiries.

That is why the first step before switching on automation is to set up the transfer of target actions to the analytics system. Without this data, the algorithm has nothing to learn from. How to build data collection is covered in our article on analytics.

When automation works poorly

Automated strategies are not a universal solution. There are situations where they lose to manual management.

• Too little data. The algorithm needs a flow of target actions to find patterns. With a handful of conversions a month it learns from noise.
• A wrongly set goal. Optimising for enquiries when they are not passed to analytics turns into optimising for anything at all.
• Constant changes to settings. Every serious intervention resets the accumulated learning, and the ramp-up period starts over.
• A narrow or specific audience. When there are a few hundred buyers in the whole country, statistics never accumulate, and audience expansion brings untargeted impressions.
• A sharp change in conditions. A new range, a different price, a changed season - past patterns stop working, and the algorithm needs time to readjust.

A separate trap is expecting results in the first few days. Automated strategies have a learning period during which the figures are unstable. Conclusions drawn on day three are usually wrong.

What remains human work

The algorithm optimises what it was told to, within the conditions it was given. Everything that defines those conditions stays with the specialist.

The offer. No system compensates for a price above the market or a vague description of the service. The algorithm distributes impressions; it does not create value.

The landing page. Advertising brings a person to a page, and what happens there is beyond automation’s control. If the enquiry form is inconvenient, the algorithm will have nothing to learn from. The typical causes are covered in our article on a website that gets no enquiries.

Goal setting. The decision about what counts as a conversion is made by a human. And that decision determines everything else.

Checking the quality of enquiries. The system sees the fact of an enquiry, but not whether it was a genuine one or whether it led to a deal. Linking up with customer records closes this gap - more about it in our article on CRM.

Protection against wasted spend. Automation cannot tell a real visitor from a software one, so traffic quality control remains manual. How it works - in our article on fraud traffic.

The order for moving to automated strategies

1. Set up goals in analytics and make sure enquiries are passed correctly.
2. Accumulate statistics: the algorithm needs a flow of target actions, not isolated cases.
3. Start with one campaign rather than switching the whole account at once.
4. Set a cost-per-conversion or budget limit based on your economics, not on a desired figure.
5. Do not touch the settings during learning - let the system accumulate data.
6. Compare the result with the manual-management period by cost per target enquiry, not by the number of clicks.

This order protects against the main mistake - switching on automation and immediately starting to correct it. How to read the resulting figures is described in our article on data analysis and returns.

What this gives a small company

The main benefit for a small business is not saving budget but saving attention. Manual advertising management demands daily involvement, and an owner usually does not have that time.

The second effect is less dependence on qualifications. Automation evens out the result: it will not deliver the outstanding figure a strong specialist would achieve, but it will protect against gross mistakes.

The third is the ability to scale. Adding a new line of business does not require a proportional increase in the effort spent managing bids.

At the same time, automation does not do away with the specialist: it changes their work from setting bids to working with the offer, the audience and analytics. Campaign setup and the move to automated strategies are part of our advertising and contextual ads setup. It is also worth reading our breakdown of advertising setup for small businesses.