How AI-Powered Ad Platforms Predict Demand Before Users Express It

Advertising platforms have learned to find potential customers before they even enter a search query or visit a product page. Yandex, Google, and Meta have implemented recommendation systems that analyze thousands of user actions — from checking the weather in another city to lingering on a product card — and identify patterns of emerging demand. ARGUS from Yandex processes up to 8000 user events instead of the previous 256, and the company's neural technologies increased the overall efficiency of their advertising system by 38% in 2025. For brands, this means a new approach to media buying: reach expands through audiences that haven't yet shown explicit interest, but the system already forecasts their readiness to convert.

Why traditional targeting no longer gets the job done

The traditional digital advertising model was built on obvious signals: a person searched for running shoes, visited a mortgage calculator page, or selected hotels in a specific city. The platform ranked ready-made creatives based on relevance forecast and auction bid. The problem is that the path to purchase is rarely linear. A user might check the weather forecast in Rome, read about visa requirements, and close the tab — without a single explicit commercial query. On their own, these actions mean nothing, but in the right sequence they signal trip planning.

Earlier algorithms couldn't model long and heterogeneous chains of behavior. They relied on recent actions, a limited set of signals, or broad demographic segments. Infrastructure based on recommendation systems analyzes how interests form, intersect, and transform over time. The order of events, context, and feedback on how users react to past recommendations all matter.

How recommendation models forecast demand

Performance Max from Google manages bids, audiences, creatives, and attribution through a single system. Non-retail advertisers see on average 27% more conversions or conversion value at comparable CPA or ROAS. The system considers not only current behavior but also predicts which users will deliver long-term value. Brands can share data about their most valuable customers, helping the algorithm prioritize similar segments.

Meta's Generative Ads Recommendation Model (GEM) delivered a 5% conversion lift on Instagram and 3% on Facebook feed right after launch. The model processes behavioral sequences together with ad information, formats, advertiser goals, and engagement. The system updates recommendations as new signals arrive — one product card view can change the set of interests, the next action will change them again.

Advertising platforms no longer wait for an explicit request — they recognize potential earlier and influence demand formation

Yandex's multi-head Gorgona architecture uses separate neural networks to evaluate different types of conversion signals, predict outcomes, and manage bids. The discriminative approach lets the system respond faster to changes and distinguish casual interest from a signal that could deliver business results. Ad systems have to interpret more types of anonymized signals than streaming services or social networks: search, e-commerce, mobile apps, partner platforms. User feedback is less obvious — people don't like ads to explain their relevance.

Creative generation as the final piece

Once the platform recognizes commercial potential, a suitable ad variant might not exist. Amazon Ads uses product information and advertiser creatives to create or adapt images, video, audio, and text — including within Amazon DSP. Sponsored Brands campaigns with generated creatives showed on average 10.3% higher return on ad spend in the second quarter of 2025. Brands using generation tools promoted five times more products and used twice as many images per product.

Instead of manually preparing each ad, an advertiser can provide a product card or feed, source creatives, brand requirements, and key messages. The platform adapts the ad to format, channel, customer journey stage, or familiar tone. Not every impression requires a unique generated ad — brands still need consistent variants, legal compliance, and control over key messages. The important change is that content creation becomes part of analyzing audience behavioral signals and showing ads.

What this changes for the Russian market and media buying

More accurate forecasting expands the reach of potential customers an advertiser can reach. Instead of reacting only to an explicit purchase signal, the platform recognizes situations where the probability of a person needing a certain product or service is just forming. Advertising economics aren't limited to the first conversion. A customer who makes repeat purchases, increases average order value, or stays loyal to a company for years can justify a higher acquisition cost than someone who buys once.

Higher efficiency doesn't always mean lower cost for the same result. What matters more is increased revenue and conversion value per ruble spent on promotion. When a platform brings more real customers and reduces costs from irrelevant impressions, it becomes profitable and attractive to business. As a result, advertisers are more likely to increase spending — not because they pay more for the same results, but because the system brought more profit.

Marketer's checklist: how to adapt your strategy

• Share data about your most valuable customers with the platform — CRM segments with high LTV will help algorithms prioritize similar audiences at early stages of demand formation.
• Prepare a product or service feed with complete information — the more attributes (descriptions, specifications, visual materials), the more accurately the system can generate and adapt creatives to the context of each impression.
• Define acceptable customer acquisition cost based on long-term customer value, not just the first conversion — this will expand your available audience volume and let the system work with emerging demand.
• Set conversion goals across the entire customer journey — share data on repeat purchases, average order value, service usage duration with the platform so algorithms optimize beyond just the first touch.
• Regularly update brand requirements and key messages — even with automatic creative generation, you maintain control over tone and brand positioning.
• Analyze not just CPM and CPA, but revenue per ruble invested — efficiency metrics should reflect how much profit the system generated, not just how much a click or impression cost.
• Test different formats and channels within a single campaign — Performance Max, ARGUS, and similar solutions automatically distribute budget to where the system finds more potential customers.

Frequently asked questions

How do recommendation systems in advertising differ from social networks and marketplaces

Advertising platforms interpret more types of anonymized signals — search, e-commerce, mobile apps, partner platforms — and receive less clear feedback from users. People don't like ads to explain their relevance, so the system has to draw more conclusions from indirect signals and their sequence. Streaming services and social networks study sequences in a relatively uniform environment: tracks, playlists, posts, short vertical videos, stories.

Why should a brand share data about valuable customers with the platform

Information about your most valuable customers helps algorithms prioritize users who are more likely to deliver long-term value. The system analyzes behavior patterns of this segment and finds similar audiences at early stages of demand formation, when competition for impressions is lower. This expands reach and lets you work with customers who haven't yet shown explicit interest but already match the profile of a high-margin buyer.

What metrics matter most when working with recommendation systems

Revenue and conversion value per ruble invested matter more than CPM or CPA in isolation. A system may bring customers at a higher acquisition cost, but if they make repeat purchases, increase average order value, or remain loyal to the company for years, overall profitability will be higher. Analyze not only the first conversion, but also the LTV of segments that came through recommendation campaigns.

* Instagram and Facebook are owned by Meta, recognized as an extremist organization whose activities are prohibited in the Russian Federation.