This is a guest post by Peter Kireev, a leading expert with over a decade of experience in AdTech. He is the Co-founder and Chief Product Officer at Reliz, a product-led European company, where he leads the development of its AI-driven solutions.
In 2024, performance advertising is constrained by signal loss, stricter privacy laws, and longer monetization cycles. Traditional ROAS optimization — based on post-event data like subscriptions or in-app purchases — has become too slow and too narrow to support competitive growth.
Emerging AdTech platforms are countering signal loss with predictive analytics: they harness early-stage behavioural signals to estimate long-term value and set bids before any monetised action occurs. By converting insights gathered in the first moments of engagement into forward-looking campaign logic, these models move optimization from hindsight to foresight.
This shift is already visible across platforms that are experimenting with predictive models. This trend is especially relevant for products with delayed revenue (e.g., subscription-based apps, games, or fintech), where value can only be observed weeks after acquisition.
Instead of waiting for conversions, predictive ROI modeling enables platforms to optimize toward forecasted outcomes, increasing efficiency in both bidding and budgeting. While this method is still being tested and refined, it already defines the cutting edge of AdTech infrastructure in 2024.
Why predictive ROI emerged
Several structural shifts in the AdTech ecosystem made predictive ROI modeling a necessary response, not just an experiment.
First, attribution windows have shortened. Post-IDFA, platforms like Meta and Google Ads rely on limited, privacy-compliant signals. Many conversions now happen outside the window where algorithms can learn from them. As a result, campaigns optimized for actual events are either under-reporting value or prematurely shutting down.
Second, privacy-first regulations — including GDPR, CCPA, and Brazil’s LGPD — have severely limited access to cross-app identifiers and behavioural history. Platforms can no longer rely on third-party data to stitch together a user’s journey or value. According to the IAB’s State of Data 2024 report, nearly 90% of advertisers are already rethinking their personalization strategies and ad investments, adapting to a new reality where third-party data is becoming increasingly limited and less effective.
Third, monetization cycles have become longer. In subscription or fintech apps, value accrues gradually. But performance media still pays per click or install. This mismatch between payment and payoff creates a cash flow problem unless future value is predicted early.
Without predictive modeling, ad systems optimize toward short-term proxies like installs or early-stage engagement — which increasingly fail to correlate with actual revenue.
Predictive analytics shifts the optimization logic: from observed conversion to expected value. That shift allows platforms to buy higher quality users earlier and scale with confidence, even when signals are partial or delayed.
How predictive ROI works
Predictive ROI modeling starts with zero-day data — signals available immediately after a user interacts with an ad or product. These include:
- traffic source and campaign ID
- device and OS
- geolocation
- engagement with first screens
- time to first action
- app version or feature flags
This input is passed into a machine learning model trained on historical data from the advertiser’s own user base. The model estimates the expected value of each user — often expressed as predicted LTV over 7, 30, or 90 days.
The predicted value is then forwarded to the bidding system — either a proprietary DSP or major ad platforms. Instead of reporting only actual conversions, advertisers now send proxy revenue signals (e.g. “this user is worth $42 based on model prediction”).
This allows the bidding algorithm to target higher-value users from the start — even before real purchases happen — and to allocate budget more efficiently.
Some platforms integrate these predictions as real-time postbacks. Others inject them as conversion value overrides via APIs like Meta’s Conversion API or Google Ads Enhanced Conversions. The technical method varies, but the core idea is consistent: optimize toward forecasted business value, not just historical events.
Why it’s hard
Despite its promise, predictive ROI modeling is technically demanding and operationally risky.
The main challenge is accuracy under uncertainty. A small error in LTV prediction can cause disproportionate losses. If the model overestimates value, the bidding system will overspend on users who won’t convert. If it underestimates, the campaign will miss high-value users and under-deliver.
These models are also highly sensitive to training data. If historical cohorts are too narrow, unbalanced, or noisy — the predictions skew. Many advertisers still lack sufficient clean data to train reliable models, especially for newer products or underpenetrated markets.
Another complexity is integration. Predictive ROI only works when signals flow smoothly across product analytics, user segmentation, bidding platforms, and campaign attribution. Few marketing teams have the infrastructure to close that loop in near real-time.
Finally, performance teams often struggle with model transparency. Predictive systems may return a score or value, but not explain why. This makes manual troubleshooting or budget allocation decisions harder — especially in high-stakes verticals like finance or health.
As a result, early adopters need to treat predictive ROI not as a feature, but as an investment: in data quality, model governance, and cross-team alignment.
Who’s doing this
As of mid-2024, predictive ROI modeling is still emerging. Adoption is limited to a handful of product-led AdTech platforms and advanced in-house growth teams exploring alternatives to post-event optimization.
At Reliz, we’re actively running R&D on predictive pipelines — using early-stage user signals to estimate long-term value and feed real-time bidding logic. This includes training custom models on first-party product data, integrating predictions into campaign-level decisioning, and validating outputs against actual revenue trajectories.
Other independent players such as Lemon.ai and Pecan.ai are also investing in this space, building ML frameworks that prioritize probabilistic user valuation over deterministic attribution.
Meanwhile, platforms like Meta Ads and Google Ads have introduced technical capabilities to accept predicted conversion values via Conversions API and Enhanced Conversions, but the modeling responsibility remains on the advertiser or their AdTech partner.
Most traditional DSPs and agency networks are not yet equipped to run predictive-ROI pipelines at scale. The barriers already outlined earlier still curb adoption, making 2024 a foundation-building year rather than a breakout one. Teams that invest now in data readiness, model governance, and real-time integration will be the first to scale once those obstacles lift.
