Guest
This article is written by Anastasiya Starovoytyva, Senior Content Marketing Manager at SplitMetrics — a mobile marketing performance company helping Fortune 100 companies achieve 25%+ revenue growth on Apple Ads and beyond.
The more Apple Ads campaigns you run, the more decisions there are to make. Which bids should you change? Should you focus on lowering acquisition costs today, or on the revenue those users might generate over time? And, once you’ve made those decisions, you still have to actually make the changes.
As part of our work in mobile marketing performance, we at SplitMetrics wanted to put a number on one of the most routine tasks in Apple Ads: changing a bid. We found that a single change takes around 1.5 minutes to analyze and execute. At 1,000 changes, that’s roughly 25 hours of work.
That made us wonder: can this task actually be automated? And if so, what kind of AI can help, and how much of the optimization process can it really take on?
Let’s explore the topic in more detail.
How Apple Ads optimization works
In Apple Ads, you set a maximum CPT bid for each keyword. But deciding what that bid should be depends on several factors:
- Efficiency: CPA, CPI, ROAS, conversion rate and conversion volume
- Масштаб: impression share, search volume and room to grow
- Budget: whether a campaign is constrained or spending on weaker opportunities
- Конкурс: whether a higher bid could bring more impressions
- Seasonality: whether demand is likely to change
- Downstream results: trials, subscriptions or revenue that may only become clear later
At a small scale, you can make these decisions on your own without spending too much time on them. But across thousands of keywords and a large portfolio, doing this repeatedly becomes a significant amount of work.
Why automated rules may not be enough at scale
If you’re running Apple Ads and using a campaign management platform, you’ve probably used automated rules. They can take care of straightforward changes based on the thresholds you set.
However, those thresholds have to be defined in advance. Once you start adding different conditions, you end up with a growing set of rules to maintain. And even then, they only respond to the conditions you’ve anticipated.
A rule might say to increase a bid when CPA is below a certain threshold. But it doesn’t necessarily account for whether the keyword has enough conversion volume to make that signal reliable or whether the additional conversions are likely to have lower downstream value.
How AI can help with Apple Ads optimization
Instead of relying entirely on a fixed set of rules, an AI-driven optimization system can assess campaign data, estimate the likely outcome of different actions and determine what to do next.
For Apple Ads optimization, that can mean:
- Adjusting bids. Increase or decrease CPT bids based on performance and the expected opportunity for additional conversions.
- Moving budget. Shift budget toward campaigns with stronger opportunities while staying within spending limits.
- Scaling keywords. Identify keywords that perform well and still have room to generate more volume.
- Reducing wasted spend. Lower bids on keywords that consistently fail to meet the account’s efficiency target.
- Responding to market shifts. React to changes in impression share, conversion rates, competition and seasonality without waiting for the next manual review.
- Optimizing for business outcomes: Look beyond immediate acquisition metrics and optimize toward the outcomes that matter most to the business, such as CPA, CPI, ROAS, trial starts, subscriptions or downstream revenue.
Two types of AI are particularly helpful here: predictive AI, which forecasts what will happen, and prescriptive AI, which decides what to do.
Predictive AI: forecasting the outcome
Predictive AI uses historical and current campaign data to estimate how a bid change could affect impressions, spend, conversions, CPA or other metrics.
For example, a keyword with a $10 CPA might generate more conversions if its bid increases — but the additional conversions could come at a much higher cost. A predictive model can estimate the expected outcome at different bid levels before the change is made.
The same applies to budgets. If one campaign is constrained while another is consistently underspending, the system can estimate where additional budget is most likely to generate incremental volume.
Prescriptive AI: choosing the action
If a model predicts that increasing a bid will generate more conversions but at a higher CPA, the system needs to determine whether that trade-off is reasonable. The answer may depend on the target CPA, available budget, expected revenue, conversion volume and other business constraints.
Prescriptive AI uses these predicted outcomes together with the campaign objective to determine the action. That might mean increasing or decreasing a bid, moving budget between campaigns or leaving a campaign unchanged.
Why combining both matters
Prediction without action leaves the decision with the marketer. Prescription without a strong prediction layer has limited information about what the proposed action is likely to achieve. Combining the two creates a continuous optimization loop:
Predict → decide → act → measure → learn.
The system forecasts the likely outcome of a change, determines the action that best supports the campaign objective, executes it, measures the result and uses the new data to improve future decisions.
Bids, competition, search volume, budgets and conversion performance change around the clock. A system that can both anticipate those changes and act on them can optimize campaigns 24/7 rather than relying on fixed rules or manual intervention.
What AI Apple Ads optimization looks like in practice
Gismart Piano is an example of how this approach can work in a real Apple Ads account.
The growth team used to manually review metrics such as CTR, conversions, CPI and post-install behavior to manage its campaigns. As they looked for ways to improve efficiency while growing, it tested an automated bid optimization approach using SplitMetrics’ Samba model family.
According to Gismart Piano, this new approach led to a 7% reduction in CPA, CPI below its KPI targets and a 56% increase in trial sign-ups within two weeks.
The team also spent less time on routine bid management and was able to put more of its time into things like keyword and geographic expansion.
Why not just use ChatGPT or Claude?
It’s reasonable to ask whether a general-purpose AI model such as ChatGPT or Claude could be used if you give them access to your account. They can, potentially, analyze campaign data, but they won’t be very helpful. They lack the infrastructure needed to optimize Apple Ads, including:
- Current Apple Ads campaign and keyword data
- Historical performance
- Campaign and keyword context
- Conversion and attribution data
- MMP and downstream revenue data
- Campaign objectives and business constraints
Without that context and access to live Apple Ads data, they fall back on generic advertising knowledge or simple assumptions.
Finally, an LLM typically stops at the recommendation, while a specialized optimization system can actually perform the job.
So, can AI help with Apple Ads at scale?
AI can handle an enormous volume of optimization work, and at a much quicker pace — up to 10× faster than a UA manager. Instead of manually reviewing and updating thousands of keyword bids, you can rely on AI to execute routine optimization decisions while working toward performance objectives such as CPA or ROAS.
However, the type of AI matters.
For Apple Ads optimization, predictive AI can estimate what is likely to happen after a bid or budget change, while prescriptive AI can use those predictions to determine the best action. When the two are connected, the system can make changes, measure the results, and learn from them.
A general-purpose AI alone doesn’t make the cut because it lacks the specialized signals, campaign context and execution capabilities needed to make and measure these decisions reliably.
The most practical approach is to build a system that can handle the day-to-day optimization work on its own, with you in control — and free to focus on a bigger-picture view.
