PMAX BUDGET: HOW THE ALGORITHM ALLOCATES SPEND

The Performance Max budget is one of the elements advertisers manage most mechanically, you set a number, press save, and move on. But in PMax the budget is not just a spending cap: it is a signal the algorithm uses to calibrate its behaviour, together with the bid target and the conversion data available.

Understanding how PMax budget allocation works, that is, how the algorithm decides how much spend to direct to each channel, each auction, and which time windows, has concrete implications for campaign optimisation. A budget that is too low relative to the objective generates instability. A budget that is too high relative to the bid target creates inefficiencies. A correctly sized budget, on the other hand, gives the algorithm the flexibility it needs to operate efficiently.

In this article I analyse how PMax allocates spend across channels and auctions: starting from the official Google documentation, and noting where direct experience on real accounts leads to more precise conclusions about mechanisms Google describes only partially.

How Budget Works in Performance Max: the Official Version

According to the official Google Ads budget documentation (support.google.com/google-ads/answer/1704443), the budget for Google Ads campaigns, including PMax, is an average daily budget. Google can spend up to twice the daily budget set on days with higher traffic (over-delivery), compensating with lower spend on subsequent days. Over the course of the month, total spend does not exceed the daily budget multiplied by the number of days in the month.

For PMax campaigns specifically, the official documentation specifies that the budget is allocated automatically across all available channels (Search, Shopping, Display, YouTube, Gmail, Maps, Discover) based on the conversion objective. There is no manual control over the percentage of budget directed to each channel, this is one of the key differences compared to standard campaigns.

What the official documentation does not describe in detail is the mechanism by which the algorithm decides, auction by auction and channel by channel, what bid to assign and therefore how much budget to consume. This is where the concept of Performance Max budget signals comes in: the budget is not just an external constraint, it is a parameter the algorithm uses to calibrate its own bidding behaviour.

What experience shows: the daily budget set in PMax implicitly communicates to the algorithm the "volume of opportunity" the advertiser wants to capture. A very low budget relative to available opportunities does not just lead to spending less, it causes the algorithm to become more selective in the auctions it enters, effectively raising the implicit minimum bid for each placement. This mechanism is not documented by Google, but it explains why increasing the PMax budget can improve CPA in certain contexts: the algorithm accesses a wider set of auctions, some of which have a lower cost.

PMax Budget Distribution Across Channels: What You Can and Cannot See

The PMax budget distribution across different channels is one of the least transparent aspects of the campaign. Google does not show, in the standard interface, a breakdown of spend by channel. There is no report that says "40% of the budget went to Search, 25% to Display, 20% to YouTube" with the corresponding CPA per channel.

This opacity is intentional from a product design standpoint: PMax is built to optimise spend holistically across channels, without the advertiser interfering with the distribution. Google believes that showing the per-channel breakdown would lead advertisers to intervene manually on partial data, worsening overall performance.

From experience on accounts with PMax active for at least 6 months, distribution patterns emerge that repeat consistently across different verticals:

• Search and Shopping absorb the main share, in almost all e-commerce and lead gen accounts analysed, Search and Shopping collect between 55% and 75% of total PMax campaign spend. It is the channel with the lowest CPA and highest conversion rate, so the algorithm progressively concentrates the available budget there.
• Display and YouTube cover the tail, Display typically absorbs 15–25% of the budget, YouTube 5–15%. These channels have a higher CPA and lower conversion rate on a last-click basis, but contribute to awareness and retargeting in ways that the standard PMax report does not correctly attribute.
• Gmail, Maps and Discover are marginal in budget terms, in almost all accounts analysed, these channels absorb less than 5% of total PMax spend. They have significant impressions but very low CPCs.

The distribution is not static: it changes over time as the algorithm accumulates conversion data by channel. In the first few weeks, the distribution is more uniform (exploration phase). After 6–8 weeks, it tends to progressively concentrate on the channels with the best historical performance for that specific account.

Inference on channel distribution: the concentration of budget on Search and Shopping observed in most accounts is not random, it reflects the structure of conversions. PMax optimises for the conversion objective, and conversions tend to be more frequent and cheaper on Search than on Display. The practical result is that a PMax campaign with a limited budget behaves, in practice, very similarly to a Smart Shopping campaign, with an additional tail on Display and YouTube that grows as the available budget increases.

Over-Delivery and Under-Delivery: How Google Manages Daily Spend

The over-delivery and under-delivery mechanism is documented by Google in the About average daily budgets section (support.google.com/google-ads/answer/6385083). The principle is simple: the budget set is a daily average, not an absolute cap. Google can spend up to 200% of the daily budget on individual days (over-delivery) and less than the budget on other days (under-delivery), balancing the monthly total.

This mechanism exists to allow the algorithm to capture high-value traffic peaks without being rigidly constrained to the daily limit. It is designed to maximise conversion value over the long term, not to respect a point-in-time limit.

What the documentation does not sufficiently emphasise are the operational implications of this mechanism for those who monitor spend daily:

• Days with double the expected spend are not a problem, they are the algorithm's normal behaviour in response to high-value traffic peaks. Intervening by lowering the budget after an over-delivery day can interfere with the learning cycle.
• Persistent under-delivery indicates a bid target problem, if the campaign systematically spends less than 70–80% of the set budget for more than several consecutive weeks, the problem is almost always not the budget, it is the target ROAS or tCPA being too restrictive, limiting the auctions the algorithm can compete in.
• The correct monitoring window is monthly, not daily, looking at PMax daily spend produces false alarms. The correct metric to monitor is monthly spend relative to expected monthly budget (daily budget × days in the month).

Inference on over-delivery: on accounts where over-delivery recurs frequently, days with spend at 190–200% of the daily budget, those days almost always coincide with the best conversion performance of the month. The algorithm correctly identifies high-opportunity windows and concentrates spend there. Attempting to limit this behaviour by lowering the budget or adding daily spend caps interferes with one of the most effective mechanisms in PMax's automatic optimisation.

Target ROAS and Budget: the Interaction That Determines Real Spend

The target ROAS PMax budget relationship is the most critical, and most misunderstood, aspect of Performance Max budget management. Target ROAS is not just an efficiency objective: it is an implicit constraint on the auctions the algorithm can enter, and therefore on the volume of budget it can actually spend.

The mechanism works as follows: for each auction, the algorithm estimates the user's conversion probability and the expected conversion value. If the bid required to win the auction would imply a ROAS below the set target, the algorithm does not participate, or participates with a bid so low it does not win. The result is that an excessively high target ROAS leads to underspending the available budget: the campaign cannot find enough auctions that satisfy the ROAS constraint, and the budget remains partially unused.

According to the official Target ROAS documentation (support.google.com/google-ads/answer/6268637), Google recommends setting target ROAS based on account history, not on ideal objectives or industry benchmarks. A target ROAS too far from historical performance causes the algorithm to overly restrict the set of auctions it enters.

From experience on real accounts, the optimal relationship between budget and target ROAS follows a practical logic:

• If the campaign spends less than 80% of the budget for more than 2 weeks, the target ROAS is probably too high. The correct solution is to lower it by 10–15% and observe whether spend increases in subsequent weeks, before modifying the budget.
• If the campaign spends 100% of the budget and ROAS consistently exceeds the target, there is budget "left on the table": increasing the budget could capture additional conversions at equally efficient ROAS. The budget is the constraint, not the bid target.
• Target ROAS should not be modified by more than 15–20% at a time, large changes in either direction restart the algorithm's learning phase, causing instability in subsequent weeks. Changes to target ROAS should be made in gradual steps with at least 2 weeks between one intervention and the next.

Inference on the budget/target ROAS relationship: on accounts where the budget was increased while keeping the same target ROAS, in most cases a proportional increase in conversion volume is observed, with ROAS remaining stable or even improving slightly in the first few weeks. This happens because a higher budget allows the algorithm to access a wider set of auctions, some of which have very efficient CPA but require an absolute bid larger than a reduced budget would allow. This is one of the most important counter-intuitives in managing PMax campaign budget.

Practical Strategies to Optimise PMax Budget

Translating the understanding of PMax budget allocation mechanisms into concrete operational decisions requires a reference framework. Here are the strategies that produce the most stable results on accounts of different sizes, based on official sources and inferences from direct experience:

• Size the budget based on target CPA, not industry benchmarks: the minimum effective budget for PMax is calculated by multiplying the historical target CPA by the number of desired daily conversions. If the target CPA is €30 and you want 3 conversions/day, the minimum effective budget is €90/day. Below this level, the algorithm does not collect enough daily data to optimise bids reliably.
• Do not lower the budget during the learning phase: in the first 6 weeks from activation or after a significant change to the campaign, lowering the budget reduces the data available for optimisation and extends the instability period. If the budget must be reduced, do so gradually, by no more than 20% at a time, with at least 2 weeks between one reduction and the next.
• Use separate budgets for PMax campaigns with different objectives: if you have PMax campaigns with different objectives in the same account (e.g. one for e-commerce and one for lead gen), keeping them with separate budgets allows spend to be optimised independently. Campaigns with different objectives in the same account compete for budget in a non-transparent way if they share budget through a portfolio.
• Monitor the spend/budget ratio on a monthly, not daily, basis: set an alert when monthly spend falls below 75% of the expected monthly budget, this is the reliable signal of a problem (target ROAS too high, disapproved products in the feed, conversion tracking errors), not daily fluctuations.
• Increase the budget in steps of 20–30% every 2 weeks rather than doubling it all at once. Gradual increases allow the algorithm to progressively adapt its bid strategy, without restarting the learning period. Increases above 50% of the budget in a single change can temporarily destabilise performance for 1–2 weeks.

Inference on budget management over time: on mature accounts with PMax active for more than 6 months, a recurring pattern is observed: the campaign progressively learns to spend the budget more efficiently, same spend, growing ROAS, without the advertiser doing anything. This is the effect of the continuous accumulation of conversion data that improves the precision of the algorithm's estimates. Intervening frequently on the budget in this phase, lowering and raising it in response to weekly fluctuations, interrupts this automatic improvement cycle and produces on average lower performance than would be obtained with a less interventionist approach.