How to Scale Ad Budget Without Resetting the Algorithm

Not what folklore says. Google's documented triggers for the Learning status are bid strategy events, and a budget edit is not among them. That single correction changes how you scale, because most teams are protecting against the wrong thing.

Google's page on the duration of the learning period lists four categories that put a bid strategy into Learning. A new strategy, where the bid strategy was recently created or reactivated. A setting change, where a setting for the bid strategy was changed. A composition change, where campaigns, ad groups or keywords have been added or removed. And an ad group target change, which Google says may in some cases trigger a Learning status on Shopping campaigns.

Notice what is missing. There is no line saying that raising a daily budget flips the strategy into Learning. The cautions about budget are real and we quote them in full below — but they live on different pages, they are worded as risk rather than as a switch, and they carry conditions the folklore version has dropped.

The honest mechanical picture is this. Raising a budget does not tell the system to forget anything. It changes the volume of auctions the strategy has to fill at a given target — the system has to find more of the same kind of user, and the marginal user is usually more expensive to win. Google does not publish a model for how far a given budget increase perturbs delivery, and we are not going to invent one; there is no publishable number for "a X% budget rise costs you Y days". The damage from scaling badly shows up as cost, not as a status label. Our companion piece on why CPI rises as you scale covers that economics in depth. This article is the operational side: how to move the number without paying for calibration twice.

Across the 300+ apps we have managed since 2013, the most common scaling failure is not a reset at all. It is a team making four changes in one afternoon, seeing performance move, and having no way to attribute the move to any one of them.

Google states it can take up to 3 weeks or 1-2 conversion cycles for the bid strategy to calibrate to the new objective, although it can be faster depending on the amount of conversion data present. That is the only duration figure Google publishes, and the qualifiers are part of it.

The same page names three factors that determine where inside that range you land: the number of conversions that your campaigns, ad groups, keywords or products obtain; the duration of your conversion cycles; and the bid strategy itself, with Maximize Conversions and Maximize Conversion Value given as the examples.

The second unit is the one worth internalising. Google offers conversion cycles as an alternative to weeks, which means the clock is not really a clock — it is a data counter wearing a calendar's clothes. A campaign generating a high volume of installs a day calibrates on a completely different timeline from one generating a handful, and no amount of waiting substitutes for volume.

Google's tips for maximising your App campaign add the front end of that timeline, noting that App campaigns can take a few days to start gathering information. Between the two pages you get a shape rather than a schedule: a few days before there is anything to look at, and potentially up to three weeks before a strategy has settled against a new objective.

What neither page provides is a mandatory waiting interval between changes. Teams routinely quote one to each other — a fixed number of days you must leave a campaign alone — and we cannot source it. That figure is not publishable, so we do not print it. What is sourced is the conversion-count condition in the next section, and it is a better gate than any calendar rule because it scales with your campaign instead of ignoring it.

Because Google states that making changes to your in-flight campaign before the first 100 conversions have registered may disrupt learning and result in poorer performance. That sentence, from Google's best practices for App campaigns, is the closest thing to a hard gate in the documentation.

Every word of the condition travels with the number. It is the first 100 conversions, not a rolling window. It applies to an in-flight campaign, meaning one already running rather than one being set up. And Google says such changes may disrupt learning — it is a stated risk, not a guarantee of damage. Quoting it as "you cannot touch a campaign until 100 conversions" overstates it; quoting it as a vague suggestion understates it. It is a threshold below which you should expect changes to cost you.

The gate is more useful than a calendar rule because 100 conversions is a count, and counts scale with spend. Two campaigns on identical daily budgets can sit on opposite sides of it at the same moment: at a target of ₹40 per install a given budget clears 100 installs quickly, while at ₹400 the same budget takes an order of magnitude longer. Those are arithmetic examples on figures we chose, not benchmarks — we do not publish a CPI benchmark we cannot source.

It is also why launching several small campaigns instead of one larger one so often underperforms. Splitting a fixed budget across four campaigns divides the conversion signal four ways, and each fragment needs its own hundred conversions before edits are safe. You have not diversified; you have quadrupled the calibration bill. Our guide to building a mass acquisition programme covers when the consolidation argument stops applying.

Google's stated guideline is to avoid drastic changes, giving as its examples changing the budget by more than 20% or changing your CPI by more than 20%. That is a guideline about magnitude, and the only percentage Google puts a number on.

The exact wording on the App campaign tips page is to try not to make drastic changes to your campaign, for example changing the budget by more than 20%, or changing your CPI by more than 20%. Two things follow from how that is phrased. Budget and bid are each named separately, which means a 20% budget rise combined with a 20% bid cut is not one change within the guideline — it is two changes that happen to be individually within it. And the guideline is expressed as a percentage, so it compounds: a campaign that takes a 20% step every time it clears the gate doubles its budget in four steps.

The fourth step is the one teams break under pressure. Performance dips after a rise, someone reverses it, someone raises it again, and the campaign has absorbed three changes in a week. A reversal is itself a change rather than an undo — there is no undo button on a bid strategy.

Google publishes minimum budget-to-bid ratios for App campaigns, and they differ by campaign type — so a bid that is fine at one budget is starved at another. This is the check most teams skip, and it explains a large share of campaigns that quietly fail to spend.

The best practices page states them directly: for ACi tCPI and ACpre, the budget should be at least 50 times the bid; for ACi tCPA, the budget should be at least 10 times the bid; for ACe tCPA, the budget should be at least 15 times the bid. Google uses its own campaign-type shorthand on that page and does not expand the abbreviations there, so we quote them as written rather than guess at expansions.

The arithmetic in those cards is the ratio applied to a bid we picked for illustration, not a recommended bid. The point is the relationship: the ratio is a floor on budget given a bid, which means raising a bid without raising the budget can push a campaign under its own floor.

This also reframes what a budget increase is for. If a campaign sits at exactly its ratio floor, the first tranche of any budget rise is not buying scale at all — it is buying the headroom the bid already needed. If your campaign is not spending its budget in the first place, the ratio is one of the first things to check, and our diagnostic on an App campaign that will not spend works through the rest of the fault tree.

Google's own description of App campaigns is that you provide some text, a starting bid and budget, and the languages and locations for your ads, and the system tests different asset combinations using your text ideas, images, videos and assets from your store listing. Read that as assets rather than finished ads. The system tests different ad combinations and serves the best-performing ads. Bid and budget are two of the very few inputs you control. Setting them out of proportion to each other wastes one of them.

It is an average, not a cap — Google states your campaign spend will never exceed 2 times your average daily budget on a given day, and that in a given billing period you are never charged more than 30.4 multiplied by your average daily budget amount. Both halves of that matter when you are scaling.

Google's page on choosing your bid and budget sets out the mechanism. The daily figure is the average you are comfortable spending per day; individual days can run above it because traffic fluctuates, bounded at twice the average; and the billing period is bounded by the 30.4 multiplier, Google's figure for the average number of days in a month.

Two operational consequences follow directly.

• A single overspending day is not a signal. It is the documented behaviour of an average daily budget, bounded at 2x. Reacting to it by cutting the budget introduces a change the campaign did not need, and that change is subject to every caution in this article.
• Judge spend over the billing period, not the day. The 30.4 multiplier is the number that governs what you can actually be charged, so a period-level view is the one that matches how the product bills.

A finance team working from the daily number will be surprised by a 2x day; a team working from the 30.4 multiplier will not. That alignment prevents the worst kind of intervention — a mid-flight cut driven by an accounts query rather than by performance. We work through the cash side in our guide to UA budget and cash payback, and the return side in the ROAS and CAC guide.

Scale the existing campaign wherever it is viable, because creating a new one starts a new bid strategy — and a bid strategy that was recently created or reactivated is the first documented trigger for the Learning status. Duplication is the one move that reliably does the thing everyone is trying to avoid.

The instinct runs the other way. When a campaign performs well, teams want to clone it and run the clone at a higher budget, on the theory that the original stays safe. But the clone has no calibration at all: it is a new bid strategy, it starts in Learning, and it has to accumulate its own conversions before edits to it are outside the caution zone. You have not protected the original; you have bought a second calibration.

The same page lists composition changes — campaigns, ad groups or keywords added or removed — as a trigger in their own right, so restructuring is not a neutral tidy-up. Settle on a structure before you scale rather than during.

There are legitimate reasons to open a new campaign: a genuinely different objective, a different market, a different campaign type where the bid semantics differ. Those are cases where the existing strategy is calibrated to the wrong thing and the recalibration is the point. What does not justify a new campaign is a desire to spend more money on something that already works. That is what the budget field is for.

When the existing campaign genuinely cannot absorb more spend at an acceptable cost, the answer is usually upstream of bidding altogether. New creative expands the set of auctions you can win without asking the same strategy to reach further into the same audience, and our creative strategy guide covers how to feed that pipeline. In our experience that lever is under-used precisely because it is slower to pull than a budget slider.

The shape is similar and the published numbers are not — TikTok, for instance, states that volatility typically starts to decline after about 25 campaign results or 7 days from when the campaign enters the learning phase. A single scaling rule applied across networks will be wrong on at least one of them.

Note what that TikTok sentence actually claims. It describes volatility starting to decline after about 25 results or 7 days — not a hard exit from the phase, and not a threshold below which changes are forbidden. It is a different kind of statement from Google's 100-conversion caution, and the two should not be merged into a shared house rule. TikTok's page also names the behaviours that interrupt progress: pausing campaigns or ad groups, making edits that retrigger the learning phase, and setting unreasonable budget allocations or creative volumes.

Meta's learning phase documentation is the obvious third comparison, and we are deliberately not quoting a figure for it. The Business Help page did not return readable content when we checked it, and our rule is that we do not print a number from a page we could not open and read. If you have seen a specific Meta optimisation-event threshold quoted in a deck, treat it as unverified until you have the page in front of you. Our overview of Meta app install campaigns and the comparison in TikTok app install campaigns cover the platform mechanics we can source.

One variable per change, steps within Google's 20% example, spaced by the documented conversion gate rather than by the calendar, with a written log of every edit. That is the whole method, and its value is that every component is traceable to a published sentence.

Two limits are worth restating so nobody over-reads this article. Google does not publish a required interval between changes, so any waiting period in your runbook is a convention you chose, not a rule you inherited — label it as such. And Google does not publish how a budget change propagates through delivery, so the honest framing of a step is that it increases the volume the strategy must fill at your target, with a cost you observe rather than predict.

Scaling is not the same problem as efficiency. This article is about not damaging what you have while you spend more; the reason the cost per install rises anyway as you spend more is a separate mechanism, covered in our piece on diminishing returns. If you are planning a step change in spend and want the cadence pressure-tested against your actual conversion volumes, send us the campaign structure, or see how we run this in our user acquisition work.