It is the question we get more than any other, in some form: what should I be spending? People expect a percentage of revenue, or an industry benchmark, or a number that sounds responsible. The useful answer is none of those. It comes from the machine you are buying from, not from your budget.
Both major ad platforms optimize by learning from conversions. Not from clicks, not from impressions — from completed actions. Everything about how much you should spend follows from one fact: an optimizer with too few examples cannot optimize.
So the rule we give clients is deliberately blunt.
The rule
Your daily budget floor, per platform, is one target cost per conversion.
If a lead is worth $150 to you, the floor is $150 a day on that platform. Not $40. Not “let’s test with $500 a month.”
Why the floor is a floor
The instinct with a small budget is to run a small version of the campaign. Spend a third, get a third of the results, scale when it proves out. That is how nearly every other business expense behaves, and it is precisely wrong here.
Paid platforms do not deliver linearly at the bottom. Below a certain spend, the algorithm never accumulates enough conversion data to learn who to show your ad to. It keeps guessing. You do not get a smaller version of the result — you get an expensive random sample, and then you conclude the channel does not work for you.
One conversion a day is the point at which the system has something to learn from every single day. It is the minimum at which the campaign is a campaign rather than a donation.
The learning phase, plainly
Every new campaign or ad set starts in a learning state. During it, delivery is deliberately exploratory and performance is unstable — costs swing, results are erratic, and nothing you see is representative. The platform is running experiments on your money because it has to.
The exit conditions are published, and they are specific:
| Platform | To leave learning | Scope |
|---|---|---|
| Meta | ~50 optimization events within 7 days | Per ad set |
| Google (Target CPA) | ~30 conversions in 30 days recommended | Per campaign |
| Google (Target ROAS) | ~50 conversions in 30 days recommended | Per campaign |
Read those numbers against the floor and the tension is obvious. One conversion a day is seven a week. Meta wants fifty. The floor does not get you out of the learning phase. It gets you into the game.
The floor is not what good costs. It is the line below which nothing can work.
The part people skip
We keep both numbers in front of clients on purpose, because quoting only the floor sets up a disappointment and quoting only the stable figure ends the conversation. The honest version has two lines: here is the minimum at which this is worth attempting, and here is what it costs to be past the guessing.
Roughly, the stable number is seven times the floor — seven conversions a day gets you to about fifty a week. That is a real gap, and pretending otherwise is how agencies end up defending a campaign that was never funded to succeed.
What that looks like in practice
Target CPAs below are illustrative, chosen to show the arithmetic across a realistic spread. Yours comes from your own numbers, not from this table.
| Business | Target CPA | Daily floor | Daily for stability |
|---|---|---|---|
| Home services — drain repair Emergency intent, short consideration | $85 | $85 | $595 |
| HVAC install lead High ticket, longer decision | $220 | $220 | $1,540 |
| Dental new patient Local, high lifetime value | $130 | $130 | $910 |
| B2B demo request Small audience, expensive click | $400 | $400 | $2,800 |
The B2B row is where this gets uncomfortable, and it is the honest case for not running the channel. If $2,800 a day is not real, the answer is not to run $300 a day and hope. It is to pick a channel whose economics you can actually fund.
Per platform, not in total
The rule is per platform because the learning is per platform. Google does not benefit from what Meta learned. Splitting one floor across two channels does not give you two half-funded campaigns; it gives you two campaigns that both stay stuck.
This is the single most common way we see budget wasted, and it is always well intentioned. A business with $3,000 a month spreads it across Google, Meta, and a third platform to hedge. Every one of them sits below its floor. Twelve months later the conclusion is that paid does not work for their industry, when the actual finding is that it was never funded past the exploratory phase on any channel.
One channel, properly funded, beats three that are starved. If the budget only clears the floor once, then you run one platform.
Which channel goes first
Start where intent already exists. Somebody typing “emergency drain repair” into Google has already decided; you are competing to be chosen, which is a shorter argument to win.
Social is where you build the demand that search harvests.
That ordering is why we treat paid search as the entry point for most service businesses, and social as the layer that makes the search cheaper over time. Someone who already recognizes your crew converts at a different rate than someone meeting you in a results page for the first time.
It is also where this connects to the other thing we published this month. Our audit of five years of paid social found that the content which actually performs is content about people — and that the reason most businesses do not run it is not taste, it is that nobody ever captures it.
The evidence behind that
Faces beat facts: what $800,000 of paid social actually showed.
Basis & limits
- This is a framework, not an audit. It is our recommended approach, reasoned from published platform mechanics and from running these accounts. It is not a measured finding, and we have not tested the floor as a controlled experiment.
- Platform thresholds are the platforms’ own guidance. Meta publishes the ~50 events per ad set per week figure; Google publishes conversion-volume recommendations for Smart Bidding. Both are guidance rather than hard switches, both change without notice, and both should be checked against current documentation before being quoted to a client.
- The example CPAs are illustrative. They are chosen to show the arithmetic across a realistic spread, not drawn from a specific account. Do not cite them as benchmarks.
- The 7× stability figure is arithmetic, not a result. It is what fifty weekly events implies at one conversion per unit of target CPA. Real accounts reach stability at different points depending on conversion window, audience size, and how many ad sets the budget is divided across.
- It assumes you know your target CPA. If that number is a guess, fix that first — every figure here is a multiple of it, so an inherited or invented CPA makes the whole calculation confident and wrong.