The 50-conversions rule that quietly decides your PPC results

Every Google Ads and Meta account has a number that decides whether the money works, and almost nobody talks about it. It isn’t your budget. It isn’t your CPC. It’s roughly 50 conversions per campaign, every 30 days — the volume smart bidding needs before it can actually learn who to bid on.

Below that line, the algorithm is guessing. Above it, it’s optimising. Most of the accounts I audit are sitting just under it and wondering why the results feel random.

What “Learning (limited)” is really telling you

Put a campaign on Target CPA or Target ROAS and you’ll sometimes see a status label: Learning (limited). Google phrases it politely. What it means is blunter: this campaign doesn’t give me enough conversions to optimise, so I’m running on thin data.

Smart bidding works by pattern-matching. Every time someone converts, the system gets one more example of what a buyer looks like: the device, the hour, the query, a hundred signals you never see. Fifty a month is roughly where the patterns become statistically real. Give it five, and every conversion swings the model wildly. That’s not learning; it’s reacting to noise.

So when you raise the budget on a campaign converting six times a month, you don’t get six times better bidding. You just get more impressions bought by a system that still has no idea who’s worth it. A bigger budget without more conversions just makes the guessing more expensive.

What actually counts as a conversion

Here’s where a lot of accounts quietly sabotage themselves. The 50 isn’t 50 of anything. It’s 50 of the one action you told the platform to optimise for.

If your primary conversion is a purchase and you get 40 sales a month, you’re under the line, even if you also log 300 “add to cart” events and 900 “page view” ones. Those secondary actions don’t feed the bidder you set to chase purchases. I regularly see accounts with thousands of conversions in the dashboard, nearly all of them the wrong ones, and the purchase bidder starving on 30 real sales.

So before anything else, count how many of those you actually get in a month. That’s the number smart bidding lives on.

Why untracked conversions starve the bidder

Now the cruel part. The bidder only counts conversions it can see. If a chunk of your sales never make it back into Google — lost to consent denials, ad blockers, a checkout on another domain, a click ID stripped in a redirect — then a business doing 60 real sales a month might be feeding the algorithm 40.

You were over the line in reality but under it in the data, and the campaign runs limited because the receipts never arrived.

This is why I won’t touch a paid account without checking measurement first. Fixing server-side tracking often recovers 20–40% of conversions that were dying in the browser, and that alone can lift a campaign out of Learning (limited) without spending a cent more. The conversions were always happening; Google just couldn’t count them.

The fix nobody wants to hear: consolidate

The instinct when results go soft is to split things up: a campaign for this product line, one for that region, one for the new offer. It feels organised. It’s also the fastest way to guarantee every campaign stays under 50.

If you’ve got 60 conversions a month spread across five campaigns, you don’t have one healthy account. You have five starving bidders. Pull them into one or two and the same 60 conversions land in the same place, and the machine finally has enough to learn from.

Fewer, fatter campaigns beat many thin ones almost every time. It’s the single change that turns the most accounts around.

Why “let’s just test it for a month” fails

A client says: give it a month, see if Google Ads works for us. Reasonable on the surface. Then do the math on the learning curve.

Smart bidding wants around 50 conversions to exit the initial learning phase, and that phase re-triggers for a week or two after every meaningful change. If you’re getting 15 conversions a month, you will never finish learning inside a 30-day window. You’ll spend the whole month funding the guessing phase, judge the campaign on its worst data, and conclude “paid doesn’t work for us.” It might. You just never let it out of the tutorial.

A fair test is measured in conversion volume, not calendar days. Two hundred conversions tell you something real. Thirty tell you nothing except that you didn’t reach thirty-one.

The honest math

This is the part where a lot of agencies go quiet, so I’ll say it plainly. If your economics can’t realistically produce ~50 conversions a month for a campaign, no amount of “optimisation” will save it. Not mine, not anyone’s. Smart bidding has nothing to chew on.

Usually there’s a route to the threshold:

• Consolidate campaigns so the conversions pool instead of scattering.
• Fix the tracking so real sales stop vanishing before Google counts them.
• Widen a conversion that’s been defined too tightly to fire often.
• Raise the budget — but only because the volume is genuinely there to buy.

Sometimes none of those apply, and the honest answer is that paid search is the wrong channel for you this quarter. I’d rather tell you that than bill you €800 a month to babysit a campaign that mathematically cannot optimise.

That’s also why I don’t take on paid media under about €1,000/month in spend — below it there’s rarely enough conversion volume for management to earn its keep. Our PPC management runs from €500 to €3,500 a month depending on scope, and the first thing it buys is this diagnosis: can this account realistically clear 50 a month, and if not, what has to change first? It’s the same discipline behind the FMCG and energy accounts we’ve run: the volume was there, so the job was keeping the bidder fed with clean data.

Get a campaign over that line and everything downstream gets easier. Bids stop lurching, cost per sale settles, and the budget you already spend works harder because the algorithm knows what it’s aiming at.

If your campaigns feel like they’re guessing, they probably are. Start with the PPC service, or if you suspect the real leak is measurement, see how pricing works and where we’d look first.