TL;DR
Value optimisation tells Meta to bid more for people likely to spend more, instead of bidding the same for every purchase. It runs inside Advantage+ Sales campaigns, and practitioner guides suggest it starts to stabilise with roughly 30 to 50 high-value purchase events a week, while Meta's own learning-phase guidance stays at about 50 optimisation events a week. The catch is that it only works if your purchase values are accurate, complete, and sent quickly. Feed it sloppy values and it will confidently optimise toward your most expensive mistakes. Get the data right first, then switch it on, then judge it on revenue and margin, not on cost per purchase.
What value optimisation actually does
Normal purchase optimisation treats every conversion as equal. A 20 euro order and a 400 euro order are the same event, so the system chases the cheapest conversions it can find. That is fine when your orders are all roughly the same size. It is a slow bleed when they are not, because the cheapest buyers are often the smallest buyers, and the system will happily fill your order book with them.
Value optimisation changes the target. Instead of minimising cost per purchase, it maximises predicted value. Meta builds a predicted lifetime value model, pLTV, from your purchase history and bids higher for people who look like big spenders. In practice that means fewer, more expensive conversions that are worth more. Or at least that is the deal when it works.
The important part is the "when it works". Value optimisation is not a switch you flip on a broken account. It is an amplifier. It amplifies good signal and bad signal with equal enthusiasm.
The minimums, honestly stated
Meta has not published a hard minimum for value optimisation. Practitioner guides report that it can stabilise with roughly 30 to 50 high-value purchase events a week, and Meta's standard learning-phase guidance is about 50 optimisation events a week. Treat both numbers as what they are: reported experience, not a promise.
What matters more than the count is the shape of your data. Value optimisation needs spread. If 90 percent of your orders sit between 45 and 55 euros, there is nothing for the model to learn, and you will pay for complexity you do not need. If your orders range from 25 to 500 euros, and the big ones share traits the model can find, then it has something to work with.
It also needs truth. Three things break it before it starts:
Missing values. If some purchases arrive without a value, or with a placeholder value, the model learns that some customers are worth zero. Fix your event setup before you touch bidding.
Slow values. If your real order value only settles days later, after refunds, upsells, or subscription upgrades, the model is optimising on early numbers that do not match reality. Send the best value you have at purchase time, and be honest about what changes later.
Dirty values. Shipping, tax, discounts, and refunds all change what an order is worth. Decide what "value" means for your business, apply it the same way every time, and send that. A model trained on gross revenue will chase gross revenue, which is not the same as profit.
How to switch it on without regretting it
Do not flip your main campaign to value optimisation on a Friday. Run it as a test, side by side, with the same creative and the same audience, and give it time. Value models learn slower than purchase models, because they need enough big orders to separate signal from noise. Two weeks is a reasonable minimum before you judge it. A weekend is not a test.
Judge it on the right numbers. Cost per purchase will probably rise. That is the point. You are paying more per order to get better orders. The numbers that matter are revenue per purchase, total revenue, and, if you can track it, margin. If revenue per purchase rises faster than cost per purchase, you are winning. If cost per purchase rises and revenue per purchase does not move, switch it off and fix your data first.
One more warning. Value optimisation loves your existing best customers, which means it can quietly narrow your prospecting toward people who look like the people who already buy big. Watch your new-customer share while it runs. If it collapses, you are harvesting your own audience instead of growing it, and that trade has a short shelf life.
When not to use it
Skip value optimisation if your order values barely vary, if you get fewer than a few dozen purchases a week, or if your tracking is not clean yet. In all three cases, standard purchase optimisation with a cost control will do the same job with less risk. The fanciest bidding strategy in Ads Manager cannot fix a leaky bucket. It just charges you more per litre.
FAQ
What is Meta value optimisation?
A bidding option inside Advantage+ Sales campaigns that tells Meta to maximise predicted purchase value instead of minimising cost per purchase. Meta bids more for people its pLTV model predicts will spend more.
How many conversions does value optimisation need?
Meta has not published a figure. Practitioner guides report roughly 30 to 50 high-value purchase events a week as a workable range, and Meta's standard learning-phase guidance is about 50 optimisation events a week. Volume matters, but clean and varied value data matters more.
Will my cost per purchase go up?
Probably. Value optimisation trades cheaper conversions for more valuable ones, so cost per purchase rising is expected. Judge it on revenue per purchase, total revenue, and margin, not on cost per purchase alone.
Does value optimisation work for lead generation?
Not in the same way. It is built for purchase value. For leads, the equivalent is sending back real lead quality or closed-won value through offline conversions or the Conversions API, so the system can learn which leads are worth money.
What breaks value optimisation most often?
Bad value data: missing values, placeholder values, slow values that arrive after the fact, and inconsistent definitions of what counts as value. Fix the data before you switch the bidding on.
How long should I test it?
Give it at least two weeks with the same creative and audience as your control. Value models learn slower than purchase models, and a few days of data will mostly show noise.
MetaMaxd is a Meta ads agency. We would rather run boring purchase optimisation on clean data than fancy value bidding on a mess. If your tracking needs sorting before your bidding does, that is where we start.


