You can hit your Meta target dead on and still lose money. I watched an account doing around $20k a month buy purchases at $29 against a $30 goal: green checkmarks everywhere, while the owner kept asking why the cash left over after product and shipping had not moved.
The number you type into Ads Manager is not a report. It is an instruction about which people to chase. A cost per result goal is an average target, not a ceiling, so some results can cost more than your number. The ROAS goal is its value counterpart. One steers delivery toward affordable events; the other steers it toward purchase value while defending a return floor. Choose the wrong instruction and the target can look healthy while the business does not.
Free tool: the ROAS Calculator works out your return on ad spend, your break-even ROAS and the budget a revenue goal takes.
I will use one store throughout: it sells a $28 cleanser and a $140 set, and both fire the same Purchase event. The question is not whether either customer is good. It is what Meta can tell about the difference from the goal you give it.
First, the goal changes what counts as a good auction
Every impression on Meta runs through an auction. Meta weighs a bid, its estimate that a person will take the chosen action, and ad quality in what it calls Total Value in this breakdown of the auction. Your goal influences what the system is willing to bid for an opportunity. It does not force a purchase at the price you typed in.
The main delivery choices fall into three useful buckets:
- Highest volume uses the budget to seek the most events and lets cost float.
- Cost per result goal, often called cost cap, aims for an average cost near your number and can underspend to protect that average.
- ROAS goal aims for a minimum return on ad spend and can also underspend rather than pursue sales it expects to fall below the floor.
Bid cap is different. It limits the per-auction bid, not your reported CPA, which is why I almost never start there with a small store. For the cleanser store, the useful first decision is simpler: should Meta treat every Purchase as one equal win, or should it care how much each Purchase is worth? Answer that before you choose a number.
The second prediction is what separates CPA from ROAS
With a cost goal, the crucial prediction is whether a person will complete your event. With a value goal, Meta must also predict how much that event is likely to be worth. That is why a ROAS goal depends on purchase value and currency reaching the system, and why value bidding has little to distinguish when products all cost the same.
On our store, a cleanser purchase and a set purchase each count as one Purchase. To a cost goal, they are equal results. A value goal can distinguish the $28 order from the $140 order. It may be willing to bid differently for people likely to buy the set, provided its prediction supports that bid. It is not reading the shopper’s mind or guaranteeing a bigger cart. It is using the purchase-value signal the store sends it.
Goals also affect pacing. Meta can leave budget unspent when it cannot find enough opportunities it expects to meet a tight target. A delivery reference describes both cost-cap underspend and under-delivery from a ROAS goal set too high. I have seen a $25 cap against a $34 uncapped average spend only $40 of a $300 budget. The ad set was active. The number was simply too restrictive for the opportunities available. Underspend is not the same as cheaper results. Before celebrating a strict goal, check whether it bought enough results to matter.
A cost goal buys events, even when their values differ
A cost goal tells Meta to pursue your selected event at an average cost near your target. It does not ask whether one Purchase leaves more money after fulfillment than another. As one comparison of the bidding modes puts it, CPA treats a $20 order and a $200 order as the same win.
For our store, the $28 cleanser buyer may be easier to convert than the $140 set buyer. If so, a cost goal can fill the account with small carts while keeping CPA near $30. The result is not an algorithm malfunction. Meta followed the instruction: find Purchases at roughly that cost. Use a cost goal when each conversion is worth roughly the same to you, not just when each conversion has the same event name.
The same distinction matters outside ecommerce. On lead forms, cheap-event buying can find people willing to submit a form without finding people likely to become qualified calls. On purchases, it can favor lower-friction single-item orders. I used to tell clients that a low CPA meant an account was healthy. I was wrong. A low CPA tells you what the events cost. It does not tell you what remains after the order is fulfilled.
Nor can you solve a weak CPA result by typing in a heroic number. Practitioner guidance on cost goals calls for daily budget around 5x the goal to reach 50 conversions in a week, roughly 50 to 100 weekly conversions for a cost goal to settle, at least three weeks before scaling, and an initial goal 10 to 20 percent above the recent average. Those are planning benchmarks, not a promise that delivery will cooperate. The principle is less fussy: give the system room and enough events to learn before tightening its leash.
A ROAS goal buys value, then protects its floor
A ROAS goal makes purchase value part of the decision. Instead of counting the $28 cleanser and $140 set as identical wins, it can pursue buyers likely to place larger orders. That requires value and currency on Purchase events: value optimization uses the Maximize value performance goal and needs those signals. It also works best when baskets vary. If everything in your catalog costs $29, there is little extra value for the system to learn to prefer.
Setup deserves attention here. The value-optimization discussion also calls for Pixel and Conversions API signals and enough purchase volume before imposing a ROAS goal. I would test highest value without a floor first. That gives you an uncapped return to compare with the target you want to set, rather than asking the account to clear a number chosen by hope and a spreadsheet.

A ROAS floor introduces a tradeoff: Meta may reach fewer people and spend less to defend it. Highest value seeks the most purchase value with the budget, while a ROAS goal can hold back spend to stay near its weekly target. The same discussion describes fewer purchases with higher combined value under value optimization. I look for around 50 or more purchases a week before expecting a goal to settle, and 100 plus if the value signal is predicted lifetime or profit value. Those numbers guide whether I would try the test; they do not make the outcome automatic. Judge it on return and contribution, not order count alone.
Same spend, different buyers: read past the CPA
Put $5,000 behind each approach and the reported winner depends on which column you read. A side-by-side comparison shows a conversion-focused side with 170 orders, a $29.41 CPA, a $72 average order value and 2.45x ROAS. The value-focused side has 120 orders, a $41.67 CPA, a $118 average order value and 2.83x ROAS.
Read that slowly. The conversion side wins on order count and cost per order. The value side wins on return because its average basket is 64 percent larger. A higher CPA can accompany more revenue from the same spend. Neither CPA nor ROAS alone tells you profit; you still need to know the variable costs behind those baskets. But the example shows why declaring victory from the cheapest Purchase can be expensive.
This is also where the customer mix can shift. Cheap-event buying can lean toward easy converters, including past buyers and low-friction single items. Value buying can narrow delivery toward people the system expects to place larger orders. For our store, the practical question is whether the margin lives in the set rather than the cleanser. If it does, the campaign with fewer orders and a higher CPA may be doing the more useful job.

Calculate the floor from what an order leaves behind
Before setting either goal, work out what an order is worth after variable costs. Start with average order value. Subtract product cost, shipping, payment fees, and pick and pack. Divide what remains by revenue to get contribution margin. Break-even ROAS is 1 divided by that margin: in the worked example, a $100 sale with $60 in product, shipping and fees leaves a 40 percent margin, so break-even ROAS is 2.5.
Now give our store a $70 average basket at 40 percent margin. It has $28 left before advertising. That makes $28 its break-even CPA on that average basket. I would use a $22 cost goal and leave $6 of buffer, because averages move. The goal is not a guarantee of $22 purchases; it is an instruction that gives the account room to spend below the point where the average order stops covering its variable costs and ads.
ROAS expresses the same break-even point from the other side. At 40 percent margin, the store needs $2.50 in revenue for every $1 of ads to cover those costs. The formula counts variable costs, not rent or salaries. A 2.5 floor sits at that break-even point if delivery actually meets it and the margin holds. A 3.0 floor leaves more room above variable costs, but it can also restrict spend. A stricter number cannot create profitable auctions that were not there.
Here is the decision in plain English: use a cost goal when baskets cluster near one value. Consider a ROAS goal when the $28 cleanser and $140 set share a Purchase event but have very different paydays. In either case, calculate the economics before asking Meta to enforce them.
When Purchase is too thin to support the goal
Sometimes neither Purchase goal has enough signal to settle. If every basket is $29, value bidding has little to prefer; a cost goal or highest volume is the simpler starting point. If the account gets eight purchases a week, a tighter goal does not manufacture data. As this discussion of Meta’s auction and learning phase explains, a new or edited ad set needs around 50 results in the week after the last edit to exit learning.
Change the event before you keep tightening the number. For ecommerce with too few purchases, I might test an earlier, higher-volume event such as Initiate Checkout with value. For a subscription business, subscription start may better represent the outcome than a first order. For lead generation, a qualified call can be more useful than a form submit when there is enough volume to optimize for it. Those are different fixes for different signal problems, not interchangeable settings. The test is whether the event gives Meta enough information about the outcome the business actually wants.
Four questions for whoever runs the account
I would ask these before accepting a green CPA cell as an answer:
- What event are we optimizing for, and what value do we pass with it? A Purchase count cannot tell the cleanser from the set unless the value signal arrives.
- What are break-even CPA and ROAS on variable margin? Show where the chosen goal sits relative to both.
- What did uncapped delivery do before we added a floor? A baseline is better than guessing what the account can clear.
- How many optimization events arrived in the last seven days, and did we edit anything midweek? A thin or interrupted learning window changes how much faith to put in the result.
If a manager cannot answer those without opening three dashboards, that is the tell. I have seen the same participation tradeoff on Google, where a tight target can shrink the auctions an account enters, as discussed here for Google Ads. On Meta, the practical check is the same: do not praise a floor without checking what delivery it gave up to meet it.
Now say the store adds a $180 bundle beside its $28 cleanser. Run the model before touching Ads Manager. A $30 cost goal still counts each Purchase as one event. If cleanser buyers remain easier to convert, delivery can keep favoring them; CPA may look steady even though the new bundle has not improved the business. A 2.8 ROAS floor can distinguish the order values and favor people likely to buy the bundle, but it may deliver fewer orders at a higher reported CPA. Check contribution and volume before calling that a win.
If the bundle carries 50 percent margin after product and shipping, its break-even ROAS is 2.0; a 2.5 return leaves room above those variable costs. If the account gets only 20 purchases a week, neither tight Purchase goal has the roughly 50 weekly results used as a learning benchmark. I would stay on highest volume while the bundle builds history rather than force a floor the account cannot reliably judge.
Pick the buyer you need, then give Meta the number that helps it find that buyer. Cost goal when events pay you about the same. ROAS goal when one event hides two very different paydays. That is the work I want from whoever runs the account, and why I moved this part to continuous management at groas: check the goal against margin, pass purchase value cleanly through Conversions API, and move the floor only after uncapped delivery shows what the account can actually clear.

