Meta Ads Performance Goals: Match Optimization to Evidence
Which optimisation event should I choose on Meta?
Choose the deepest event, usually Purchase for a store, that still reaches about 50 a week per ad set, the figure Meta's learning phase needs. If it can't, Meta's own fix is a more frequent event, such as add to basket. The trade-off: a shallower event is cheaper to optimise but sits further from revenue.
For the surrounding account decisions, compare Meta Ads Performance Dashboards: A Decision Workflow and use Meta Ads Reporting Stacks: A Requirements-Led Selection as the next diagnostic.
In short
Pick the deepest event that gets enough volume. For a store that is usually Purchase. Meta’s learning phase needs about 50 optimisation events a week per ad set, and if an ad set can’t reach that, Meta’s own advice is to move to a more frequent event such as add to basket. Shallower goals also need less budget: Meta suggests a daily budget of at least ten times the goal’s average cost.
This choice comes after the campaign objective; for that first step, see Which Facebook campaign objective should an online store use?
Optimization-event selection matrix
| Business outcome | Candidate optimization event | Required signal evidence | Risk if signal is thin |
|---|---|---|---|
| Recognized purchases | Valid Purchase tied to the declared recognized order state (performance goal: Maximise number of conversions) |
Reader-set minimum valid purchases per decision window, with Meta’s about-50-a-week reference as the floor; valid purchases ÷ eligible orders documented | Delivery can be unstable or interpreted from too few outcomes |
| Purchase value, not order count | Purchase with a value (performance goal: Maximise value of conversions) |
Values sent through the Conversions API and Meta pixel; Meta suggests 50 or more conversions a week | Meta says value optimisation can bring fewer conversions and a higher cost per result |
| Qualified leads | Server-verified qualified-lead transition, not raw form submit (performance goal: Maximise number of qualified leads) | Reader-set count and qualification coverage = qualified events ÷ eligible submitted leads | Optimizing raw or mislabeled leads can favor low-value volume |
| Checkout creation | Accepted InitiateCheckout from checkout service |
Reader-set event count plus downstream purchase rate = recognized purchasers ÷ eligible checkout starters | Proxy may improve while purchase economics deteriorate |
| Landing-page view | Valid page view after outbound click (performance goal: Maximise number of landing page views) | Landing-view coverage = valid landing views ÷ outbound clicks | Optimizes a distant proxy and says little about recognized revenue |
Choose the deepest event aligned with the business outcome that also meets your valid-volume and integrity thresholds. Meta’s reference point for volume is about 50 optimisation events in the week after the last significant edit; below that, the ad set is likely to be marked Learning limited. Use it as the floor and put your own integrity checks on top; if the desired event is thin, document the proxy risk and an exit condition.
The performance goal names come from About performance goals; the value row from About maximising the value of conversions.
Interpretation boundary
Use the optimization-event depth matrix only for its stated decision. Select the deepest valid event aligned to the business outcome that meets the reader’s reconciled volume and integrity thresholds. A thin or mislabeled event requires a documented proxy risk and exit condition, measured against Meta’s own volume figure rather than one you invent. Reader-supplied thresholds remain inputs, not universal standards.
Documented thresholds
Meta publishes the rules below. Each links to the Meta Business Help Centre page it comes from, checked on 1 Oct 2026; Meta revises these pages without notice, so re-check anything load-bearing before you act on it.
- The volume floor. Ad sets usually leave the learning phase after about 50 results in the week after their last significant edit (About the learning phase); an ad set unlikely to reach about 50 optimisation events in that week is marked Learning limited (About learning limited).
- Meta’s fix for a thin event. Among its fixes for Learning limited, Meta suggests choosing an event that occurs more frequently, for example moving from purchases to add to basket (About learning limited).
- Deeper goals cost more. A conversion may cost more than a landing page view, so Meta says a daily budget should in general be at least ten times the average cost of your performance goal: at $5 per link click, at least $50 a day (About performance goals).
- Value optimisation has entry requirements. Maximise value of conversions needs value signals sent through the Conversions API (required) and the Meta pixel; Meta suggests budgeting for at least 50 conversions a week and testing for at least three weeks (About maximising the value of conversions).
- Changing the event restarts learning. Any change to the optimisation event is a significant edit (Significant edits and learning phase).
Can software help?
Bach.ai connects to your Meta ad account, audits it, ranks what it finds by estimated revenue impact and proposes specific fixes. Its conversion-volume check counts each ad set’s optimisation events per week against the 50 mark and, when an ad set falls short, proposes consolidating ad sets or a higher-volume event. It applies a change on Meta only after you approve it.
FAQ
How do you choose the deepest valid optimization event in Meta Ads?
Select the deepest valid event aligned to the business outcome that meets the reader’s reconciled volume and integrity thresholds. Meta’s volume reference is about 50 optimisation events in the week after the last significant edit; check it in Ads Manager by adding the Optimisation events column (Columns, then Customise columns).
What should you do when the preferred optimization event has thin or unreliable signal?
A thin or mislabeled event requires a documented proxy risk and exit condition. Meta’s own fix for a learning-limited ad set is to optimise for a more frequent event, for example add to basket instead of purchase, or to combine ad sets so that one ad set gets the volume.
What can an optimization-event matrix support without proving incremental impact?
It supports the bounded operating choice encoded by the optimization-event depth matrix. It cannot replace missing source records or turn platform credit and observed association into incremental impact. Meta counts optimisation events under your chosen attribution setting, so the counts are attributed results, not proof of cause.
What does an “ad set may get zero purchases” warning mean?
Whatever the wording, the documented version of that problem is Learning limited: an ad set unlikely to get about 50 optimisation events in the week after its last significant edit. Meta’s causes include a small audience, low budget, low bid or cost control and an infrequent event; its fixes include combining ad sets or a more frequent event.
Method and sources
Every Meta rule and goal name on this page was checked against the linked Meta Business Help Centre page on 1 Oct 2026. Your thresholds and integrity checks come from your own event logs and order ledger. If you optimise for value, the bid side is covered in how do I choose a Meta bid strategy?; for the terms, see the Meta ads glossary.
Sources: Meta Business Help Centre, About performance goals, About maximising the value of conversions, About the learning phase, About learning limited, Optimisation events, Significant edits and learning phase (checked 1 Oct 2026).