The optimisation loop: read, decide, change, wait
Every optimisation decision goes through the same four stages, and skipping one is where accounts go wrong. Read means looking at the right window, the right level and the right metric: results over a period long enough to contain real conversions, at the ad-set or ad level where the decision will be made, on the metric the campaign is meant to move. A day of cost per acquisition (CPACost of each conversion.) on an ad set with a handful of results is noise. Decide means asking whether the number is outside the range you would expect from chance and from the week-to-week movement of the auction, and whether a cause is visible. Change means one thing, with a reason written in change history, because a change without a reason cannot be evaluated. Wait means leaving the edit alone long enough for delivery to re-learn and for the attribution window to fill in, which for conversion campaigns is usually days rather than hours.
The loop has a natural rhythm. Daily checks look only for breakages: spend stopped, an ad rejected, a tag gone silent. Weekly reviews make the real decisions. Monthly reviews ask structural questions about campaign structure, creative supply and measurement. The temptation to act daily on conversion metrics is the most common way to keep an account permanently in learning.
What to change and what to leave
Meta documents that significant edits restart the learning phase, and delivery works best with stable ad sets and enough results. That shapes the list of levers.
| Lever | Touch it when | Leave it when | Why |
|---|---|---|---|
| Creative | An ad's results decay, or a new concept is ready | An ad is still winning and fresh | Creative is the strongest lever and the safest to add to |
| Budget | Results are consistently good or bad over a full window | A single day looks off | Large budget changes restart learning; small ones usually do not |
| Audience | An audience has clearly saturated or is the wrong people | Delivery is finding results on broad targeting | Narrowing often raises cost per thousand (CPMCost per 1,000 impressions.) without improving results |
| Placements | A placement shows poor results with enough volume to judge | Volume per placement is thin | Removing placements shrinks the auction pool |
| Bid strategy | Economics demand a cost or return guardrail | Volume is the goal and results are acceptable | Caps trade volume for predictability |
| Optimisation event | Measurement shows the current event is wrong | The event works and has volume | Changing it is a full restart |
| Attribution setting | Reporting is being reconciled with another system | Mid-test | Changes what is counted, not what happened |
The pattern is that creative is the lever to work hardest, budget the one to move gently, and the structural settings the ones to get right once. The creative hub covers how to keep a supply of new ads flowing, which matters more than any targeting adjustment.
The core guides
- Scaling Meta campaigns — vertical scaling through budget steps that respect learning, horizontal scaling through new audiences, creative, placements and markets, the bid strategies that make scaling safer, and the marginal-return rule for knowing when to stop.
- Frequency and creative fatigue — how FrequencyAverage times a person saw an ad. accumulates, the difference between an audience that is saturated and a creative that has worn out, the signs in the data, the frequency controls Meta provides, and refresh cadence.
- Troubleshooting performance drops — a diagnosis sequence for sudden drops that starts by checking whether the drop is real, then works through learning resets, auction shifts, fatigue, policy and site changes.
Read the core guides in that order if the account is growing; start with the guide to troubleshooting performance drops if something has just gone wrong.
Further guides
Further guides cover what sets the price of results, spending less without breaking what works, and planning around a sale or seasonal peak:
- What drives the cost of Meta ads — how the auction prices impressions, what moves the cost per thousand, how click-through and conversion turn it into a cost per result, and estimating costs before launch.
- Reducing spend on Meta — where to cut first, lowering versus pausing, consolidation and what to protect, without sending working ad sets back into learning.
- Planning Meta Ads around a sale or peak — launching early enough to clear learning, readying creative and audiences, budgeting the window and reading results afterwards.
Targets from economics
Optimisation without a target is just activity. The target comes from the business, not from the platform: what a conversion is worth, what margin it carries, how much of that margin can be spent acquiring it. For e-commerce the anchor is Break-even ROASMinimum ROAS to avoid a loss., the return on ad spend at which a sale neither makes nor loses money after cost of goods and fulfilment; the break-even calculator derives it from margin, and the ROAS calculator turns a target ROASRevenue generated per unit of ad spend. into the spend a given revenue allows. For lead generation the anchor is the value of a lead, worked back from close rate and customer value.
Two refinements matter on Meta specifically. First, the platform's reported results are attributed under its own window and include modelled conversions, so the target should be set on the number the business trusts, with a known relationship to the platform's number; the measurement hub covers the discrepancies. Second, the target that matters for scaling is the marginal one: the cost of the next conversion, not the average. An account can show an average cost well under target while the last increment of budget bought conversions well above it. The scaling guide works through that rule, and the media planning framework sets the budget context it sits in.
Common optimisation problems
| Problem | What is usually behind it | Where to go |
|---|---|---|
| Results swing wildly day to day | Reading too short a window with too few conversions; edits restarting learning | The loop above; the learning phase guide |
| Doubling the budget halved the efficiency | Budget step too large; marginal cost rising past the average | Scaling guide |
| Costs creep up with no change made | Frequency climbing on a fixed audience, or a winning ad wearing out | Frequency and fatigue guide |
| Performance fell off a cliff overnight | Measurement break, rejected ad, site change, or an edit that reset learning | Troubleshooting guide |
| Spend will not reach the budget | Bid or cost cap too tight; audience too small; ad rejected | Scaling guide, bid strategies section; troubleshooting guide |
| Cheap results that do not turn into revenue | Optimisation event too high in the funnel; attribution setting too generous | Campaign objectives guide; attribution settings guide |
| Every ad set lives in learning | Too many ad sets splitting too little budget; constant edits | Campaign structure hub |
| Traffic has never converted since launch, as opposed to falling | Goal, events, placements, landing page or attribution set up wrongly at launch | Clicks but no conversions troubleshooting |
| Click-through is weak from day one rather than decaying | A failing hook, format, audience or offer layer | Low click-through rate troubleshooting |
| An ad set has underspent its budget ever since launch | A cost goal below the going rate, a squeezed audience, a rare event, overlapping ad sets or trimmed placements | Meta ads not spending budget troubleshooting |
| Lead forms fill up but sales cannot book the leads | A volume form, no qualifying question, submissions as the goal, a loose seed or slow follow-up | Low-quality lead ads troubleshooting |
| Cost per result rose and nobody can say why | The cost read as one number rather than split into the price of impressions, click-through and conversion | What drives the cost of Meta ads |
| A budget cut lost far more results than the saving suggested | Every ad set cut equally, or working ad sets sent back into learning | Reducing spend guide |
| A sale or seasonal peak delivered little while campaigns were still learning | Campaigns launched or edited too close to the window | Sales and seasonal peaks guide |
The author's Meta Ads expertise page describes how this loop runs on real accounts and the review cadence behind it.
