Optimisation is a loop, not a lever
New performance marketers look for levers: a bid to raise, an audience to add, a setting to change. The platforms encourage this with recommendation panels that always have another suggestion. Most of those changes are noise. Automated bidding re-learns after each one, the daily figures move for reasons unrelated to the change, and after a month of daily adjustments nobody can say what caused what. The experienced approach is slower and produces more: fewer changes, each with a hypothesis, a fair comparison and a written result, so that the account's history becomes a record of what is known to work.
The experimentation loop framework describes the general cycle; applied to paid acquisition it has five turns. Observe where the result is short of the target and at which stage of the chain. Form a hypothesis about a lever and the effect it should have on cost per acquisition (CPACost of each conversion.) or return on spend (ROASRevenue generated per unit of ad spend.). Design a test the platform cannot bias, whether an A/B TestControlled comparison of two variants. on creative, a split on landing pages or a geographic holdout. Read the result against what was predicted, not against what was hoped. Roll out, record, and choose the next hypothesis. Reporting is what closes the loop: a decision log that says what was tried and what happened is worth more than any dashboard.
Test, scale, report
| Activity | What it decides | Cadence | Guide |
|---|---|---|---|
| Test | Which lever to pull next and whether it worked | Continuous, one or two tests live at a time | Testing and scaling |
| Scale | Whether more budget buys results at an acceptable marginal cost | After a test wins and the guardrails hold | Testing and scaling, Target-led paid media budget |
| Report | What changed, why, and what the review should decide | Daily checks, weekly optimisation, monthly review | Performance reporting |
| Prove | Whether the spend adds results at all | Quarterly or at each major budget change | Performance reporting |
| Restore | How money returned after a cut goes back in, and how far it should go | When a restore trigger agreed at the cut is met | Restoring paid media spend after a cut |
The order matters. Scaling what has not been tested spends more to learn the same lesson; reporting without tests describes the weather; testing without reporting loses the knowledge within a quarter.
The core guides
Testing and scaling performance campaigns orders the levers by expected effect and cost, explains how to design a fair test when platforms overlap and audiences are shared, shows how to read results without being fooled by noise or by platform attribution, and sets out the four scaling paths, budget, audiences, channels and markets, with the guardrails that detect diminishing returns before they become losses.
Performance marketing reporting and reviews builds the cadence of daily checks, weekly optimisation and monthly review, gives a report structure of plan, efficiency, volume and marginal view, explains how to reconcile platform numbers with analytics, and introduces the decision log that turns reviews into a record. It routes the attribution choice and the diagnosis of discrepancies to the analytics root, and shows where IncrementalityConversions caused by marketing. tests and modelling answer the questions dashboards cannot.
The statistical mechanics of testing, sample sizes, significance and the common errors, live in the CRO testing section, which this hub links rather than repeats. The measurement plan framework is the document that ties the core guides together: what will be measured, by what rule, reviewed when.
Further guides
Further tutorials turn the targets into a budget that finance can sign and bring spend back after a cut:
- Agreeing a target-led paid media budget — a budget that follows results: the target, a stopping rule, cash and payback limits, review points, platform settings and reporting.
- Restoring paid media spend after a cut — bringing money back when a restore trigger is met: the starting settings, what decayed, the order of return, targets and budgets changed apart, paused campaigns, prospecting before remarketing and the marginal cost at each step.
Creative and landing pages as the biggest levers
Across platforms, the two levers that move results most are the ones that sit outside the platform's settings. Creative decides whether anyone stops, and in paid social it is close to the whole game: the platform's delivery system finds the people, the creative decides whether they act. A new concept that works will outperform any amount of bid and audience adjustment, and creative fatigue is the most common reason a campaign that worked stops working. The landing page decides whether a click becomes a result, and its conversion rate multiplies everything upstream; doubling it halves the cost of every result without touching the campaign. The landing pages section of the CRO root covers how to build and test them, and the testing guide here sets out when a landing page test is the right next move.
The practical consequence is that the optimisation calendar should be dominated by creative production and landing page tests, with bidding and targeting changes as the smaller, less frequent adjustments they are. Accounts where the reverse is true, daily bid changes and a creative refresh twice a year, are the ones whose results drift down regardless of effort.
Working with automated bidding across platforms
Every major platform now bids for you. Google's help centre documents bidding strategies that aim at a target cost per action or return on spend; Meta's documents the equivalent cost and return controls. The systems work by predicting, auction by auction, the probability that a given impression leads to the conversion they have been told to value, and bidding accordingly. In practice they do this better than a person can, provided three conditions hold: the conversion event is right, there are enough conversions for the prediction to be learned, and the target is one the economics can actually support.
Optimising with automated bidding is therefore mostly about those conditions. Feed the right event with values where they exist. Consolidate structure so each campaign sees enough results. Set targets from the economics and change them gradually, because large moves send the system back into learning. Resist the daily tweak. And watch for the failure mode automation cannot see: it optimises towards reported conversions, so it will happily spend more on the audiences most likely to convert anyway, which is why the remarketing layer and the holdout test matter more under automation, not less.
A subscription box swaps daily tweaks for a loop
Pantry Crate, a meal-kit subscription, had been adjusting bids and audiences daily across Google and Meta while its cost per new subscriber crept from £52 to £68 over six months. The new lead froze bidding and targeting, and ran one test at a time. A creative test with four new concepts against the control found one that cut Meta's cost per subscriber to £49. A landing page test moving the price and the first-box offer above the fold lifted conversion rate from 3.1% to 4.4% on both platforms. A geographic holdout showed remarketing added only a fifth of the subscribers it claimed, so its budget was cut by half and moved to prospecting. After one quarter the blended cost per subscriber was £45, with a decision log of twelve entries explaining every change.
Common performance optimisation problems
| Problem | Usual cause | Guide |
|---|---|---|
| Results drift down despite daily changes | Changes without hypotheses; automation re-learning after each | Testing and scaling |
| A test "won" but the rollout did not improve anything | Noise read as signal; platform attribution flattered the variant | Testing and scaling, CRO testing |
| Efficiency collapsed when budget was raised | Scaled past the marginal cost ceiling with no guardrails | Testing and scaling |
| A fixed monthly budget stops spend that would have paid, or pushes spend that will not | Budget set as a number rather than an agreement on a target, a stopping rule and a ceiling | Target-led paid media budget |
| Money returned after a cut buys less than it did before the cut | Everything switched back on at once, onto stale lists, tired creative and a smaller remarketing pool | Restoring paid media spend after a cut |
| Creative worked for a month and then stopped | Fatigue; no production pipeline for new concepts | Testing and scaling |
| Clicks are fine, results are not | Landing page, not campaign, is the constraint | Landing pages |
| Weekly review produces no decisions | Report built around platform metrics rather than plan versus actual | Performance reporting |
| Platform totals exceed what the business recorded | No reconciliation; attribution overlap unrecognised | Performance reporting |
The performance marketing hub places optimisation within the whole system; the fundamentals hub covers the targets this loop optimises towards.
