With few levers and a lot of automation, management quality is measured in discipline: what you touch, when and on what criteria. Here is the routine.
Learning and what resets it
After creating the campaign or making a large change there is a learning period — on the order of one to two weeks — with volatile results: costs above target, uneven spend, volume in fits and starts. What restarts that countdown: changing the bid strategy, moving the target or budget a lot, changing the conversion event, and refreshing all creatives at once.
The practical consequence is uncomfortable but simple: a freshly touched app campaign cannot be judged. If your client asks for explanations on day three, the right answer is to show them the learning calendar you agreed on, not to touch anything else.
Change cadence
- One lever at a time: budget or target, never both.
- Moves of around 20% maximum, with two or three days between changes. The reference is repeated across documentation and agency practice because underneath sits the same probabilistic engine as in the Smart Bidding module.
- Scale when the campaign is stable, not when it has a good day: the practical reference is about a week of consistent costs before raising budget.
- Come down slowly too: abrupt cuts are large changes and reset learning just the same.
When to stop scaling
When cost per install or per action rises persistently above your profitable threshold and doesn't recover within a few days, you have hit the ceiling of that segment with those creatives and that goal. The ways out: new creatives (lesson 6), another market (lesson 4), or accepting the ceiling. Raising budget further only buys worse volume.
Diagnosis tree
1. The campaign isn't spending. Target too demanding for what the system can find; budget below the minimums; insufficient creatives to access the surfaces. Order of review: assets → target → budget.
2. Cost per install spikes. Did something of yours change (bid, assets, budget)? Did the listing change (update, reviews, average rating)? Is there seasonality or new competition? Have the creatives run out (same volume, rising CPI)?
3. Lots of installs that never open the app. You are optimising the wrong metric, and possibly buying junk traffic. Climb a rung on the bidding ladder and go straight to lesson 10 (quality and fraud).
4. Volume has stalled. Segment ceiling: new creatives, new market, or loosen the target to enter more auctions.
5. The numbers don't match across systems. Normal, especially on iOS (lesson 3): different windows and models. Pick one official source, document the difference and decide with one.
What to check every week
A short, sufficient dashboard: spend and its split by campaign; cost per install and per value event; day-1 and day-7 retention by campaign or country (lesson 10); revenue per install if you monetise; asset ratings; and the listing's average rating. With that you know whether the problem is in the campaign, the creative, the listing or the product.
💡 Ninja trick: log every change with its date and reason in a document — it is the manual version of the "timestamp" trick from the Lead Scoring module. In a system with two weeks of latency, remembering what you touched ten days ago is worth more than any report.
⚠️ Pitfall: on Friday afternoon, changing the target, the budget and uploading new creatives "so it works over the weekend". On Monday you will have a campaign relearning from scratch and no way to know what caused what.
What you should remember
- Learning takes 1-2 weeks; large changes reset it.
- One lever at a time, ~20% max, every 2-3 days.
- Scale on demonstrated stability; stop scaling when cost doesn't recover.
- Diagnose in order: assets → target → budget → listing → product.