Handing ad decisions to software can save hours, and it can also spend money quickly and quietly. The answer isn’t to avoid automation. It’s to give it room in stages, the way you would with a new colleague.
Stage 1: Watch
Let the software say what it would do, and change nothing. Run this for long enough to see an ordinary week and an unusual one. Ask yourself:
- Would you have made the same call?
- Does it explain why, with numbers you can check?
- When it’s wrong, is it wrong in a harmless direction?
If you can’t tell why it did something, stop there. A decision you can’t check is a decision you can’t trust.
Stage 2: Approve
Let it propose real changes that you confirm one at a time. This is where you learn its habits, and where you’ll spot rules that don’t suit your store.
Stage 3: Small, safe moves only
Allow it to act alone only on moves that are cheap to get wrong. Lowering a budget or pausing a campaign limits your spending. Raising a budget does the opposite, so keep that decision for yourself.
Limits worth setting first
| Limit | Why it matters |
|---|---|
| Total daily budget | Your worst-case spend for a day |
| Maximum change per step | Stops one bad reading from causing a big swing |
| Cooldown | Gives a change time to show its effect before the next one |
| Automatic changes per day | Caps how much can happen while you’re away |
| Protected campaigns | Keeps your proven campaigns out of reach |
Questions to ask any tool
- Where are my limits enforced? They should be checked on the server, not just in the screen you see.
- How do I stop everything at once?
- Is every action recorded, with who did it and why?
- What happens if a request fails halfway? It shouldn’t retry blindly and create duplicates.
- What does it do when data is missing? It should say so, not guess.
Keep looking
Automation doesn’t remove the need for judgement. It changes what you spend it on. Check the log weekly, and whenever your store changes, such as a sale or a new product line.
You can test a change against a set of limits in the demo.