How to choose a bid strategy

A budget is how much money is allowed to leave. A bid is what you are willing to pay for the thing you want, and it is the number that competes.

Between those two sits a spectrum of how much of the pricing decision you hand over. At one end you state a price and the system uses it. At the other you state an outcome — a cost you want to average, a return you want to hold — and the system prices every auction itself, aiming at your number and missing it in both directions while it learns.

That is the whole taxonomy, whatever the platforms call their versions of it. The decision is how far along that spectrum to sit, and everything below is about which inputs move you.

How much do you know about what a result is worth. If every conversion is worth roughly the same and you know the number, you can state a price and mean it. If conversions vary by an order of magnitude and the variation is visible to the platform through the values you send, an outcome-shaped strategy has more to work with than you do. If conversions vary and the variation is invisible to the platform — it happens in your billing system three months later — then neither of you knows, and stating a confident number is worse than not stating one.

How much can you afford to lose while it learns. Handing over pricing means a stretch of delivery that is worse than what you had while the system works out who to show ads to. That cost is real and it is paid up front. A campaign with a small budget and a short window may never get past it, which makes the more automated choice the wrong one regardless of its merits at scale.

How much volatility the business absorbs. Outcome-shaped bidding is an average. Some days will be much worse than the average and some much better, and the arithmetic only works if you can let the bad days happen. A business that cannot — a fixed margin per unit, a cash position that will not tolerate a bad week — wants a ceiling rather than a target, which is a different decision covered in When to use a cost cap.

A target is not a cap.

A target is a goal you hand the system. It will come in above it and below it, and the promise, where a platform makes one at all, is about an average over time rather than about any single result. Reading a target as a ceiling is the most expensive misunderstanding in this subject, because it produces a month of results above the number and a conversation about whether the platform is lying.

A cap is a refusal: where a platform offers one, it holds the line and takes the volume loss instead. It buys predictability with reach.

Whether predictability is worth reach is a question about your business rather than about advertising, and it is the question this whole decision reduces to.

It will not tell you that one strategy beats another, and it will not print a table of thresholds. A figure stated as a fact about the world needs a real source, and the numbers circulating about how many conversions a system needs before it stabilises are vendor claims rather than sourced facts. Nobody outside a platform can tell you where those lines are.

What is true and checkable is your own account's history: how long previous campaigns took before their cost per result settled, how much it moved afterwards, and what happened the last time somebody changed a strategy mid-flight. Those numbers are in the account and they are the ones the decision should run on.

A bid change in Tempio is a change request like anything else. The payload carries whichever fields apply — an amount, a target cost per acquisition, or a target return, which is treated as the ratio it is rather than as money. Guardrails are checked when the change request is written, again at approval, and once more immediately before a platform is called, using the same functions each time so the sentence you read at the gate and the decision at the apply cannot disagree.

The arrangement, stated once: your connections read your accounts, a change is written as a draft on the record with the state it would change from and the checks it passed, applying it is gated on approval — by you, or by a rule you wrote — and no platform write reaches a customer account until that platform has granted write access. Applying a cleared change to a platform is in private beta.

A rule you wrote can clear a bid decrease under a cap it states in a currency it states. An increase always waits for a person, as does any change whose direction cannot be read against the entity's previous value — no pre-image, or a field the account's mirror never held. "Cannot tell" is a refusal rather than a pass at a gate that spends money.

The part that matters afterwards is the record. Every applied change is a row with what it moved and from what, so "what did we change on the fourteenth, and what did it do" is a question with an answer rather than a discussion. That is what makes a strategy change readable, and a change you cannot read the effect of is worth less than one you can.

If you want one sentence to act on: start a new campaign on whatever asks the platform for the least, get enough volume that its numbers mean something, then add a constraint if and only if the business needs one. Change one thing at a time, change it deliberately rather than often, and write down what you expected before you look at what happened.

Set a target CPA that holds is the next decision if the answer was a target, and How budget changes restart learning is what every one of these changes costs in delivery.

Why this guide prints no benchmark table, and what else is refused for the same reason, is What we refuse to claim, and why.