Reading the data, and killing what does not work

An advertising account produces an enormous amount of data, almost all of which is capable of making you feel busy without making you money. The useful subset is small, and it starts with a question most accounts cannot answer honestly: which clicks turned into customers?

Conversion tracking comes first

Until a sale, enquiry or booking can be traced back to the search that produced it, every other number is decoration. Clicks and impressions tell you about attention, not money, and optimising towards them produces cheap attention from people who were never going to buy.

• Track the thing that is actually worth money, not the step before it. A form submission is not a customer, and if the two diverge you will optimise for the wrong one.
• Track phone calls if people phone you. A large share of local businesses measure the half of their response they can see and conclude the channel is weak.
• Count a conversion once. Duplicate tracking inflates everything and is depressingly common after a site is rebuilt.
• Check that it still works after any change to the site. Broken tracking looks exactly like a sudden drop in performance, and people often react to the drop instead of the cause.

Checking it properly

Most broken tracking is broken quietly, and the account looks like a performance problem rather than a measurement one. Four checks, none of which takes long.

1. 01Submit your own form or make your own booking, end to end, and confirm the conversion appears in the account. Do this from a phone as well, because mobile often uses a different template.
2. 02Confirm it appears once. Two tags, or a tag firing on a page that reloads, will double every number you have.
3. 03Check what counts as a conversion. If page views of a thank-you page count, confirm that page cannot be reached any other way, including by somebody bookmarking it.
4. 04Set the conversion window deliberately rather than leaving the default. For a service people consider for two weeks, a short window hides real conversions and makes the channel look worse than it is.

Reading movements without fooling yourself

Small numbers move a lot for no reason. A keyword with eleven clicks and one sale has a conversion rate that will swing wildly on the next click, and acting on that swing is how accounts get worse one confident decision at a time.

A test you can do in your head, which catches almost every bad call: ask what happens to your conclusion if the next two conversions land on the other side. One keyword at two conversions from forty clicks, another at zero from thirty. If the next two conversions go to the second keyword, they are level. You do not have a result, you have noise with a preference.

Give a change long enough to produce enough events to judge it. What counts as enough depends on how often you convert, but the instinct to decide after a day is almost always wrong, and it is strongest exactly when you have just spent money.

Killing what does not work

Most accounts are carrying something that has never produced a customer and has been running for months, kept alive by the thought that it might come good. Decide the threshold in advance, before you are attached to it.

1. 01Set a cost ceiling per keyword: the point past which, with no conversion, you stop. Work it out from what a customer is worth rather than from how you feel.
2. 02Review against that ceiling on a schedule, not when you happen to look.
3. 03Pause rather than delete, and write down why. Six months later you will want to know whether you already tried it.
4. 04Move the budget to what is working rather than banking it. The point of cutting is to reallocate.

One decision, worked through

Illustrative. A customer is worth 600 in profit and your ceiling is therefore, say, twice that in spend with nothing to show: 1200. Three keywords are in front of you.

• One has spent 1400 and produced nothing. It is past the ceiling. Pause it, with a note. This is not a close call and it will still feel like one.
• One has spent 300 and produced nothing. It is nowhere near the ceiling. Leave it alone; there is no information here yet and pausing it buys you nothing but a tidier screen.
• One has spent 900 and produced one conversion. It is inside the ceiling and it is working. Leave it, and check again when it has roughly doubled its spend.

The discipline is that all three decisions were made by the number you wrote down before you looked. That is the point of writing it down: on the day, the first keyword will have a story attached and the second will feel like it is failing.

A keyword that cost three times what a customer is worth and produced nothing is not unlucky. It is answered. The hard part is that the decision feels like giving up, when it is the only way to find out what the rest of the account can do with that money.