Module 6 - Testing and learning

Why is testing and learning something to be careful with?

Testing and learning is one of my favourite subjects, and it genuinely never stops, but the first thing I want to warn everyone about is not to get lost in it.

Do not make so many changes that you have no idea what is working, and do not spend half a day going through very specific analytical detail unless you are spending tens of thousands a month, because then you can afford to. It is very easy to spend an entire day, and I am guilty of this myself, looking at accounts and ending up so overwhelmed that you do not know what to do next. Especially when you are starting off, it helps to be aware of just a few things and to test only one or two things at a time. If you have got a low budget, or even a decent budget, you could test a few things at once, but do not go down the rabbit hole and do not overthink it.

What can a simple test tell you?

A simple example is that once, for a high ticket product, we thought there were ten sales, and Google had attributed eight of them to TV screens, which is a device within YouTube.

That is normally one of the things you would take out, but because eight sales had come from it, we decided to take that campaign out and give TV screens its own specific budget to see how it performed. That is a very simple version of a key insight you can act on. You look at your account, and one of the metrics I would always go for is cost per conversion, and then cost per acquisition, which I covered in an earlier module. Focus on that first, rather than getting lost straight away in whether it was women or men, or whether there were more sales on tablets.

Why shouldn't you trust your instincts straight away?

Because something that did not work on another platform might work very well on YouTube, so it is almost like starting on square one.

You want to test a few specific things at the beginning, and after a month or two you will know what is working. It is something you build up over a few months, until you know the audiences that work best and what to say in your videos. You might also find, a couple of months in, that certain devices are not right for you. For example, you could turn TV screens off if the domain you are sending someone to is complicated, or if your brand is not well known on Google, because people cannot be bothered to type the address in when they see it on a screen they cannot click. Normally you would turn TV off anyway, because you would ask people to click the button or the link below, and you cannot click on TV.

How does testing spiral into a rabbit hole?

It happens when one finding leads you to start a new campaign for every variable at once, until you have far more campaigns than you can properly read.

You might run one campaign for a week and see that over half the sales were on mobile, so you create a mobile campaign, then notice that a lot of those buyers were women, so you create a mobile campaign for women and another for men, which is already four campaigns. Then you notice that Google spent a hundred pounds on one video and fifty on another, so you want to test the budget on both, and test all of the new mobile campaigns, and suddenly you are at eight campaigns before landing pages have even come into it. You end up testing so much at once that you cannot tell what is working, so keep it as simple as you can, especially when starting out, so you know exactly which hooks and which audiences are working, and where to start when it comes to scaling.

Why does testing and learning never really stop?

Because even once you know what is working, an audience can start to fatigue, meaning everyone in that target market has already seen the ad, so results begin to drop and you need to test new audiences.

You are always going to be looking at your ads, your cost per conversion and your cost per acquisition, and always trying to make one percent improvements every day. As Simon says elsewhere in this course, if you are making constant one percent improvements every day or every week against someone who does a spray and pray approach, throwing a few ads out and leaving them running for months, you are going to completely wipe the floor with them.

How long does YouTube's learning phase take?

It varies. New ads can learn within the first week to the point of profitability, but it can take up to 60 or even 90 days, depending on how big the audience is and how many leads or sales you get.

The more leads you get, the more it learns, and the more sales you get, the more it learns. If you unlock conversions as a goal for your campaigns, you might duplicate the campaign, change it to cost per conversion, and test that. Be patient, because you might get good results in week one and then see things dip or tank in week two. At that point it could be something to do with placements, it could be bots, or a few other things, but once you deal with them the learning starts again.

What did a retargeting insight actually look like?

For one client, add-to-carts were much higher on mobile phones among an older demographic, but they never actually paid on their phones, always finishing the sale on desktop or laptop.

So we set up a retargeting campaign to the desktop, because we knew that was where they would finish the payment, and retargeting is the cheapest and most direct way to get back in front of someone right when they are on the device they are going to buy on. You would test that for two or three weeks, and if it works well you increase the budget, while running a few other audiences at the same time. Testing gradually shows you what your instincts are actually built on: data from your own account, and eventually the two work together, at which point your YouTube account really starts to move.

What should you test first: ad creative?

Ad creative, and specifically the hooks, because too often people go live with just one creative to start getting paid, and if the hook or the story in the middle is bad, the whole ad fails even if targeting and the landing page are excellent.

The hook is the first six seconds of the video, and it is one of the most important things to test. Test different hooks, at least two or three, maybe five or more, around themes like wealth, health or family, or different stories, or even the same script with different b-roll footage. If you have a lower budget, test two or three at once in the same ad set so Google can learn from them.

What comes after the hooks?

The story, meaning the part in the middle that pushes on pain points and shows you are the expert.

Test different approaches here to see what works best in your niche. I have seen people run three strong hooks where the middle did not correlate with them, so the hook got people watching but the story did not resonate, and they left. Whatever hook you use, make sure the story stays on the same point, though it does not need to be entirely different for each hook.

Do you need to test the call to action?

Not really, at least not at the start. Normally you would just say click here to find out, or click here to buy, either pointing to the button directly or mentioning it somewhere in the video.

That is not something worth testing early on. If you are on a small budget, start with the ad creatives and one or maybe two audiences.

What should you test next: targeting?

Competitor audiences, because YouTube allows intent based targeting, letting you reach anyone who has been browsing or engaging with a competitor's website.

Google adds those people to a list you can target, saying essentially, what about us, we are better than them. They are already engaged in the market and likely to have the problems your product or service solves, so it is a safe and good audience to start with. If you have a bigger budget, put all your ad creatives in one campaign and target one audience at a time, rather than putting several audiences in the same campaign, because Google will simply pick the best audience in that campaign and put all the budget behind it.

What is placement targeting?

Placements let you choose specific YouTube videos or channels for your ad to appear on, built using software covered elsewhere in the course that pulls active, regularly posting videos in your niche.

For a coaching product you might choose big influencers in that space, and for health you could search YouTube for health content and add those channel URLs to your placements list, so your ad shows on very targeted, specific videos. On a budget of around a thousand pounds a month, I would run three creatives against two audiences, including competitor targeting and placements, and leave that for at least two weeks, ideally three or four, to get through the learning phase, checking in to make sure conversions are coming through.

How do you decide on bidding strategy and campaign objectives?

Slowly, and only once your creative and targeting are working, because bidding is about finding the right price to get your ad served without wasting budget.

On YouTube you might bid a set amount per video view, and spending ten pounds a day at a low bid could get you thousands of views, whereas a CPM campaign, where Google charges per thousand impressions, needs a high enough bid or your ad simply will not get shown. Bidding a little more can sometimes get your ad shown ahead of others and improve traffic quality, so it is worth testing once the rest is dialled in.

Campaign objectives are linked to this. If your account average cost per conversion is fifteen or twenty pounds and you set a target of ten because that is what you want, you are mostly just crossing your fingers, and the real effect is that Google will not show your ad at all. Google will suggest a realistic target, for example twelve pounds if that is close to your average, and it is worth letting Google take the wheel there. Optimising for cost per conversion rather than cost per click tends to bring people who are more likely to actually convert, fill in a form or buy, rather than just watch a video.

When should you look at demographics and devices?

Last, because there is no way of knowing this data until you have tested everything else, but it becomes valuable once you have enough of it.

If, after two months, eighty percent of conversions are coming from the twenty one to thirty five age range, you can build a campaign specifically for them. If that same group is buying mostly on mobile because they are younger and more tech savvy, you can combine age and device into one targeted campaign with its own budget. It is also worth checking, once you know most of your traffic is mobile, whether your landing page actually resizes and looks good on a phone, because if it does not you could be losing a lot of traffic and sales.

What does learning actually mean in practice?

Learning is simply looking at what worked best, using the conversions you set up, whether that is page views, landing page forms, or full sales, and reading the numbers Google gives you.

If it is set up correctly, Google will show you how many conversions and sales you got, which campaign produced them, whether they came from mobile, and whether buyers were mostly women or men. Going back to the earlier point, do not test fifty things at once. Test one or two, so that if conversions go down you know the decision was wrong, and if they go up you know exactly what change caused it. Test twenty things at once and it becomes an overload of data where you do not know where you stand, and you are more likely to give up out of frustration.

If a campaign has spent hundreds of pounds with zero conversions, check your CRM first to rule out a tracking error, but if tracking is working on your other campaigns, it is safe to assume whatever that campaign is doing is not working, so pause it, turn it off, or duplicate it and try something different. Do not start a campaign completely from scratch unless you are testing a genuinely new audience or creative. If you have a big campaign running to a broad audience across five creatives and all devices, duplicate it before narrowing it down to the one creative or audience you want to isolate, because that keeps everything YouTube has already learned rather than starting again from zero.

Where do you actually find this data inside Google Ads?

Inside the account, under demographics, conversions, placements, audiences and devices, all broken down by campaign or across all campaigns at once.

In one remarketing campaign, clicking demographics and gender showed eighteen conversions in total, split around eleven men to seventeen women, nothing dramatic enough to act on. Household income can be tracked in some countries, including America and Canada, but not in the UK. Under placements, you can see exactly where an ad was shown, including the occasional odd or low quality placement that is worth excluding, and there is a whole section elsewhere in the course on exclusions, including flags for suspicious activity such as unusually cheap clicks paired with a very high click through rate.

Under audiences, you can compare how different audiences perform, even when Google has funnelled most of the budget into the strongest one. Under devices, you can see the split across TV screens, tablets, mobile phones and computers, along with the bid adjustments applied to each, and use that to decide where to increase budget or cut spend, for example dropping tablets if they are getting the budget but very few conversions. You can view all of this at the level of a single campaign, or switch to all campaigns to see the full account picture, including which audiences and conversion actions, such as time on landing page versus calls, are performing best.

I hope that was helpful. If you have any questions, you can email me as always at s.media, or ask in the Facebook group or wherever else in the community, and I am happy to help.