Hello, everyone! My name is Yana and I’m a Google logistician. I’m engaged in delivering ads to the target audience of a business with the necessary parameters. I scale online business sales in different countries using the same AI tools of Google Ads.
In today’s video, we’re going to analyze several advertising campaigns. We’ve had a bit of a break here. I’ll explain everything now. The topic is swimming pool accessories. We had several southern states in the US to target. So, what do we have? We have several variants of advertising campaigns. At first, we launched much more of them. Which, in principle, is evident from the amount of traffic we received in a certain week of December.
How many calls and sales will I get by ordering contextual advertising from you?
But what were the difficulties we faced? The difficulties were most likely on the client’s side. He had a rather slow website. Then, in the process, it turned out why it was quite slow. This is a confirmation of the fact that no matter how high-quality the audience is – if your site does not meet the minimum technical requirements from the point of view of the customers themselves, then it is impossible to talk about something super effective, no matter what kind of traffic you get. Let’s take the daily data. You can see here that there were not many of them. But, for example, the day when we gave a little more clicks, that’s out of the question. Here, for example, the number of purchases was much better.
Why did we have this gap period? We didn’t advertise at all and had no clicks. Because this whole period of time, when we told the client that we really have a problem with the fact that you have some nuances with the speed of the site. That is, not all clients want to deal with this, to believe in it. Until you show the real number of clicks, you compare it with the number of sessions, which is many times less. That is, accordingly, it is clear that no matter what traffic we bring, the speed of the site has a very negative impact on it. Of course, the client has been negotiating with the contractor on how to eliminate this whole story all along, so this month of launch is stretched over two periods of time.
That’s why it’s basically out of the question to say that this is a super cool result or something else. It took us a couple of weeks to launch and a couple of weeks to produce. This is one month of broken history. Therefore, it is better not to break a month. That is, it should be a systematic growth, scaling in order to be able to directly influence the training mechanics of AI campaigns.
At the beginning, we launched several formats of advertising campaigns that were supposed to provide feedback. But then it became clear that we had no choice but to turn to two basic campaigns, to fix them, to fix the speed of the site. Here are the two basic ones. Why two basic ones?
One of them is Pmax. By the way, it worked quite well and was even cheaper for some periods. The rest is a regular product for manual clicks. Why did we launch it for manual bids? To see and feel the feed itself, because we had a certain type of pool cleaning product and not all search queries were suitable for us to target with Pmax. In order to evaluate the quality of the feed, whether it needs to be improved, and so on, what is the real cost per click, plus or minus.
Because if you look at it, we had a super cheap CPC and a super high CPC. We needed to understand why and how we could either really go to a super-hot auction with a higher CPC and get sales from there, or we still had a problem with the feed itself and the quality of all these clicks was quite low.
Because even with the help of scripts, the Google API will not show you all the queries generated by Pmax. So you shouldn’t hope that you can use a script to accurately, one hundred percent, determine which search queries prevailed or did not prevail in these Pmaxes. Because if you analyze your auction statistics and try to compare it with reality, you will see that it will be very different for the queries that Pmax will draw you.
That’s why we had this ordinary product. We didn’t need all of them. Of course, at different times we had higher and lower CPCs. If you understand correctly, the CPC was higher for the simple reason that we were trying to pump it up. Although here, it remained practically the same in order to initiate a significant part of the impressions.
In fact, we had sales conversions from only one product item, despite the fact that several of these products generated some kind of traffic.
Analyzing the value of cost per click
Did we overpay for CPC? In principle, there is a certain logic behind the overpriced bids, because we had a broken start-up period. Here, we did not fully understand whether we were collecting the right or wrong traffic, whether it was completely correct or not. Therefore, it is predictable. But even here, you can see that conversions appeared in a regular product.
This is another confirmation that the traffic was correct, selected correctly, segmented correctly. Even in the tops for impressions and queries that led to conversion.
If we had continued, if this period had not been interrupted, if it had been systematic and correct, we would have managed to lower the CPC quite significantly, as it is in Pmax. In Pmax, if we look at the purchase, it is still roughly equal to what competitors paid. But since we did not know whether we would generate the right traffic or not, because the product overlapped with other broader queries and there was seasonality during this period, this decision was reasoned and deliberate on the part of the specialist.
Could you have gotten more conversions?
Could we have made even more conversions? I think, a priori, yes. We could have really made even more sales. It was possible to make this traffic spike here and generate even more, grow to five to six conversions per day somewhere by the end of the third or fourth week of sourcing, if we had not broken this period.
How many calls and sales will I get by ordering contextual advertising from you?
Plus, in fact, we wasted some time on rolling out this whole story, thinking that we really had a problem with traffic, that some strong competitors had entered, that we hadn’t launched on time. After all, the client did launch later than the recommended time.
But it turned out that the problem was the very slow speed of the site, just very slow. We’re not talking about the seconds of the benchmark, but the fact that it took 8-10 seconds for the site to open. This is a very long time. With the high competition and the high level of user spoilage with this experience, this UX design, it was very difficult to say that anyone would have expected it. Unless we were the only store in the whole world, then maybe so, but that’s not what we had. Therefore, it is possible to reach five sales per day. Unfortunately, we didn’t make it in that time frame.
You can look here, for example, on the 16th. We’ve been tweaking it for another week, and you can see that the price is cheaper. By the way, the profitability is higher per envelope, despite the fact that the CPC, as you remember, has not fallen. This is an argument that sometimes there are even more conversions and they are cheaper when targeting in hot auctions at an even higher CPC than catching super-budget conversion clicks somewhere.
Pmax has become expensive because even with a fairly high conversion rate, the cost per click has skyrocketed, but for the mechanics of learning Pmaxes, this time period can be a variant of the norm. We are considering a week or two weeks. On a US scale, that’s not enough time to test the entire set of traffic, the types of auctions, which queries convert better, which queries have a higher conversion rate, which have a lower conversion rate. That is, to test the entire volume on a conditional $400 per week of this spin that we had here at the end of this period is very ridiculous. Considering the United States, this is a very large market. Even for the geo that we advertised to, it is very significant, large, and there is potential for growth, but not everything can be done so quickly, so instantly, so cool.
Two options for testing advertising campaigns
These are the two test options that we had for a small advertising budget, a start-up test. The client wanted to test whether there were any sales, how quickly or slowly they could scale, and so on. Here’s an example of small budgets. There were only a few states in advertising.
We had a very small set of products, up to five items in each feed. Is it possible to optimize this, to reduce the cost per conversion to even less, to 15? Yes, it is. It took some time to collect regular traffic and train the AI to make these ad campaigns cheaper. Because, as we remember, at the beginning, Pmax was even cheaper than a regular product. The regular product is now much cheaper because it is, in fact, cleaned up quite well in terms of traffic with some probable negative words. And, of course, its conversion rate was quite high. So, yes, it is possible to reduce it to a much smaller one.
How long does it take to collect analytics for AI?
The question is that you should always be prepared for the fact that if we enter the United States or, for example, Australia, you need to be patient for at least 60-90 days. Anything less makes no sense, because it will take too long to collect this analytics. Can it be collected faster? I am not sure that it can be faster, because the quality of the analytics we provide to the AI machine depends on the quality of the data. Because we remember that we are building our own AI agent here. That is, each account is actually unique in the use of the AI box. Of course, the product will be almost identical everywhere, and in the AI box, we conditionally generate our own AI agent that operates with its own database, storage location, and input/output information.
But what we want to do at the end is to fill it with the right information, i.e., the right analytics, showing what we want at the output. If you’ve ever written a really competent promo in the same GPT chat. I mean competent! Not just “Imagine you’re a super-awesome marketer and develop a strategy for me” – that’s too general. But the right one, with an emphasis on what to do, what not to do, what result to focus on, what kind of feedback should be provided, what resources to use, what database, and so on. If you’ve generated the right kind of promo, you can roughly estimate how difficult it is. Not only that, but nowadays the generation of promos on freelance exchanges can cost from $100 to $1,000 at least. And they can take up to 6 to 8 hours to create.
Imagine how complicated the story is in an advertising office now. And you could give it the maximum amount of data set, be patient, and then it would work great. No, in practice, no matter how Google sells it, their remote sales managers, I see the opposite situation. It is better when you gradually and systematically teach the AI inside your Google Ads to generate the right result. In most cases, in 8 cases out of 10, this is the best option for scaling, systematic, smooth, predictable, and psychologically comfortable in terms of costs. These micro variants of getting results are vivid examples. Can it be done cheaper? Yes, you can.
Is it possible to lower the cost per conversion?
We see that period to period… We even had a start date of November 16. We had a big break here. It really became cheaper, this price per conversion. On December 3, we had something very specific in the shot. Because AI is also trying to adapt to all formats, to all the nuances that exist. We tested the cheapest CPC, super cheap CPC, expensive CPC, average CPC, and this Pmax showed the most optimal scaling option.
Yes, it was super high here. That’s why we have one of the highest cost/conversion rates here, and it’s $28. But it can drop to $15 if this period is uninterrupted, if you systematically try to scale it, and not just rush in and overwhelm everyone.
Conclusions.
Everything is real, everything is practical, and that’s why I even love this Performance. Regardless of one’s attitude to it, AI tools will rule the day – this is the main trend. The dynamics of scaling will be carried out by them in the future. In 2025, it will be very difficult to generate large turnover on manual rates, even if you see that it has become much cheaper now. In fact, it will be very difficult to export only on manual rates for this entire period. Yes, they will work, but the main point of growth, scaling, and comfort will be provided by AIs.
This is a small, very strange discontinuous case describing why this happened. An iteration of very honest cases, not super-successful successes, but cases like this, of which there are actually a lot. It seems that publishing them would not be a super-successful case, but on the other hand, I tell you the nuances: what happened here, what influenced the result, show the dynamics, show how partial work was done. I show what is positive, what is negative, what are the rates. I even show that with a high CPC, you can get a much cheaper version of the cost per conversion.
As we saw in the test clicks, if we take these percents. I’m just showing you the conversions. Let’s see, it will probably be more convenient to see here, here is the cost conversion, you can see that there are no conversions here, the traffic is great. There are up to $30 here. And $15 is the desired price per conversion.
And the average, if we calculate for this period from January 16 to January 25 – this was the period of our sapporting, here you can see that even the average $33 is much cheaper than it was calculated in the first period.
The first period of time is the same as November, we spent money, we spent $26, we didn’t get any purchases, but there were some Add to carts, most likely you can see them now.
There were no purchases, there was Add to cart, Begin checkout, and that was it. But there was no purchase.
This is an example of not super successful cases, but honest ones.
