Reducing your CPA on Google Ads – a two-year case study

Google Ads Optimisation Case Study - Rykov Media

Running a Google Ads campaign can be tedious. It can certainly be very costly and simply unaffordable for young businesses. However, if you have a budget that you can set aside just for Google Ads, and you follow certain steps, you can eventually optimise your campaigns so you get the right ROI (Return On Investment) from the ads. It took us about two years to do so for one company. Since some of the information is confidential, we will use the name Zinga for the company, as well as (proportionately) made up budgets + we will say that the company offered an on-demand car washing service (the real company was actually offering a service rather than a product, which is quite an important detail).

Starting out

Creating the initial campaigns on Google Ads (or Google AdWords, as it was called at the time) was quite a routine task. We went through the usual stages:

  1. Looked up most researched terms in the industry.
  2. Looked up not such popular terms.
  3. And finally looked up very niche terms.
  4. Ran the above through Google Keyword Planner for historical data and to analyse bids.
  5. Created multiple campaigns with the above keywords.
  6. In total we had about 3 different campaigns, with 4-5 ad sets, each consisting of about 10-20 keywords.

Note: for all of the above, we started with simple search campaigns, that were optimised for website traffic.

The campaign then ran for about 3 weeks, during which time we observed what the ACTUAL traffic was for each keyword, CPC for each keyword and competition. (We will come back to the latter further down the line).

Adjusting our website

For anyone looking to optimise your conversion cost over time, remember that the key thing here is your service (or product) that you offer. No matter how good your keyword optimisation is, if the website is not functioning well, or if certain payment methods are not accepted, the client will not convert.

Our main mistake was not A/B testing every aspect of car washing/website pages. This could save us a lot of headaches in the future…

We had a 5-step process for anyone looking to book a service with us. From Google Analytics, we were able to determine some of the top screens in terms of the bounce rate – i.e. % of users that left that page without interacting with it. The higher the % – the more urgent the need to fix that page.

Something to remember – users will inevitably “fall off” at each stage of the process. Your goal is to determine WHY they did not proceed.

While we implemented changes, we also continued to pause some of the keywords that were spending our precious budget, but that were not/rarely converting. The conversion rate at the start was $50/conversion (made up number, but the changes are proportionate to what we experienced).

This was rather high, but the company’s management was determined to get the first 1000 customers as soon as possible, and for a time it was acquisition at all costs… until we explained that that’s not how things should be done.


While optimising our campaigns, we managed to narrow down the total number of campaigns to 2 main campaigns (about $500/day spend per campaign), where we had approximately 40 keywords per campaign. Out of those 80 keywords, 5-6 combinations were responsible for about 70% of conversions. Out of those 5-6 combinations, only one was unique to us (compared to our competition).

Tempering with the experiments

One major thing that also impacted our results and made predictions difficult was seasonality. By December we had a large influx of orders, and then for two months we suddenly hit rock-bottom. December to mid-January was a “dead” period. During this time we managed to release a new version of the website with brand-new features, which we hoped would result in higher conversion rate. The latter was difficult to estimate, as customers were coming back after a break, and to a large extent the influx of orders was due to that. At the end, we were able to estimate the impact of the last website update, only after tracking bounce rate for 3 months following the release.

Optimising for conversions was part of our strategy long-term, and so after 3-4 months of testing keywords and campaigns, we switched the campaigns’ goals to “maximise conversions”. Immediately we saw a decrease in clicks. The number of conversions on one of the main campaigns actually dropped, following the switch. As the result we kept:

a) One campaign as a “maximise  conversion”.

b) Second campaign at “maximise clicks”.

The average cost per conversion for the two campaigns dropped to about $40/conversion after 5 months (since we started).

5-7 months in

Running experiments can be scary, when you are constantly trying to minimise your spend AND increase number of orders at the same time. But it is crucial. After 5-7 months our number of conversions on AdWords was now also impacted by our social media activity. Around this time we began to run social media campaigns and giveaways more regularly on Facebook and Instagram. The attribution model that is set by default on AdWords is last click – i.e. it attributes the conversion to user’s last click. For example, if the user saw our ad on Facebook, and then clicked on our AdWords ad, Google will count this as a whole 1 conversion from its ad.

This is where most problems arise, however, when you try to optimise conversion cost, you have to find ways to calculate your ROI one way or another, regardless of the attribution model. Over the next 4-5 months, we managed to slowly but steadily attribute conversions/conversion split to social media, organic search and google ads more accurately. This coincided with another seasonality – July/August period (so 1 year after we started). This is where we got very excited, as the price per conversion dropped to $20/conversion. Our celebrations were premature.

Conversion price goes back up

The problem for any new business with seasonality, is the fact that YOU DON’T KNOW when it will strike. We were more or less able to determine specific months when orders might be up/down only after two years of being operational (and we are still learning). However, unfortunately, this can seriously impact your CPA. Following the July-August period, we saw our CPA’s jump back up to $35/conversion.

We experimented with changing the goal from maximising conversions to target CPA (i.e. only spend money on potential click, if it leads to a conversion < $25/conversion), however this dropped our traffic by over 30%, and after 1 week, we decided to revert. 1 year and 3 months in, we now changed both campaigns to maximising conversions.

Around this month, google began to actively promote it’s responsive search ads and dynamic search ads. After another two months-worth of experiments, we found that dynamic campaigns actually had a conversion cost of just $15/conversion!

Armed with the above results, we continued running “Maximise conversions” search and dynamic ads, and by month 24, our conversion cost has been more or less stable at $23/conversion.

More notes

The one and complete failure for us have been display ads. While we tried to treat it more like a lead-generation tool, the cost was not worth it at the end, so we decided to stick to what was working instead.

Auction insights can also prove useful when assessing your CPA. During some months we noticed that more competitors began to bid on keywords that we would normally use, meaning that the cost would go up. The only real way to fight this ongoing battle is to continue finding new niche keywords that might bring in results. While we experimented with more local searches (for example targeting “car wash in South Brighton”), we found that sticking to the main 10-20 keywords still brought in better results.

To sum up…

We found that experimenting with different ad types works well during the optimisation period, but it can mess with your results, if it coincides with particular seasons/website changes. A Target CPA is not always the ultimate goal, as sometimes Google Algorithm “works in mysterious ways” and a simple “maximise clicks” goal could work out better than everything else.

One thought on “Reducing your CPA on Google Ads – a two-year case study

Comments are closed.