How to do Conversion Rate Optimization: The 6 Steps of CRO for Online Businesses
In the sequel to the previous article, we’ll focus on how to do Conversion Rate Optimization in more detail. But before we get to the interesting part, here is a recap of the previous article where we had an introduction to Conversion Rate Optimization.
What Conversion Rate Optimization is,
The types of businesses it can be applied to
How to know if your business needs Conversion Rate Optimization
How to get started with Conversion Rate Optimization
In case you missed this introduction, catch up on this quick read.
The 6 step CRO process
In the previous article, we mentioned six steps that are involved in a comprehensive CRO process, these are:
Let’s go through each step in more detail to learn Conversion Rate Optimization.
1. Analytics Configurations
This step applies to online businesses that are yet to set up their analytics tracking. Because Google Analytics is widely used, I’ll use it for demonstration in this article. However, other platforms such as Segment.io, HEAP Analytics, or Mixpanel can also be used for data collection.
In this step, the most important thing is to create a tracking plan that aligns with the business key performance indicators.
For an ecommerce store, these include but not limited to:
Average Order Size
Number of Transactions
Shopping Cart Abandonment
New Customer Orders vs. Returning Customer Orders
Cost Of Goods Sold (COGS)
Product Affinity (Which products are bought together)
Product Relationship (Which products are viewed together)
This information helps in cross-selling products
Customer Lifetime Value
Revenue Per Visitor
Customer Acquisition Cost
Once the tracking plan is set up, then analytics implementation can be done.
To get all the above KPIs tracked, more than one analytics tool may be required.
For this step, the most common tools to use are:
Google Tag Manager
With Tag Manager, you can also integrate data from other platforms like marketing-based platforms.
2. Analytics Audit
This step applies to users who have already configured their analytics and are collecting data.
For those who start the process at step 1 (analytics configuration), it is best if they also audit their setup as they implement the configurations. In this case, you cover both steps 1 and 2 of the Conversion rate Optimization process.
You might be wondering why this is important while data is already being collected. This is because we need to make sure that we are collecting reliable data before we start optimizing. If we have insufficient or unreliable data as a result of poor analytics configurations or, in some cases, missing data, we will base our decisions on the wrong data and we also won’t be able to track progress effectively.
Since this is an essential step in getting everything we need for better decision making, we’ll focus on how we can audit the Google Analytics configuration and also some of the issues that can be resolved as a result.
A few cases where Analytics Audit is required
Example 1 - Duplicate transactions
This refers to cases where a single transaction is tracked more than once. The challenge with this is that it inflates the revenue and number of transactions, which gives a false impression on the performance of the business. One way this can occur is if the page users are redirected to after purchase, which also has the tracking code, loads the code more than once. It could also be caused by another reason entirely. Therefore, investigation to determine why this is happening is important.
Example 2 - Incorrect shopping funnel data
We have cases where some steps that appear later in the shopping funnel have more sessions than those that appear early in the shopping funnel. For instance, in the screenshot below, we have more checkout sessions than the add-to-cart sessions. This is strange because usually, this doesn’t happen.
On further investigation, we discovered that at some point, the add-to-cart event stopped working, causing a drop in the number of add-to-cart sessions for the selected duration.
An audit process helped identify when this issue occurred and further investigation was done to resolve the issue.
It is advisable to add an annotation in Google Analytics to explain cases where data was compromised or when a UX change is done during the Conversion Rate Optimization process. This helps explain the changes in data.
Example 3 - Bot traffic
Google Analytics has a feature to filter against known bot traffic. When this feature is not enabled, we have cases where data consists of bot traffic, which inflates the number of sessions, affects bounce rate/exit rate, and conversion rate. Due to this, we’ll never clearly know the website’s performance where these metrics are concerned.
Example 4 - Filters for internal traffic
For all online businesses, a number of people are involved in maintaining the website and these people frequently access the website to resolve things or check that everything is alright. If their data is not filtered, we’ll have cases where we have sessions that are not from real users, and this also affects the real performance of the website. Google Analytics has a filter setting option to filter users by IP address. Also, in cases where the online business deals with clients from only one country, then a filter can be created for users from only that country since they are the real target audience for the business
There are many more cases and an audit document guide is required so that everything is sorted. Here is a document you can use to guide you through the audit process.
3. Website Audit
Now that we have clean and reliable data, let’s focus on the 3rd step of the Conversion Rate Optimization process.
In this step, we look at two things:
The low hanging fruit
Hypothesis for A/B testing
Low hanging fruit
The low-hanging fruit is about the conversion rate optimization guidelines or recommendations for the different online businesses. For an ecommerce website, we have things like a sticky menu, consistency in colors for all the call-to-action buttons across the website, floating add-to-cart buttons, a search module among other things.
These are among the things that are a standard for ecommerce websites and they can be implemented even without the need for an A/B test. In this case, we go through the entire ecommerce website to figure out if all the recommended standards have been implemented.
Hypothesis for A/B testing
A hypothesis is a prediction you can make before running a test. It states the kind of change you want to make, why you want to make it, and its expected result.
In the website audit step, we also assess potential funnel leaks, what the user experience is, a product story, and how it is being told.
Here is an example of a hypothesis:
The mobile visitors have a greater scrolling distance and view little information above the fold. It could be that mobile users are not accessing the most important information as easily as desktop users.
We could have the most important elements above the fold for the product details page, these include:
Here is how you can visualize the solution if you are not using wireframes to create the variant design.
The expected result is an increase in the number of mobile users who add items to the cart, which results in more users moving further down the funnel and finally, increasing revenue.
With this hypothesis, you can then use data to get more information to support it.
More examples like this can be hypothesized at this step, then, we get to step 4 (Data Analysis) here we get more data to support our hypothesis.
4. Data Analysis
Now that we have several hypotheses to test, we can now get to step 4 of the Conversion Rate Optimization process which includes but not limited to:
Getting supporting data for the A/B testing hypothesis created at step 3
Getting data to create more hypothesis for A/B tests
Speed test analysis & device performance analysis
Shopping funnel analysis among other things
At this step, the focus is on understanding the audience and their customer journey.
Let’s look at how data analysis can help support our hypothesis created at the website audit step.
Looking at the shopping funnel data, we find that the number of mobile users is more than that of desktop users. Besides that, we see the percentage of mobile users who add items to the cart is lower than that of desktop device users.
We also check the product details page metrics on mobile devices and find that they have a higher bounce rate, lower time spent on the page, and a higher exit rate compared with desktop devices.
With this information, we can focus on the A/B testing, while monitoring these metrics for improvement.
We also focus on which pages are loading slowly and if there are devices that are performing more poorly than others like in the case of the mobile devices above.
The quality of traffic is also analyzed at this step and here are some Conversion Rate Optimization examples where traffic is concerned.
We have cases where a traffic source hardly contributes to revenue in this case, the business is better off re-allocating the marketing costs to another source.
In other cases, we find that the channel or traffic source functions best as the first point of interaction along the shopping funnel than as the last point. In this case, the marketing department can decide to allocate more resources to the channel to bring in more people.
Another example is identifying which traffic sources work best with different customer segments. This helps in target marketing, which saves on marketing resources (e.g. reducing customer acquisition costs) and also gets quality traffic to the website. This helps in increasing the conversion rate and also the number of transactions and revenue.
This is another whole topic on its own, I won’t, therefore, dwell so much on this in this article. However, here is a highlight and example.
For the most effective data analysis, segmentation is key. For example, you might find that there is a challenge with one part of the shopping funnel, however, this challenge does not apply to everyone and neither does the solution.
For example, we discovered that for one of our clients, there was a high drop-off rate at the checkout step, however, when we segmented the data, we realized that the high drop-off rate was only for the clients who were not from the country where the online shop was selling. These clients could be abandoning the cart because they don’t know what the cost of shipping is likely to be or even the import duty.
In this case, we proposed a solution that could favor the international clients.
We proposed a plugin that shows shipping estimates and import duty based on the customer’s location
We proposed to them to consider free shipping
If offering free shipping is not something they were willing to consider, then, they could have a free shipping incentive that encourages users to spend more in order to qualify for free shipping e.g. spend $90 to qualify for free shipping.
Segmentation gets you more accurate data and also helps you come up with solutions that are cost-effective.
5. Behavior Analysis
Now that we have more data we can use for the optimization process, let’s see how we can learn how users are interacting with the website.
In the Behavior analysis step, we focus on:
Examples of tools that can be used at this step are:
This step gets us more interesting data like how users are navigating the website, which elements they click on, how much further down the page they scroll, and if there are rage clicks (users continuously clicking on an element expecting a specific reaction that does not happen) or even u-turns (users keep going back and forth trying to get information).
Through user testing and surveys, we get to understand things like, if users are having a challenge with the shopping experience; if they can search and get the information they need about a product; if they understand the product or service being sold, and many more things that can help improve on the customer experience.
Heat maps and video recordings are also used alongside A/B experiments to record user behavior on the new changes or updates made on the website.
In the example below, we see from the heat maps that users hardly get to where the product descriptions are. This means that despite the importance of this section, a very small percentage of users get to read it. This knowledge can be applied to better present the
information in a way that the page is shorter and only the most important information is featured in a readable format, such as using bullet points to highlight key factors instead of paragraphs.
6. A/B Testing
A/B testing is an experimentation process, where we show two or more versions of a webpage to different segments of users to determine which version has the best returns.
The most common types of A/B tests are:
Split URL testing
Multivariate testing (MVT)
This type of testing happens when two versions of a page element or web page are shown to different segments of users to determine which one performs better.
This refers to when there is a new version of the original page existing in a different URL and this new variation is tested against the original to determine which version performs better. The original page could be having the URL www.mywebsite.com/a and the other one could have the URL www.mywebsite.com/b.
This refers to a test where a single element is tested across multiple pages. A good example is when testing a call-to-action button across multiple pages of the website to find out which version performs best.
In this type of testing, variations of multiple elements on a page are tested simultaneously to determine which combination of elements performs best.
There are many tools that can be used for A/B testing. Among them are:
Visual Web Optimizer
To run a successful A/B test, one of the main factors to consider is traffic to the website. In the Conversion Rate Optimization guidelines, the recommended number of goals that are likely to get you a clear winner is over 300 conversions per variation.
It is also best to run the tests at the start of a business cycle and let the test run for the best duration.
When calculating the test duration, the number of visitors, the amount of improvement in conversion rate you are looking to achieve, the current conversion rate, and if all the visitors or just a percentage will be included in the test need to be considered.
Statistical significance determines how likely it is that the difference between the control and the variation is not due to random chance or an error. It measures the confidence level in the experiment. For instance, if you run an experiment that gets to a 97% significance level, you can be 97% confident that the differences between control and the variation are real.
You must let the test run through its entire duration so that it reaches its statistical significance and this is because there is always something to learn from a test. You could find that during the first half of the business cycle, the variant performs best and during the last half, the control does better than the variant. If the experiment was stopped before it reached its statistical significance, then, the variant would be deployed based on insufficient data and this would be the wrong decision.
Every experiment has something to learn about, whether you get a winning variation or if control wins. This is why you should never dismiss an experiment where control wins.
Always run heat maps and video recordings during experiments. These give you insights into how users responded to each variant, which can help you apply a concept in other experiments or know why an experiment failed.
Examples of experiment results
Experiment with a winning variation
Experiment with no winning variant
What to do if there is no winning variant
In cases where there is no winning variation, the best thing to do is to analyze the results of the experiment, then, with the information in mind, go back to the data and behavior analysis steps to get more information for a new hypothesis (or improvement of the existing hypothesis).
This also applies to cases where the declared winner is the control and not the variant.
If your experiment declares a winner (that is not the control), go ahead and deploy the winner, then, iterate the Conversion Rate Optimization steps for another hypothesis. Conversion Rate Optimization is a continuous process so that a business can keep finding the best ways to serve its clients.
In learning how to do Conversion Rate Optimization, all steps are essential and need to be done to achieve the best results. The first two steps are necessary because we make sure that the data collected is credible and can be relied upon to make informed decisions.
Conversion Rate Optimization is a continuous cycle. Even if we run one experiment and get a winning variant that boosts the conversion rate, there is always more room to keep optimizing the website.
Lastly, there is always something to learn from each experiment, including the failed experiments.