The secret of high ROI advertising—data analysis and optimization

To become an excellent marketing operator/growth manager, it is not enough if you only know how to place advertisements but do not know how to improve the ROI of advertisements by analyzing data. Data analysis and optimization is not simply looking at the background data, but a core skill that runs through marketing thinking and practice - seeing the essence through the phenomenon.

Therefore, in this article, the Captain will teach you a complete set of data processes, including a series of models and theoretical foundations, as well as work steps, and then optimize advertising based on the obtained analysis results. This is definitely the essence of the essence, and everyone should read it carefully.

Table of contents

1 Three essential data analysis models for high ROI

1.1 Advertising Process Model

1.2 User Behavior Model

1.3 Advertising Conversion Funnel Model

2. The difference between data analysis dimensions and data indicators

2.1 Concepts of Analysis Dimensions and Data Indicators

2.2 Common analysis dimensions and data indicators

2.3 Data Analysis Principles

3 AB testing determines the key to optimization success or failure

3.1 AB Testing Variables

3.2 Steps of AB Testing Process

3.3 AB testing process: developing a testing plan

3.4 AB testing process: start testing as planned

4 How to create data reports

4.1 Daily and Weekly Reports

4.2 How to prepare daily data reports?

4.3 How to write a weekly report?

5 Advertising effect diagnosis and optimization (key points)

5.1 What to look at in data analysis

5.2 Advertising effectiveness diagnosis

5.3 High CPA

5.4 Low traffic

5.5 Low advertising ROI

6 All in One Practical Examples

7 Conclusion

1. Three essential data analysis models for high ROI

1.1 Advertising Process Model

From the perspective of advertising optimizers, we can divide the advertising optimization process into four stages:

Phase 1: Setting Goals

• This step is very important. There are more than ten optional targets, including brand, traffic, interaction rate, potential customer development, conversion volume, app installation, etc. In addition, the first stage also requires you to analyze the product and target population, that is, product research and user profiling.

Phase 2: Testing Phase

• First, you need to plan the account structure based on the test dimensions of population positioning (gender, age, language, etc.) and prepare the advertising materials to be delivered; then you must use the AB testing method. For example, if the advertising material is 10 videos and you want to know which customers like to click and have a good conversion effect, you must conduct AB testing.

Phase 3: Data Analysis

• Judging by the delivery data, for those with poor results, we need to dig deeper, find out the root causes, and propose new optimization plans for further testing.

Phase 4: Scaling up

• At this point, we have basically determined which groups and materials are the best, so we need to maximize the delivery effect and scale it up horizontally and vertically.

1.2 User Behavior Model

In addition to the advertising optimization process model from the optimizer's perspective mentioned above, we also need to learn the model from the user's perspective, which uses materials to influence user behavior in four steps: attracting attention, stimulating desire, gaining trust, and guiding action.

The user behavior decision model, AIDAS principle, is used here to explain the different effects of advertising on consumers. It includes the following five stages:

① Attention

Whether or not to attract user attention often depends on a matter of seconds, so we usually use a large headline and a large picture to attract visitors' attention. Usually the quality of the title and picture directly affects the click-through rate.

② Interest

Tell your customers what your pain points are and what benefits and advantages the product can bring to you.

③ Desire (inducing purchasing desire)

What can arouse consumers' desire to buy is to further tell consumers why they need the product and what serious consequences will occur if the problem is not solved, and let consumers know that the product can solve their problems well and what psychological satisfaction it can bring after solving the problem, thus satisfying your desires.

④ Action (promote purchasing behavior)

When a consumer plans to buy a product, you have to tell him to buy it now, including how to buy it and what the purchase process is. Provide clear and reliable CTA (Call on action) to simplify the registration, shopping process and website interaction as much as possible.

⑤ Satisfaction

While satisfaction can’t directly increase conversions, it’s crucial to your overall business. The cost of acquiring a new user is 2 to 6 times the cost of maintaining an old user.

So while gaining a good reputation, you can get a user to continue to repurchase your products, and the user will recommend your products to his friends.

1.3 Advertising Conversion Funnel Model

Through this advertising conversion funnel model, we divide the conversion funnel into four layers: display, click, browse, and conversion. This is easy to understand. Just look at the picture below:

Everyone should fully understand the above three models, because they are the basis for all data analysis and advertising optimization. Next, let’s talk about the dimensions and indicators of data analysis.

2. The difference between data analysis dimensions and data indicators

2.1 Concepts of Analysis Dimensions and Data Indicators

Many new students in the industry tend to confuse the two concepts of analysis dimensions and data indicators. For example, is cost per click a dimension or an indicator?

In fact, we must first think clearly about the analysis dimension, and then look at its corresponding data indicators. When we do data analysis, we first decide which dimension of data to look at, determine the dimension, and then look at the corresponding data. Only in this way will our data analysis be meaningful.

2.2 Common analysis dimensions and data indicators

The following table can better help you understand the relationship between the analysis dimensions of this data and the data indicators.

For example, for different population positioning, age and gender are different analysis dimensions, then some data generated by different delivery strategies are basic indicators, such as display volume, click volume, conversion volume, consumption number, activation number, retention number, etc.

As shown in the example above, our analysis dimension is based on region, and exposure, clicks, spending, etc. are our data indicators.

From the data in the table, we can see that the conversion rate in Thailand is the highest and the order cost is the lowest. It can be seen that from the current advertising input-output ratio, Thailand is the best. By comparison, we can immediately generate at least two tasks to do:

• Analyze why Thailand performs better than other countries and use this as a reference to optimize other countries;
• Increase the budget for Thailand and observe the data changes to obtain more revenue.

2.3 Data Analysis Principles

I believe everyone has a question: there are so many data indicators, which ones should we pay attention to? So here, I want to talk about the principle of data analysis: the value of composite indicators is greater than that of basic indicators

Because composite indicators reflect ROI more. For example, if your basic indicators are very high and your costs are also high, but there are very few clicks and conversions, then your ROI is actually very low.

However, if you only look at one basic indicator, you cannot see the result. You need to look at the CPC cost per click and the CPA cost per action to see whether the input-output ratio meets your expectations, and then decide whether to change the strategy to continue optimization, increase investment, or shut down advertising.

Composite indicators are calculated from basic indicators.

For example, CTR, CPA, CPI, CPS, and CPT.

Why are composite indicators more significant than basic indicators? Let’s take a simple example of click volume:

Company A: 1,000 clicks, cost 1,000 yuan

Company B: 10,000 clicks, cost $100,000

Between Company A and Company B, which one has better optimization results?

CPC (cost per click) = total cost / number of clicks

Company A's click cost: 1 yuan/time

Company B’s click cost: 10 yuan/time

Judging from the click results, it is obvious that Company A is better.

3. AB testing is the key to success or failure of optimization

To improve ROI, AB testing is essential.

A/B testing, also known as split testing, is a quick way to find the best strategy.

We can use this AB test to find out which ad headlines, text, images, videos, calls to action, or combinations are best for the target audience.

In addition, we can also test the performance of different target audiences in different ad placements and positions to understand who your perfect audience is and where to place the ad for better results.

3.1 Variables of AB Testing

When we do AB testing, the most important thing is to test multiple variables, including audience targeting, materials, settings, landing pages, etc.

3.2 Steps of AB Testing Process

The AB testing process is mainly divided into five steps, as follows:

① Set project goals, which are also the goals of AB testing;

② Develop a test plan: determine the version to be implemented and the diversion ratio of each online test version;

③ Test online according to the plan;

④ Collect experimental data and make judgments on effectiveness and effect;

⑤ Determine to release a new version based on the test results, adjust the diversion ratio and continue testing, or continue to optimize the iterative plan and re-test if the test effect is not achieved.

From the definition of AB testing, we can see that AB testing emphasizes testing on users with similar attributes in the same time dimension.

The uniformity of time can effectively avoid the influence of factors such as time and season, while the similarity of attributes can minimize the impact of other factors such as region, gender, age, etc. on the effect statistics.

3.3 AB testing process: developing a testing plan

First, you need to set your test goal. For example, when you launch on Facebook or TikTok, the target you set on the platform is similar, but the one set on the platform is the ultimate goal, such as conversion to purchase, and your test goal can be a part of the entire funnel, such as views (impressions), the cost of acquiring leads (action cost), the proportion of items placed in shopping carts (action conversion), etc.

If your test has multiple dimensions, such as CPC, CTR, and CVR, which metric should you prioritize? On the one hand, you need to look at your advertising goals, such as branding or conversion purchases.

If it is a conversion purchase, CPM and CTR are not the best options. In most cases, CPC can be given priority and is relatively reliable because it reduces the amount of energy expansion.

The fewer ad variables you have, the faster you can get relevant test results, and the individual ad variables make it easier to track and evaluate results.

The most important task in testing is to plan the advertising account structure, because planning the account structure is basically equivalent to planning the dimensions of the test.

Test scenario 1: Single ad group

With separate ad groups, all of your variables revolve around the same ad group. The benefit of this structure is that your same target customer won’t see you testing multiple creatives (multiple ad groups targeting the same type of people). But one big downside to this structure is that Facebook will automatically optimize your ads and you won’t get relevant results.

The advantage of the optimization is that the cost will be relatively lower, so if you are sensitive to the cost at the beginning, you can use this method. But some ads may receive very little traffic.

Test plan 2: Multiple ad groups (same) single variable comparison (rotation)

Multiple single-variable ad groups means that each different creative is in a separate ad group. If you place each creative variation in a separate ad set, Facebook will treat each ad set as an independent entity and will not automatically optimize for small results, so this is the best way to test. Rotation makes it easier to determine which material is better, and is very suitable for testing new materials.

Test plan 3: Multiple ad groups (different) single variable comparison

When there are multiple AB tests to choose from, you should prioritize the better AB tests, which can save a lot of time.

If you want to pull out all the potential influencing factors and do a test, suppose at the Ad Set level, you want to test 5 different target audiences, and at the Ad level, you have 5 different ad images, 5 different headlines, and 5 different ad copies. Then, if you only target one target audience, 5X5X5=125 ads will be presented, and you need to manage 5X125=625 ads.

If we add all possible combinations of age, gender, region, etc., the number of ads will reach 5,000 (40 Ad sets X 125 Ads). However, the more creativity the better, because your budget is limited and you need to find the factor that has the best impact on ROI (CTR or CVR). So, we need to test from the largest to the smallest impact.

3.4 AB testing process: start testing as planned

Although you can only change one factor per test, if you have enough budget and audience, you can test several AB tests at the same time. The factors of an AB test should be varied reasonably and not too many. For example, you cannot use 20 different pictures in a test. Generally, 2 to 5 are enough, and generally no more than 10 at most.

Difference test from big to small

How to test materials? We can do a difference test from large to small, testing 3-5 variables with relatively large differences, such as testing three pictures with very different styles. Then, unify the styles of the pictures that perform better above and compare different design points, such as background color, left and right positioning, whether there are icons or CTA, etc.

In addition, it is recommended to have sufficient test volume, that is, each AB test element should have more than 100 clicks or conversions, and then compare them. It would be better if there can be 300 to 500. Another question is, how long does it take to compare the results of AB testing?

The answer is to start with 24 hours and compare for 2 to 3 days , but there is no absolute answer and it depends on your budget cost requirements.

There are still many junior optimizers who don’t know how many days after AB testing it will take to verify the results and draw conclusions, 3 days, 5 days, or 2 weeks? Or are the two test results very close, such as 0.312% and 0.299%? In cases like this, it's best not to jump to conclusions. For example, in the figure below, click 27 on the left and 28 on the right. It is obvious that the test volume is not enough. If you add 2 zeros on each side, the difference will be obvious.

Testing more subtle variables is a common mistake many experienced people make when doing AB testing, such as changing only one word or line of text in the slogan, or a less obvious part of the picture.

Finally, we need to determine the optimization plan based on the AB test results.

4. How to create data reports

Producing daily and weekly data reports is one of the important daily tasks of extension personnel. Next, let’s talk about how to make your daily and weekly data reports.

4.1 Daily and Weekly Reports

Daily report: simple and clear, mainly used for daily data monitoring, with the focus on discovering problems as soon as possible.

Weekly report: tends to be an analytical report, the purpose is to analyze the causes and propose solutions.

4.2 How to prepare daily data reports?

What parts should a complete daily report include?

• Account Structure Description
• Data Source
• Perspective Segment Analysis
• Summary of key data
• Daily summary and next steps

How to generate a data source?

First, we enter the Facebook advertising report background from the advertising management tool, check the options we need, first click "Segment Selection", and then click "Indicator Selection".

It is important to note here that Facebook’s success rate is not conversion rate, but conversion divided by impressions, and conversion rate is conversion divided by clicks. Finally, we click Export in the upper right corner to export the report and select the xlsx format file. It is recommended not to select CSV because CSV will display errors for the last digits of some long strings of numbers.

After exporting the above table from the Facebook backend, we need to process the data. For example, when designing the name of the advertising group, we name it according to gender, age, hobbies, material type, time, placement, budget, etc. We need to separate them into columns to facilitate the pivot table later. For specific operations, please see the instructions in the figure below. I will not expand here.

Split fields by columns

The sorted data is shown in the figure below:

By splitting the fields into columns and analyzing the data based on the daily pivot table, we can drag and compare the test results and advertising performance of various dimensions through the pivot table.

Daily summary and plan

4.3 How to write a weekly report?

Everyone has the same habits for daily and weekly reports, which fall into the category of personal management and team management. I will share several daily and weekly reports that I used to report when leading a team for your reference.

5. Advertising effect diagnosis and optimization (key point)

5.1 What to look at in data analysis

First of all, you need to understand what data analysis is looking at? In fact, when you were making AB testing plans, you already needed to have awareness of data analysis. The setting of the test variables is actually to pave the way for the subsequent data analysis.

The tasks of data analysis can be summarized into the following three points:

1. Finding data anomalies in trend analysis

2 Comparative analysis to define the scope of the problem

3. Multi-dimensional segmentation to discover the causes behind the problem

Before optimizing, you need to plan your delivery process and adjustment strategy:

We divide ads with different performances into the following four quadrants.

For the optimization of advertisements with different performances, it is easier to jump one quadrant, but two are more difficult. When doing data analysis, we need to use four quadrants, as shown below.

5.2 Advertising effectiveness diagnosis

The next part is what everyone is most concerned about: What should we do if we find that the advertising effect is not good after analyzing a series of dimensions and indicators?

In order to give everyone a clearer understanding, I have compiled an advertising diagnostic ideas diagram. There are generally three reasons why advertising is not effective: high cost, small display volume and low input-output ratio .

The detailed disassembly can be seen in the following three situations.

5.3 High CPA

First, let's look at the high cost per click. We can break it down into high cost per click or low CTR:

High traffic cost

1 If the traffic cost is high, we can break it down into the intensified competitive environment and the high bids. If the competitive environment is intensified, then we can ask whether the ECPM value of the media is rising overall. Or, for example, if it is a holiday now, then everyone is scrambling for traffic, so the ecpm value must be rising, and the traffic cost will be high during holidays.

2 Then the second possibility is that our current bid is too high. At this time, we need to test it to see whether our test bid is too high, resulting in excessively high traffic costs.

Low click-through rate (CTR)

Then the second reason is that our click-through rate is relatively low, and the low click-through rate can be divided into three reasons:

1 The first is that our creative appeal is relatively low, and the entrepreneurial appeal is relatively low. It may be because our targeted population is too broad, which may attract many non-target audiences to see our ads, so no one clicks in the end.

2 Then the second type is our current material. In fact, no one attracts our current target audience to click on our ads;

3 Another reason may be that our current OCPM model is not stable. Why does this happen? The first possibility is that our conversion tracking settings, such as pixel settings or target settings, are not in place, so the channel media may not be able to see the real conversion data, so its traffic model is wrong. The second is that there is not enough conversion data now, because the OCPM machine needs a certain amount of time and clicks to learn.

5.4 Low traffic

Now let’s take a look at the second reason, which is that our display or conversion volume is small. We can break it down into two reasons.

The first is that we have insufficient budget, which may lead to premature decommissioning, or our budget allocation is not reasonable.

The second one is insufficient consumption. Insufficient consumption means first of all a small amount of display. The small amount of display means that our ECPM value is relatively low.

A relatively low ECPM value actually means that the CPC, CTR or CVR is too low, so we need to see which specific link has a problem. For example, if the CTR click-through rate is relatively low, then we go back to the high-cost graph just now and see how to solve the problem of low click-through rate.

5.5 Low advertising ROI

The third point is that our ROI input-output ratio is relatively low. If the input-output ratio is relatively low, there may be four main possibilities.

For example, if our targeted audience is too broad and attracts many non-target audiences, then users will naturally not complete the conversion because they don’t want to buy the product. At this time, we can try to gradually narrow our targeting range to test whether the ROI can be improved.

The second type of material actually does not attract the target audience. This requires analyzing user needs, and you can try iterating the material.

The third point is that our landing page cannot convince the target users. This is actually a problem with our landing page. At this time, we can iterate our landing page.

Finally, it is possible that our sales follow-up is not timely. If this is the reason, we need to communicate in depth with our business colleagues.

6. All in One Actual Case

Next, I will take a specific case to demonstrate the whole process to you (remember to draw inferences from one example):

First of all, our goal is to reduce the installation cost CPI/CPA of a certain game app

Goal: Optimize to less than 7 blocks

Generally speaking, we will test for about 3 days, hoping to optimize the ads to a single result of less than 7 yuan.

Now, let’s take a case study and analyze the results over three days for optimization.

By clicking on this ad group, we can see the trend of the past three days. From the above picture, we can see that the CPA is gradually decreasing, and the trend effect is getting better, which proves that there is room for optimization. Then we can do segmentation and conduct AB testing on age and material splitting.

Test plan

Then let's first develop a variable test plan, paying attention to distinguishing the test priorities

Demographic Targeting Variables

For example, we can divide the target population into three age groups: 27 to 29 years old, 30 to 32 years old, and 33 to 35 years old, and divide them into 6 groups based on gender.

Adjust based on the performance of the variable

For example, adjust the optimization direction according to the performance trend of the following material variables 

If we see that costs are gradually increasing, we will analyze the reasons according to the diagnostic decision tree mentioned above.

When you enter the ad group and find that the cost per result is gradually decreasing, this proves that the ad group is currently performing well and you can consider increasing the budget and expanding the volume.

Comparing the layouts, we found the effects of different material types

At this time, we can communicate with the designer, apply for budget and manpower to make more videos.

After further analysis, it was concluded that videos featuring female anchors performed best, followed by videos featuring new characters. The choice of female anchor was made because the audience is mostly otakus, and the new character can also attract players who like similar games.

After further comparison of different groups of people, we found that boys between the ages of 27 and 35 had the best results, and boys generally performed better than girls. Combined with the characteristics of the game (otaku games, you know), we feel that our test results are logically consistent.

Analysis and optimization by time period

In addition to daily optimization, we can also analyze data over a period of time during periodic reviews.

In addition to being used for self-review and summary, it is usually also used when we report to our boss or the client, for example, when we want to ask our boss for more resources (budget, staffing, designer resources, product optimization resources).

Let’s go back to this case:

If you extend the time and see the advertising performance during the peaks and troughs above, you will definitely want to know what caused them.

After analysis, we draw the following conclusions:

6

Summarize

Finally, let me give you a summary. You can look at the pictures directly, but the captain reminds everyone that because products and businesses vary greatly, these are just reference suggestions. Don’t copy them blindly, and be sure to learn from them.

The content of the article and the PPT have unknowingly exceeded 10,000 words. In fact, there is much more to share. Any one of the PPTs can be talked about for at least an hour, and if they are all expanded, they can be written into a book. I have written down the whole process here as much as possible, I hope it will be helpful to everyone.