What are the four categories of data analysis? The role and use of "descriptive," "diagnostic," "predictive," and "prescriptive" analytics to answer marketers' questions

As a result of the term "data-driven" taking on a life of its own, we often see workplaces where the goal is to look at numbers. However, simply looking at data does not create any value beyond understanding the current situation. What is really needed is to analyze the data and determine the next step.

Furthermore, even when the term "analysis" is used, there are different interpretations and meanings depending on the person, such as some people seeing analysis as finding the cause and others seeing analysis as predicting the future.

Given these differences in interpretation, a common framework is necessary for effective analysis—that is, analysis that leads to decision-making or problem-solving. The four analytical approaches at the core of this are "description, diagnosis, prediction, and prescription." This article explains how marketers should understand the differences between these four approaches and how to use them. By clarifying the purpose of each analysis and the insights it provides, more accurate data-based decision-making becomes possible.

Why is it important to know the types of data analysis?

Decisions are determined by the quality of the questions

When it comes to decision-making, the difference between good and bad marketers isn't their ability to read data, but their ability to determine "what questions to ask," in other words, the quality of the questions they ask.

When sales fall, the analytical methods required and the insights gained are completely different between finding out "by what percentage" and "why sales fell," "which segments are experiencing the decline," and "what factors influenced it." Furthermore, if you also want to consider "what will happen next month?" and "what steps can be taken to recover," a different approach will be required.

Knowing "what kind of analysis should be done?" will raise the level of data utilization.

Many companies often simply produce and look at the numbers without considering the type of data analysis they are conducting, but no matter how detailed the analysis is, if it is not suited to the purpose, the desired answer will not be obtained.

By being aware of the four types of data analysis, you can go beyond a vague desire to "see data" and ask a specific question: "What do I want to know?", and organize your approach to that question. For example, do you need a descriptive question like "I want to know this week's conversion rate," or a diagnostic question like "Why was last month's campaign successful?" This question will help you organize the approach you should take, including the data you need to look at and the appropriate analytical method.

Once the question is clear, you can easily choose the appropriate data collection method and analysis technique, which will improve the level of data utilization and lead to more accurate decision-making.

Let's take a look at what analytical methods are appropriate for the various questions marketers face. Here, we'll divide data analysis into four categories: descriptive analysis, diagnostic analysis, predictive analysis, and prescriptive analysis, and explain the role and application of each.

Descriptive analytics: Analysis that captures "what happened?"

Descriptive analytics is an analysis that uses data to clarify what has happened or what the current situation is. All decision-making begins with understanding the facts.

Usage examples

It is used to express the current situation numerically and visualize trends and patterns, such as sales trend graphs, shipment numbers by region, monthly retention rates, website access analysis, etc. Checking the number of sessions and bounce rate in Google Analytics is also a typical example of descriptive analysis.

Don't just say "I see," but get into the habit of asking questions

A common problem with descriptive analysis is simply checking the numbers and thinking, "Oh, so that's how it is." However, unless you attach meaning to the facts the data shows, you won't be able to fully utilize the value of the data. For example, if you find that the number of new customers acquired is 80% compared to the previous month, it's important not to dismiss this as a "decline" but to ask the next question: "Is this due to seasonal factors, competitive factors, or the impact of our own company's initiatives?"

Even when looking at numbers in weekly reports or dashboards, it's important to ask yourself, "What should I investigate next?" and "What hypotheses can I come up with?" By going beyond mere description and linking it to questions, your analysis will gain meaning.

・Related articles:Introduction to hypothesis thinking for marketers: The basics and practice for using it in practice and getting results 

Diagnostic analysis: Analysis that answers "why did it happen?"

Diagnostic analytics is an analysis that explores the causes of phenomena revealed by descriptive analytics. By digging deeper into why the results were achieved, the structure and causes of the problem can be understood.

Usage examples

It is used when investigating the cause of a sudden drop in CVR, when comparing the effectiveness of advertising measures, when verifying why sales are not increasing in a particular region, etc. Typical methods include segment analysis, interpreting the results of A/B tests, and cohort analysis (grouping users and analyzing the behavior of each group).

A perspective that explores causation rather than correlation

The most important thing in diagnostic analysis is not to confuse correlation with causation. Even if there is a correlation, such as "sales decrease on rainy days," it does not necessarily mean that "rain is causing sales to decrease." There may be a hidden causal relationship, such as many people refraining from going out on rainy days. In the field of marketing, this distinction tends to become blurred. Before seeing a correlation, such as "sales increased in months when advertising expenses were increased," and concluding that "advertising expenses drove sales," it is important to consider other factors as well.

A perspective that questions assumptions

Diagnostic analysis requires a perspective that challenges assumptions. Rather than assuming "this is probably the cause," formulating multiple hypotheses and verifying them with data will greatly affect the quality of decision-making. If you fail to identify the cause, you will end up taking measures that are off the mark, wasting time and resources.

・Related articles:The Basics and Importance of Causal Inference in Marketing

Predictive analytics: Analysis that looks ahead to "what will happen if we continue as we are?"

Predictive analytics is the analysis of past and present data to predict what may happen in the future. It uses trends and patterns to predict what will happen next.

Usage examples

It is used in many areas of marketing, such as forecasting next month's sales, determining the risk of customer churn, simulating reactions to new products, predicting LTV (customer lifetime value), etc. It is also applied to recommendations, personalization measures, and budget allocation by channel, and these also fall within the realm of "prescriptive analytics," which will be discussed next.

The future is not something to "predict" but something to "prepare for"

A common misconception about predictive analytics is that it is expected to be an absolute tool that can predict the future. However, in reality, predictive models only show "what is likely if current trends continue," and do not guarantee an absolute future. Predictions are like navigation. They are not perfect maps, but rather guideposts that tell you things like "If you continue this way, there will likely be traffic jams" or "This way might be smoother."

Understand the assumptions behind the predictive model

When using predictive analytics, it is important to understand the assumptions on which the model is based. Because predictive models are based on past data and trends over a specific period, they have the limitation of being vulnerable to changes in the environment. For example, a model trained on data from before the COVID-19 pandemic is not suitable for making predictions during the pandemic. Understanding the assumptions and applying them flexibly is the key to utilizing predictive models for decision-making.

・Related articles:Turning environmental change into your ally: Marketing strategies to win in the market 

Prescriptive analytics: Analysis that guides "How can we do better?"

"Prescriptive analytics" is an analysis that derives optimal actions based on the results of predictions. It is the step that moves from "predicting" the future to "changing" it.

Usage examples

"How can we allocate next fiscal year's marketing budget across channels to maximize sales?" Answering this question is a classic example of prescriptive analytics.MMM (Marketing Mix Modeling)Typical examples include budget optimization simulations, dynamic price adjustments, and approach optimization to maximize the effectiveness of measures for each customer segment.

It doesn't always give the right answer

A major misconception about prescriptive analytics is that it is a magic wand that will automatically determine the best course of action, when in reality there are many elements that require human judgment, such as setting business goals, defining constraints, and determining priorities.

Check the assumptions behind the analysis results and suggestions

It is important to consider the analysis results and suggestions from the perspective of "What assumptions were the suggestions based on?" and "Do those assumptions match reality?" Prescriptive analytics is not a substitute for human judgment, but rather a tool to support better decision-making. By combining data and human judgment, better actions become apparent.

・Related articles:What is MMM? Explaining its features, procedures, examples, etc.

Select the type of analysis based on the question, and organize it to avoid confusion in practice

So far, we have explained the four types of analysis individually, but in actual marketing, it is important to use them consecutively.

Furthermore, even when using the same data, the analysis required will differ depending on the question being asked. The fact that "sales have fallen" requires completely different approaches depending on whether you are asking "how much have they fallen?" (description), "why have they fallen?" (diagnosis), "will they continue to fall?" (prediction), or "how should we deal with it?" (prescription).

While talented marketers may be able to intuitively distinguish between these two categories, being aware of the four categories of data analysis will help them put this intuition into words. This will help create a common understanding within teams and across the entire organization, enabling more consistent data usage.

Common marketing challenges and the type of analytics they're suited to

This article explains four types of analysis, but in practice you may often find yourself struggling to choose which specific method to use (multiple regression analysis, MMM, decision tree, etc.).

The following article organizes 13 representative analysis methods by business issue and explains how to choose the right one to achieve results. If you would like to proceed with selecting a specific method, please read it.

·Related article:The Right Way to Analyze Marketing Data: 13 Essential Methods for Improving Results and How to Choose Them

How to effectively communicate the results of each analysis to stakeholders

The full value of analysis cannot be realized unless the results are properly communicated to stakeholders and used to drive decision-making. Different analyses have their own optimal communication methods.

How to communicate descriptive analysis: "Share the facts and provoke further questions"

When communicating the results of descriptive analysis, many people tend to just list the numbers and leave it at that. However, what stakeholders want is not the numbers themselves, but what the numbers suggest.

Tips for communicating:

• Instead of saying, "Sales were up XX% from last month," explain key changes, such as, "Sales were down XX% from last month, reaching their lowest level in the past six months."
• "To clarify the cause of this change, we would like to next look at the trend in △△," providing context to the numbers and leading to the next action (= diagnostic analysis).

How to communicate diagnostic analysis: "Clarify the hypothesis and verification process"

In diagnostic analysis, it is important to show the logical path of "why a certain conclusion was reached." Stakeholders need to understand the assumptions, constraints, and verification process of the analysis in order to judge the validity of the conclusion and make a convincing decision.

Tips for communicating:

• "We formulated three hypotheses and analyzed the data from the past three months for each. As a result, we found that factor XX had the greatest impact. However, although there is a correlation between XX and sales, we have not yet been able to identify a causal relationship." Explain the hypothesis and verification process clearly, distinguishing between correlation and causation.
• Clarify your assumptions and constraints by saying, "This analysis assumes that △△ will not change. Also, we have excluded the impact of □□, but it may not be negligible."

How to communicate predictive analytics: "Inform decision-making, including uncertainty"

The most important thing when communicating the results of predictive analysis is to avoid the illusion of certainty. Predictions are merely outlooks into the future and therefore necessarily involve uncertainty.

Tips for communicating:

• Show the assumptions and uncertainty range of the forecast, such as "If the current trend continues, there is a high possibility that XX will occur next month, and it is expected to fluctuate within a range of ±△%."
• "However, if the following factors change, the forecast will change," clearly indicating that the results will fluctuate if the assumptions change.

How to communicate prescriptive analytics: "Organize options and their characteristics, and provide information for decision-making"

In prescriptive analysis, it is important to create a foundation for making a decision by presenting multiple options and their respective advantages and disadvantages, rather than asserting that "this is the optimal solution."

Tips for communicating:

• Organize your options by saying, "We have considered three options. Plan A is ○○, Plan B is △△, and Plan C is □□."
• Show the criteria for making decisions, such as "If cost is the most important factor, then plan A is appropriate, and if speed is the most important factor, then plan B is appropriate."

Coordinating communication by stakeholder

When sharing the results of data analysis, it is important to adapt the way you communicate depending on the other person's position and interests.

For example, it is necessary to provide information that leads to specific actions for field personnel. It is important to present information that leads to measures or next actions, such as, "The XX element of the XX campaign was effective. It can be used in XX measures" (diagnostic analysis), or "By changing XX to XX, we can expect to improve XX" (prescriptive analysis).

Meanwhile, management is required to clearly and quantitatively communicate the impact on the business as a whole. Summarizing key points that directly impact decision-making, such as "Sales fell XX% compared to the same period last year, resulting in a loss of XX billion yen compared to last year" (descriptive analysis), or "If no measures are taken to address XX factor, there is a risk of XX occurring in six months' time" (predictive analysis), should be presented succinctly.

Common "analysis problems" in practice and how to deal with them

Even if you understand the above, there are still some typical challenges you may face when incorporating data analytics into your daily work.

Common problem 1: I'm told to analyze something, but I don't know what to look at.

You've been told to "just do the analysis," but you don't know which indicators to look at or where to start. This is a typical example of what happens when you start an analysis without a clear question. If your question remains unclear, no matter how much time you spend collecting and analyzing data, you won't gain any useful insights.

countermeasure:First, start by clarifying "What do you want to determine?" and "What do you need to know to do that?" Once the question is clear, the necessary indicators, analytical methods, data, and aggregation granularity will naturally become clear. Conversely, "analysis that begins with an unclear question" will almost certainly end up as a report that you can only look at.

2. We have dashboards, but no one uses them for decision-making.

Even if you create a beautiful dashboard, there are surprisingly many cases where it is not looked at on a daily basis, or if it is looked at, nothing changes. This is because the design stage was unclear about who should look at it, when, and why. Also, the numbers displayed may not pose questions that lead to action.

countermeasure:When creating a dashboard, it is important to clarify who will be checking the changes in these indicators and what decisions and actions will be taken based on them. Ideally, the dashboard should not simply be a list of descriptive numbers, but should incorporate a process that leads from diagnosis to prediction and then to prescriptive decisions.

Common 3: "I understand the numbers, but how do I proceed?"

In this case, analysis results are available, but they are unable to translate into action. Even if you know that "CVR is declining," you may not be able to take the next step and ask yourself, "So what should we improve?" or "What should we try?", and the results remain at the report stage.

countermeasure:When answering your analysis results, make it a habit to always ask "So what?" and "What's next?" It's important not to just understand the numbers, but to always include action suggestions. For example, if "CVR has dropped," make hypotheses such as "Is it due to the creative?" or "Did it change the inflow channel?" Then, to verify your hypotheses, it's important to think about specific actions to take, such as A/B testing or improving the message.

Problem 4: No one can translate analysis results into meaningful action

Another major reason why "action cannot be taken" is the lack of personnel who can bridge the gap between analysis and business.

To translate the analytical flow of description → diagnosis → prediction → prescription into business outcomes, a "translator" who can connect the meaning of the data with on-site judgment is essential. For example, data scientists are analytical professionals, but may not be experts in information and prioritization in the business world. On the other hand, marketers may have business acumen but struggle to understand and interpret analytical methods.

countermeasure:This translator-like role does not necessarily have to be a full-time position. It can be an in-house marketer with strong analytical literacy or an analyst with business understanding. It is also effective to utilize an external partner or consultant.

Furthermore, if you end up collaborating with a data scientist, it's important to respect each other's expertise and communicate in a common language. For example, rather than a request like, "Please forecast next quarter's sales," it's better to say, "We'd like to forecast next quarter's sales using predictive analytics and derive actionable measures using prescriptive analytics. Our recent sales slowdown is due to an increase in existing customers abandoning our services and a decline in revisit rates. Therefore, we'd like you to analyze metrics related to repeat visit rates and LTV." This will clarify what the data scientist needs to do and the output the marketer expects, leading to more constructive communication. It's not a problem if the hypothesis is wrong. In fact, having a hypothesis clarifies the starting point for the analysis and improves the accuracy of the dialogue with the analyst.

What marketers need is not "understanding everything" but "the ability to ask questions and apply them"

So far, we've introduced four analytical approaches in detail, but the most important thing for marketers is the ability to convert business issues into "questions." The first step to success is being able to convert vague issues such as "We want to increase sales," "We want to reduce churn," and "We want to increase customer satisfaction" into specific questions such as "For which segment, for which product, how much do we want to increase sales by, and by when?" and "What do we want to improve?"

Furthermore, not all marketers need to have advanced analytical skills. For example, a chef can understand the characteristics of ingredients and create excellent dishes without having specialized knowledge of agriculture or fishing. In the same way, marketers can compete just fine if they understand how to use analytics. What's important is to understand the meaning of the analytical results and have the perspective to consider how to use them in decision-making.

In conclusion

Now that you've finished reading this article, take a moment to reflect on your own team. What questions are you asking and what analysis are you conducting?

If you feel like you're just looking at data or reports are piling up, start by redesigning your question. If you already have a clear question, take another look at whether you've chosen the most appropriate analytical method for that question. Analysis only creates value when the question and method are a good match. We hope this article will help you delve deeper into your questions and help you use data to achieve results.

If you have concerns like "I want to use data more strategically" or "I'm analyzing it, but it's not leading to the next step," then please give it a try.Please contact XICA.

For over 10 years, XICA has worked with over 300 companies on analytics that support marketing decision-making. By utilizing advanced methods and data science, including MMM, we can derive actionable strategies that can be implemented even in complex environments.

Our support goes beyond simply providing analysis. We accompany our clients from start to finish, starting with clarifying the business issues they need to address, to figuring out how to incorporate analysis into their decisions and how to make it take root in their organizations. Through "usable analysis," we improve our clients' decision-making capabilities and provide support that directly translates into business results.

Frequently Asked Questions (FAQ)

Q1. Do the four analyses have to be carried out in order?

No, they don't necessarily have to be in order. Your business situation and challenges will determine where you start.

Starting with a description (most common)

• Understand the current situation with regular reports and dashboards
• If you notice something unusual, go for diagnostic analysis
• Once you see a trend, move on to predictive analytics

Cases start with diagnosis

• If the problem is already clear (e.g., "CVR has dropped sharply")
• Identifying the cause is the top priority

Cases that start with prediction

• Budgeting and planning periods
• If you want to know first what will happen next fiscal year

However, prescriptive analytics cannot be performed without diagnostic or predictive results, so it is often performed after other analyses.

Q2. I was told to "analyze" something, but I don't know where to start.

First, start by clarifying what you want to determine and what you need to know to do so.

Specific steps

1. Check your goals：What decisions do you want to make from this analysis? (budget allocation, continuation/cancellation of measures, planning new measures, etc.)
2. Convert it into a question: "I want to increase sales" → "In which segment do you want to increase sales, by when, and by how much?"
3. Identify the type of analysis :
   • Need to understand the current situation → Descriptive analysis
   • I want to know the cause → Diagnostic analysis
   • Want to predict the future → Predictive analytics
   • Deciding on a course of action → Prescriptive analysis
4. Check the data you needWhat data do you need to answer that question?

Once the question is clear, the necessary indicators, analytical methods, and data granularity will naturally be determined. Conversely, if the question remains vague, no matter how much time you spend, you will end up with a report that is just something to look at.

Q3. I created a dashboard, but no one uses it.

This is a very common issue, and it often stems from a lack of clarity about who should be looking, when, and why.

Points to be reviewed

• Identify the user: Management? On-site managers? Practical staff?
• Decide when to watch: Before the weekly meeting? During the monthly review? During an anomaly alert?
• Design for action: Who will do what if this indicator changes?

Tips for improvement

Rather than simply listing numbers, incorporate a flow that connects diagnosis → prediction → prescriptive decision. For example, instead of just displaying "CVR dropped 15% compared to last week" (description), you could also include "CVR of major landing pages dropped. This may be due to changes in inflow paths" (diagnostic suggestion), which will make it easier for the viewer to imagine the next action.

Q4. I have the analysis results, but I don't know what to do next.

This is a case where the analysis stops at "description" or "diagnosis" and does not reach "prescription."

(I.e.

Make it a habit to always ask yourself the following two questions about your analysis results:

• "So what?"What does this result mean?
• "What's next?": What exactly should I do?

Example :

• ✕ End with "CVR has decreased"
• "The conversion rate has dropped" → "Is it due to the creative? Or a change in the referral channel?" (hypothesis) → "Verify with A/B testing" or "Analyze by referral channel" (next action)

It is important not to just grasp the numbers, but to always include action proposals.

Q5. What are some tips for working effectively with data scientists?

Respecting each other's expertise,Communicating in a common languageIs important.

• ✕ Ambiguous requests : "Please give me your sales forecast for next fiscal year."
• 〇 A clear request "We would like to forecast next quarter's sales using predictive analysis and derive actionable measures using prescriptive analysis. Sales have been slowing recently, and we hypothesize that this is due to an increase in the number of existing customers abandoning the business and a decrease in the rate of return visits. Therefore, we would like you to analyze indicators related to repeat rates and LTV as well."

Key Points

• Clarify what you want to know (purpose)
• Share why you want to know (business background)
• If you have a hypothesis, communicate it (it's okay if it's wrong; it will actually give you a clear starting point for your analysis).
• Communicate how to use it (how to use it in decision-making)

It's okay if your hypothesis is wrong. In fact, having a hypothesis will give you a clear starting point for your analysis and improve the accuracy of your dialogue.

Q6. How should we be aware of the difference between correlation and causation in practice?

It is important not to confuse "things just going together" (correlation) with "cause and effect" (causation).

Examples of common misconceptions

• "Sales also increased in the months when advertising expenses were increased" → Therefore, advertising expenses boosted sales
• "Sales decrease on rainy days" → Therefore, rain decreases sales

How to check in practice

1. Consider other factorsWere there any other changes during that time?
2. Check the time contextDoes the cause occur before the effect?
3. Validate with multiple data: Do the same relationships appear across different time periods or segments?
4. Experimentally confirm: Is it possible to intentionally change the conditions and verify them using A/B testing, etc.?

It is difficult to prove a complete causal relationship, but even just having the suspicion that "this might be a correlation" can prevent you from making an inappropriate judgment.

Q7. If the accuracy of a predictive model is low, is there any point in using it?

Yes, there is. The purpose of prediction is not to "predict the future perfectly" but to "make better decisions."

The value of prediction

• Even with a 60% accuracy rate, the quality of your judgment will definitely improve compared to intuition alone.
• It will be an early warning that "if we continue like this, it will be dangerous"
• You can compare multiple scenarios ("Plan A is likely to result in XX, while Plan B is likely to result in △△")

the important thing is

• Use with an understanding of the uncertainty of predictions
• Don't take prediction results as absolute, but combine them with other information
• Regularly verify accuracy and continue to improve

A forecast is like a navigation system. It's not a perfect map, but it's valuable enough to serve as a guidepost that tells you, "If you continue on this path, you're likely to get stuck in a traffic jam."

Q8. What are some tips for explaining analysis results to management?

For the management,Communicate the overall business impact succinctly and quantitativelyIs required.

✕ Explanations to avoid

• A litany of technical terms ("The coefficient of determination was 0.8 and the p-value was less than 0.05")
• Detailed explanation of the analysis method ("Using multiple regression analysis...")
• Reporting numbers without any conclusions (ending with "Sales have dropped")

〇 Effective explanation

1. State the conclusion first: "If we implement the XX measure, we can expect an improvement of XX billion yen."
2. Show business impact in monetary terms: "Sales fell 15% compared to the same period last year, resulting in a loss of 3 million yen compared to last year."
3. Present options and criteria for decision-making: "Plan A prioritizes cost, while Plan B prioritizes speed. The risk is ____."
4. Clarify next actions: "First, we will implement XX and then verify the effectiveness after two weeks."
5. Be honest about your uncertainties: "If no measures are taken against XX factor, there is a risk of XX in 6 months."

What management wants is not detailed analysis, but information to help them decide, "So, what should we do?"

Author information

XICA Co., Ltd. Marketing Team