Interval analysis

Video tutorial

Overview of Interval Analysis

In versions 1.11 and later, Sensors Analytics provides "interval analysis".

Product, operations, and marketing personnel all need to monitor the conversion rate of specific business processes in their daily work. Besides using funnels to look at conversion rates, measuring conversion also requires analyzing the distribution of conversion time. Interval analysis addresses this type of problem and need. By calculating the time interval between two events in a user behavior sequence, the conversion time distribution of each business conversion stage can be obtained.

Interval analysis can help you answer the following questions:

The registration process includes complex operations such as real-name authentication. We want to know the time distribution of the entire process from the start to the end of the registration.
For e-commerce products, analyze the time distribution from when a user first opens the app or completes registration to when they place their first order.
Analysis of investment and wealth management products: Distribution of time intervals from new users completing card binding to completing their first investment.

View an example of the interval analysis function application.

Interval Analysis Function Introduction

Select initial and subsequent behaviors

There are two possibilities for the initial behavior and subsequent behavior:

Initial and subsequent behaviors are distinct events. For example, internet finance products might select "registration successful" and "investment successful" respectively, while e-commerce products might select "add to cart" and "submit order." This analysis of user conversion time within the business process indirectly reflects conversion intent, allowing for targeted optimization of product experience and operational strategies.
The initial behavior and subsequent behavior are the same event. For example, online education products analyze the time interval between two classes to assess the user's motivation for learning. E-commerce products analyze the time interval between users' repeated purchases of daily necessities to predict the timing of the next purchase and make accurate recommendations.

Set initial and subsequent behavior filtering conditions

Depending on the specific analytical needs, filtering criteria can be added to initial or subsequent behaviors. Taking luxury e-commerce as an example, we might want to analyze whether users whose previous order exceeded 10,000 yuan are likely to place another order soon.

The underlying assumptions of this analysis are that users who place higher-value orders have greater trust in the platform, and secondly, these users are generally wealthier. The analysis will verify these assumptions. If they hold true, it indicates that users still have a strong willingness to purchase after completing a high-value order, and offering coupons can encourage repeat purchases.

At this point, both the initial and subsequent actions should be set to "Pay Order". Adding the filter "Order Amount Greater Than 10,000" to the initial action will satisfy the analysis requirements.

Set user filter criteria

Filter by user attributes to select the desired analysis targets. For example, only view the time intervals of behavior of female users in Zhejiang Province.

View by attribute

Interval analysis includes the following three grouping methods.

Grouped by initial behavior attributes
Grouped by subsequent behavior attributes
Grouped by user attributes

If you choose to view grouped by initial behavior, such as 注册成功 If 注册渠道 is triggered, then the users will be grouped according to the attribute value of 注册渠道 that is triggered when each user triggers 注册成功. Each user will only appear in one group.

Select aggregation time unit

Selectable data aggregation time unit:

By day
By week
Monthly

It's important to note that interval analysis does not support "by hour" aggregation. This is because "by hour" requires that the user's initial and subsequent actions occur within the same calendar hour; otherwise, the actions cannot be correlated, leading to inaccurate analysis results. See below for details. The calculation rules for pairing initial and subsequent behaviors .

Set associated attributes

It supports setting attributes to associate between two consecutive events. The attributes associated with different events can be the same or different, but the attribute types must be consistent. For example, an e-commerce platform is conducting a marketing campaign. Besides monitoring the user's behavior from the product details page to completing the purchase, it also needs to accurately determine whether the user's behavior is related to this marketing campaign. Therefore, it needs to add the marketing campaign ID attribute to the 浏览商品详情页 and 支付完成 events. This attribute can then be used as the association ID to ensure that users are matched strictly according to this pattern.

View analysis results

The results of interval analysis are presented in the form of box plots. Box plots provide a simple way to summarize a dataset using only 5 points. They can display the maximum, minimum, median, and upper and lower quartiles of a set of data.

You can see the entire time period A 事件 → B 事件 The maximum, minimum, median, and average interval times of the overall situation;

The calculation rules for each indicator in the analysis results are as follows:

Interval Count: The number of pairs that complete the interval transformation according to the interval analysis calculation rules within the selected time range (a field in the downloaded data).
Number of converted users: How many people completed the interval conversion within the selected time period. It is possible that one person will complete multiple interval conversions.
Average number of intervals per user: Number of intervals / Number of converted users (a field in the download data).
Interval conversion duration: Within a selected time range, after calculating the interval conversion pairs according to the interval analysis calculation rules, the interval conversion duration of each pair is statistically calculated.
Maximum value: The maximum interval conversion time.
Minimum value: The minimum interval conversion time.
Median: Sort the average interval lengths of each person from largest to smallest and take the middle value.
Upper quartile: Sort the average interval duration of each person from largest to smallest, and take the value at the 1/4 mark.
Lower quartile: Sort the average interval duration of each person in ascending order and take the value at the 1/4 mark.
P90: Sort the average interval duration of each person in ascending order and take the value at the 9/10 mark.
P10: Sort the average interval duration of each person in ascending order and take the value at 1/10.
Average value: Sum of interval conversion times / Number of intervals.
Average conversion time per user: Sum of average intervals per user / Number of users who convert.

Expand the table to view the details and see the interval time details for each group.

How are the initial and subsequent actions paired?

Suppose the user's behavior sequence over a certain period of time is: A → C → A → B → B → A → B → B → A → A → C → D → A → B

At this point, we analyzed what the user did. A 事件 and The time interval of B 事件 will be calculated according to the following rules:

In the event After A 事件, find the distance A 事件 Recent B 事件 represents the first interval. Continue searching for pairs of A and B from this interval onwards, ensuring that intervals do not overlap, and so on.
Selecting the aggregation time unit, such as by day, week, or month, will limit the pairings. A 事件 and B 事件 Occurred on the same day, week, or month; for example: if aggregated by day, the user's A 事件 The time was 23:50, and it happened. The time for B 事件 is 00:10 the next day. These two events cannot be paired.
If we disregard the aggregation time unit, the interval pairing result is: A → C → A → B → B → A → B → B → A → A → C → D → A → B

Typical use case examples

Taking financial investment products as an example, whether users actually invest is crucial to the business. Therefore, in product operation strategies, new users might receive red envelopes (cash gifts) to attract investment. At this point, new users could be selected as follows: 注册成功 and 投资成功 Two events were used to obtain the time distribution of users' first successful investment. At the same time, the impact of various strategies on conversion time was analyzed by configuring filtering conditions for event attributes.
See the image below for details:

Taking short video products as an example, these products offer a wealth of video content for users to view. Content is the core value that short video apps provide to users. How do we measure whether users perceive this value? Here, we choose to judge based on whether a user has watched a short video in its entirety , and we also select the time spent by new users from 启动 App to 完成播放 as the analysis object. If users generally need a long time to complete the conversion, it indicates that the cost of video screening for users is high, and users are very likely to churn. Therefore, we select the time interval from 启动 App to 完成播放 for new users as the optimization target.
See the image below for details:

Interval analysis typically reflects business conditions, and time intervals are often not suitable as optimization metrics. However, they can help us explore potential problems. For example, if the median time between order submission and payment is 5 minutes, it means that half of the user payment orders take more than 5 minutes. In this case, we should proactively consider potential problems: is it a bug in the payment function, or is there another issue causing payment failure? We need to combine event analysis, funnel analysis, and other functionalities to pinpoint the problem.

FAQ

Can the interval analysis function be used to analyze page dwell time? What is the difference between time interval and page dwell time?

A: No. Assume the registration process includes steps 手机号注册, 填写基本信息, and 实名认证. Three events. 填写基本信息 and The time interval 实名认证 is not the duration the user spent on the basic information page. This is because the user spent time on the page during 填写基本信息 and 实名认证. There may be other operations between the two actions. Once 填写基本信息 and 实名认证 are triggered, "interval analysis" can be used to calculate.