Product Operation | How to combine hot spots to make operation plans?

March 22, 2019, was originally just an ordinary Friday, but after the State Council issued the announcement at 10 o'clock , "Good news! #May1st holiday#" , everything became unusual.

The article quickly reached 980,000 views in 2 minutes after it was published, and exceeded 1.5 million views in just 10 minutes. Enthusiastic people spread the word, and the circle of friends was boiling with excitement.

As an Internet product operator , I am very happy to face such huge traffic. However, in this kind of emergency, I should accurately identify the problem and quickly come up with a new solution.

This requires operators to combine their own business status, gain real-time insight into data anomalies, and select appropriate data analysis methods to truly achieve data-driven products and refined operations.

1. Review

Let’s first take a look at what the major products did after the State Council’s notice was issued that day.

1. "Message push comes without notice"

• Official announcement! The May Day holiday has been extended to 4 days!
• Good news! This year's May Day holiday is four days
• Take 4 days off! The May Day holiday you wanted is back
• Great news! This year's Labor Day holiday is 4 days

Videos, maps, travel, social networking... news pushes from various media came one after another, and I really felt the keen sense of smell and strong execution ability of new media operators.

2. "NetEase News Hot Comments"

• It's just one day off.
• It's just two days off, just move one day of rest from the previous week and the next week to May Day.
• Still one day. Now that the company is on holiday, let them be on duty! You did this, originally...
• The best news is that we can have four days off instead of adjusting the holiday.

The creative copywriting is closely aligned with the #五一放假4天# message, and the advertising template is structured and configured to closely follow hot topics and generate traffic with one click, making it flexible and diverse.

3. "Mafengwo Travel Consumption Guide"

• According to the latest notice from the State Council, the latest version of the 2019 holiday arrangements will be updated in a timely manner.
• We have listed the 24 most suitable travel destinations at home and abroad.
• Highwan's Spring Travel Season promotion has become a hot topic. A 730-yuan coupon package for new customers is delivered in a timely manner, covering tickets, hotels, air tickets, and free travel. There is also a 688.8-yuan red envelope challenge to attract users.

For the general public, traveling during short or long holidays has become a habit and a way of life. As soon as the news of #May Day holiday 4 days# came out, travel enthusiasts instantly started the travel preparation process.

4. "Fliggy PR Data"

• "After the holiday announcement, domestic air ticket bookings from 10am to 12pm increased by more than 50% compared to the same period last week; the growth in international air tickets was even more astonishing, exceeding 150%.
• The five most popular domestic destinations are Shanghai, Chongqing, Beijing, Xi'an and Chengdu; the international cities are Seoul, Hong Kong, Bangkok, Osaka and Tokyo.

From the Fliggy platform, we can see that prices started to rise after noon. ”

…

Just like that, an unexpected holiday news of "Good news! #May1st holiday for 4 days#" instantly became a hot topic on the entire network. The media pursued the marketing, the public happily paid for it, and the browsing visit data soared instantly.

2. Analyze the problem with data

Looking back, the above actions have shown a very fast response speed, and they are well integrated with the tone of their own products, which is truly excellent.

Faced with the rapid increase of massive data, what should we do next?

You must remember these familiar scenes:

• For example, the company did a wave of promotion by internet celebrities for a period of time, and the boss wants to see the effect of the promotion, so you need to do a retrospective analysis;
• For example, if the data in a certain period of time rises or falls abnormally, your boss wants an accurate explanation, but you are confused;
• For example, one day the page views of the M site you are responsible for surge, and you need to find out the reason so as to take advantage of the situation to increase traffic again;

All of these scenarios require basic data analysis capabilities.

The value of data analysis lies in establishing a correspondence between a series of problem phenomena and business data, and through certain analysis routines, quickly analyzing and solving problems.

Next, I will focus on sharing three common data analysis methods, and then use a case study of [Analysis of the reasons for the sharp increase in data of a certain social APP during the National Day] to assist in understanding and explanation.

3. Three Data Analysis Methods

First of all, there are 9 common data analysis methods: comparative analysis, multi-dimensional disassembly analysis, funnel observation , distribution analysis, user retention analysis, user portrait, attribution search, path mining, and behavior sequence analysis.

Here we will focus on sharing the first three data analysis methods: comparative analysis, multi-dimensional analysis, and funnel observation.

1. Comparative analysis

Comparative analysis is the most basic and common data analysis method. It can intuitively show the changes in things at a certain stage, and can accurately and quantitatively express the changes/gap . It focuses on the analysis from three dimensions: "what to compare", "how to compare" and "who to compare with".

(1) What to compare

What to compare, divided into comparison of absolute value (#) and ratio value (%).

The absolute value itself is data with "value". For example, if the sales amount is 2,000 yuan and the number of views is 100 million, it is not easy to know the severity of the problem by just looking at the numbers.

The ratio value only has comparative value when viewed in a specific environment , such as active share and registration conversion rate. The ratio value alone is easily affected by extreme values.

(2) How to compare

How to compare? It can be divided into month-on-month and year-on-year comparison.

Common month-on-month comparisons include daily and monthly comparisons , which refer to the comparison of the previous time range adjacent to the current time range . They are mainly used to analyze data with short-term continuity, such as indicator setting.

Common year-on-year comparisons include weekly year-on-year and annual year-on-year, which refers to comparative analysis of data at the same position in the previous time range above the current time range . It is mainly used to observe longer-term data sets and eliminate interference from short-term data.

(3) Who to compare with

Who to compare with, compare with yourself or compare with the industry.

Compare with yourself , you can compare with yourself from different time dimensions, different business lines, and past experience estimates;

Compared with the industry , we can observe and analyze whether it is due to internal factors or industry trends. For example, when everyone is falling, can our stock price fall less than that of our peers? When everything is rising, can they all rise faster than their peers?

Now let’s go back to the above “Fliggy PR data”: “After the holiday announcement, domestic air ticket bookings from 10am to 12pm increased by more than 50% compared to the same period last week; the growth of international air tickets was even more amazing, exceeding 150%.”

Obviously,

"50%, 150%" are both proportional values;

"Compared to the same period last week..." Since the data has been rising continuously in the short term due to the news of #Labor Day holiday for 4 days#, the weekly year-on-year data is selected;

“Domestic air ticket bookings…international air tickets…” Fliggy is comparing itself. If industry data were published as a basis, it could be determined whether Fliggy is growing faster/slower than its peers.

2. Multi-dimensional disassembly

Multi-dimensional analysis is the most important way of thinking. A single indicator has no analytical value. We need to analyze it from multiple dimensions to make it meaningful and ultimately obtain more comprehensive data insights.

The essence of data analysis is to split and observe the same data indicators from different perspectives. The essence of multi-dimensional decomposition is to decompose indicators/business processes in multiple dimensions to observe data changes.

Applicable scenarios for multi-dimensional disassembly:

(1) When analyzing the composition and proportion of a single indicator , such as the number of views by column and the ratio of new and old users;

(2) Break down the processes , such as browsing and purchase conversion rates of different channels, and activity participation funnels in different provinces;

(3) Restore the scene when the behavior occurs , such as the user's level, gender, channels followed, and whether the user is in a WiFi or 4G environment.

Now let’s go back to the first scenario: “For example, the company did a wave of promotion by an internet celebrity for a period of time, and the boss wants to see the effect of the promotion, and you need to do a review and analysis…”

At this time, a multi-dimensional disassembly analysis method is needed. The general analysis idea is as follows:

(1) Analyze from the APP startup event

• View by device type , such as Android, iPhone, etc., and check the startup status of different models;
• According to the startup source , such as desktop, SMS, PUSH... different sources of startup;
• Observe by city level , such as first-tier, second-tier, third-tier and below... the launch status of different cities;
• Segment by new and old users , such as overall, new users, old users... the startup status of different user groups.

(2) Disassembly from the business process

For example, for a simple "register -> place an order -> pay" process:

• The payment funnel is viewed by channel , which may be divided into Baidu, Toutiao, WeChat public account...
• The payment funnel is divided into first-tier, second-tier, third-tier and below cities .
• The payment funnel is divided into Android, iPhone, etc.

3. Funnel observation

The funnel observation analysis method is common and familiar to us. Its operating principle is to observe the target through a series of backward-influencing user behaviors.

It is applicable to businesses with clear business processes and business objectives, but not applicable to businesses without clear business processes and with complex jump relationships.

Observe the health of core business processes through the funnel.

Let’s take stock of the pitfalls that are easy to fall into when building a funnel:

(1) First, funnel observation requires a certain time window . The corresponding time window should be selected based on the actual business situation.

• Observation by day is suitable for situations where the impact on the user's mind is only effective in the short term, such as some short-term activities (currently effective, countdown settings, etc.);
• Weekly observation is suitable for situations where the business itself is complex, the user decision-making cost is high, and it takes multiple days to complete, such as investment and financial management, account opening and capital injection;
• Monthly observation is suitable for situations where users have longer decision-making cycles, such as decorating and buying a house.

(2) Secondly, funnel observation has a strict order . You cannot use the ABCDE funnel (data from the search path only) to look at the data from ACE (data from the classification, search, and recommendation paths).

(3) The calculation unit of the funnel can be based on users or time.

• Observing users means caring about the promotion of the entire business process;
• Observing events means caring about the specific conversion rate of a certain step, but it is impossible to know the actual situation of the event flow.

(4) When the data of the result indicator does not meet expectations, you need to check whether there is only one funnel that can reach the final goal , that is, check whether there is a second pit.

IV. Case Study: Analysis of the Reasons for the Rapid Increase in Data of a Social APP during the National Day

The scenario is this: there is an anonymous social APP, similar to Tantan. The data range is between September 1 and October 14, 2018. The data surged during the National Day holiday. Let's try to analyze the reasons.

(1) First, define “data surge”

As an anonymous social product, you can choose to observe the "registration success" event.

Since the behavioral data is generated in a short period of time, we choose to focus on whether the number of registered users has increased significantly on a daily basis, and view it according to the "number of triggered users" of the "successful registration" event:

(2) Discovering abnormal positioning issues

From the above line graph of the number of users triggered by successful registration, we can see that there is a high daily growth difference in the number of registered users during the National Day holiday, which is the peak on the right side of the line.

From this, we can judge that the number of registered users increased significantly during the National Day holiday due to some reason. The specific reason needs to be further analyzed.

(3) Multi-dimensional analysis

When observing by operating system, we can find that Android's growth rate is significantly higher than iOS. iOS has a slight increase, but the increase is not obvious.

This step still cannot directly locate the problem and requires further disassembly and analysis.

According to the above figure, based on the registration method , the three registration methods of WeChat, Weibo, and mobile phone numbers all increased during the National Day period and the increase was similar. It can be preliminarily judged that the registration method has nothing to do with the data anomaly.

The above chart shows that both boys and girls have increased during the National Day holiday, with boys slightly higher than girls. However, the problem cannot be directly identified and further analysis is needed.

From the above graph, we can see that users of different age groups have increased during the National Day holiday and the increase is similar. It can be preliminarily judged that age has nothing to do with this data anomaly.

Here comes the question! According to the province observation , the above picture clearly shows that there is an abnormally high discount!

In fact, the daily month-on-month growth rate in Hainan Province increased. In addition, the month-on-month growth rate in Yunnan Province was also significantly higher than that in other provinces.

Based on the above observations and analysis, it can be basically judged that the sharp increase in data during the National Day holiday is related to the substantial increase in the number of registered users in Hainan Province and Yunnan Province. The specific reasons need to be further analyzed.

Continuing to observe by city , the filtering condition is set as province equal to Hainan Province and Yunnan Province. It can be seen that the data of Lijiang City, Dali City, Sanya City and Haikou City soared during the National Day.

Based on the above multi-dimensional analysis, it is found that the data surged during the National Day holiday, mainly due to the significant increase in the four cities of Lijiang, Dali, Sanya and Haikou .

These four cities are all tourist cities, and the period of data growth coincided with the National Day holiday.

Therefore, it is speculated that the anonymous social product may have carried out promotional activities for these four popular tourist destinations during the National Day holiday. The specific reasons for the sharp increase in data still need to be communicated and confirmed with colleagues in the market, operations, or those responsible for growth.

at last

Let’s go back to the “Good news! #May Day holiday 4 days#” mentioned at the beginning. After this announcement was issued at 10 o’clock, there was such an awesome team that they had already started a series of response measures before 11 o’clock:

(1) First, we monitored in real time that the data surged after the announcement of the May Day holiday at 10 a.m., and we deeply felt the immediate impact of the May Day holiday news;

(2) ASO has already made “air ticket booking” the second most important thing, and has also started to optimize special air tickets;

(3) Placement in information flow channels, adding air ticket materials to advertising creatives;

…

In this way, when the east wind blows, everything is ready, and a large wave of traffic dividends can be obtained cleverly and appropriately.

The above case studies all reveal the importance of data analysis capabilities, and we really need to have the core skills to understand data and gain insight into the business.

Whether it is products or operations, what we do every day is closely related to actual business. Business and data are inseparable. If we want to do a good job of business support in our respective positions, we need to have certain data analysis capabilities.

Product operators who don’t understand data analysis are doomed to have a bad life. Your boss will dislike you for not being full-stack, your colleagues will look down on you for not understanding data, and in the end, even you will doubt whether you are OK.

Once you enter the product operation industry, you have to find a way to live a good life, and data analysis is a must.

Source: Zhao Xiangwei