Operational Growth in Practice: Five Minimal Cases of Achieving Business Goals

In 20 14, I bought a MAC book as an example. My mother is not very happy. The old man was used to saving, thinking that the Lenovo porcelain I used was cheap, so he made a positive correlation with the data. I said, look, old lady, I use Lenovo notebook, slow down. Basically, I go to bed at 2 am every day. Ever since I got a MAC notebook, it's extremely fast and I have rested at 9 am. The old lady was so happy that she said directly, why don't you buy two? Then you can have a rest after dinner.

Parents in the world are cooks. Who doesn't think about children's health? In this way, I bought an ipad.

This is the persuasive function of data in real life, which has been proved in court.

The same is true for operation posts, which rely on products, technology, news engines, markets and other kitchens to eat. If there is no data to speak of, it is difficult to shake them.

However, in our business, data itself is naturally weak. Like me, I don't learn math well at school and I don't know statistics. I can't even afford a dish in the vegetable market, so I am very sensitive. What should I do? No solution, only daily experience, so starting from 20 10, I tried to speak with data and practice my logical ability. I'm usually lazy, I only know data, and I'm diligent. I read books and remember things when I have time.

Over the years, I have dealt with data product managers, data analysts and statistical directors. Every time I hear them say a new concept, I will go to Baidu for consultation. But I still don't know what confidence interval and probability distribution are. I am stupid, so I persuaded myself to put forward a concept of minimalist data and optimized source table, trying to make decisions by relying on an Excel table and simple data. Data Xiaobai can also start and quickly improve his business growth. This paper summarizes some cases in which I relied on minimalist data to improve the growth of business goals in those years.

Let's talk about four characteristics of minimalist data:

1, data is easy to obtain, such as WeChat background data, GA data, standard report data made by company data department and so on.

2, the analysis is extremely simple, and you can get insights by using Excel. I used to like to use GA, because it is powerful and free, and the data in various dimensions looks high-end and classy. However, since GA was blocked and the company stopped using it, I began to think about making a set of local data and organizing all the operating projects into a daily report. I named it the optimized source table. In this way, I can use Excel's powerful data analysis function, such as pivot table, to guide my operation.

3. The idea of data operation is simple. Operation is the process of finding problems and then solving them. Advanced data model is generally used in business decision-making, and operation is dominant and basically useless.

4. Minimalist data has some errors, and there will be personal subjective experience in it. I usually make decisions as long as I have confidence in the data. There is no accurate data in the world, only relatively accurate data. I once read a sentence: a mistake that can make people gain something is far better than doing nothing, which is very suitable for operation.

In addition, in all cases of this paper, the data part has been treated in a minimalist way, bypassing the links of burying points, formulating and implementing monitoring strategies, data cleaning, etc., and only talking about the core methods for everyone to see at a glance. There are two core words that I often say, trial and error and optimization. Trial and error is the method, and optimization is the soul.

Ok, let's take a look at the sharing outline of this article:

I often optimize the source table and record the daily data. Sometimes it's easier if the data department has standard reports.

Optimizing the source table will not bring you much value in the short term, but when the data accumulates to a certain order of magnitude, you will get a lot of insights from this table. For example, you can reasonably deduce and decompose monthly, weekly and daily goals through the daily performance table of business goals, and truly' know what you are doing every day'. Because the daily target performance table shows your business rules, especially when you are responsible for the operation of the whole product project and set KPI for subordinates, they will never say that you are patting your head again.

Let me give you an example. I learned it from the book Data Management. Since I read this book on 20 14, this technique has been applied to the operation of internet products. Following the principle of minimalism, I only say how to break it down into months, because the idea of breaking it down into weeks and days is similar. I'm just offering an idea.

First, we need to find historical data. I found the daily UV data of PC users throughout the year from the company data platform. Here I take UV as an example, and you can also take LV, Visit and PV.

Calculates the values of the nth week and the week fields based on the date data. Exclude abnormal dates, mainly holidays, because the traffic of general websites is abnormal on holidays, and exclude the date of special promotion period, which depends on the usual operation diary to record daily operation behavior. Look at this photo:

We began to see through this table and break it down.

Holiday information and special promotion days are marked during the perspective, filtered out during the perspective, and are not included in the summary table.

Did you find it? If you look at the line chart, it is obvious that business can be divided into off-season. Therefore, when we complete the annual target, we should also reasonably distinguish how much is completed in the off-season and how much is completed in the peak season. Instead of the average monthly completion.

The weight is an artificial value of the user we calculate. Generally, attach the lowest value to 1, such as May in the figure, and then divide the average UV value of other months by the UV value of May to get the weight of each month.

For example, the total UV of 20 15 is 8.8ww, and the boss will double it for you. Then you can share how much you want to accomplish every month by weight. For example, in the diagram of 65438+ 10 month, I calculated the proportion of 1 month weight to the total weight, and then multiplied it by KPI to get the KPI to be completed in 65438+ 10 month.

So, what is the purpose of decomposing KPI?

1. Make an annual operation plan as soon as possible. If you can't achieve your goal in the first month, give feedback to your boss in time and apply for resources as soon as possible;

2. Manage the boss's expectations and control the speed of kpi completion. Don't get out of control in the first month and finish it early. At that time, the boss will give you a higher kpi, and sometimes the pits are dug by yourself.

This is an activity planned by our collective team. There are five people, and I am mainly responsible for copywriting and data statistics.

At that time, the boss told the members of the new product team that you can choose 200,000 users as seed users from the existing 40 million users, and then introduce 65,438+million initial users to the new product. The average cost of each user is not more than that of 5 yuan, so you left this sentence.

We have resource and cost constraints. Throw it out first, regardless of the cost. The screening of 200,000 seed users is very important. This is the ability to segment users. We are thinking about it, and we will recruit in autumn soon. Fresh graduates have a strong desire to find a job. So we carefully selected 200,000 fresh graduates with high activity as the seeds of starting a prairie fire.

We have planned an activity to promote sharing and innovation for these users. It happened that the fresh graduates of that year arrived in autumn in advance, and many top 500 customers entered the school in advance. So, our campaign stunt started with these big customers. The theme is that some famous enterprises recruit in advance, invite students to join in getting red envelopes and give them variable rewards. There is no upper limit for red envelopes.

Next, we planned an exchange activity based on this theme. How did we do it? When I resumed the quotation, I combed the process. Let's take a look at this Excel table.

This table is the data of the whole optimization page and effect. Let me pick a few key points.

Such as channel strategy, from which channel to pull new products, which part of users are used for testing, which part is officially promoted on a large scale, and so on. Because of the particularity of WeChat platform, SMS, app push, page advertisement or interception, EDM and other channels. It is not convenient for users. We also want to do it through our own WeChat size, but the users of our WeChat fans all have identities, not just fresh graduates. Compare to compare, according to experience, choose a certain channel to do it.

Also, when planning the first draft, the activity flow and key node data must be planned and monitored in advance. This needs to be mentioned. Since it is an activity planning, it must be considered comprehensively. Otherwise, after you go online, the technology may not see the log data, such as the channel conversion rate, the number of shares brought through the channel, and the number of successful registrations. At this time, no matter how busy you are, I made a big mistake in this activity.

Don't promote the activity in an all-round way as soon as you go online. It is necessary to continuously optimize key pages or key process nodes. After the transformation of pages and processes is optimized to a certain extent, I feel that it can no longer be optimized, and it can still be pushed back through the transformation data of nodes. According to the conversion rate, we can calculate that we can achieve our goals, and then we can fully promote them.

Because this case is mainly about the conversion rate of the page, I fixed the proportional relationship between channel conversion, sharing and invitation in the first test. For example, the channel conversion rate of this activity has reached about 32%, and the ratio of sharing and invitation is 1: 5, which means that 1 sharing can bring 5 registered users.

There are also some transformation data that we can't control, such as the transformation of the registration process. The registration process is a standardized functional module, and the optimization changes greatly, so this optimization is omitted and we don't pay attention to it. Of course, in this activity, the registration process actually harmed many users.

After fixing some secondary variables, we put the optimization goal on two key pages, one is the page introducing seed users, which I call the promotion and sharing page. There is also a page that friends click to enter when users share it with friends or friends. I call it transferring album pages. These two pages, we constantly test, a total of three versions, until the sharing rate and registration success rate are optimized to the best level. To promote it on a large scale.

This link is the most critical and tangled. For a whole month, we tossed and turned. The final result is that through these three optimization rounds, we have successfully achieved our goal. It took 1 month to optimize Version 3 activity, which we think is worthwhile, because this activity process and mode have been verified by us to be feasible, can be done for a long time, and has a particularly strong expansibility. For example, we can make an activity background to copy our activity mode specifically for different user groups. So the slowness in the early stage is for the quickness in the later stage. With the background tools, you can basically do a similar activity in a week. This is the value of optimization, the cultivation of internal strength, and the purpose is to enhance the core operational capability.

Now let's talk about the optimization idea of the page. What do we think of it? In fact, there are only three key points to sum up:

For example, in the first edition of the promotion sharing page, we turned the Bole Award into a red envelope, and then the sharing transformation increased by 3 points. It is more obvious to turn the album page. We added an invitation element to the button, and we didn't talk about it for the first time, so we quickly accepted the offer. I invited the sharer to do an article. It's like your colleague telling you, let's have a big dinner tonight. You may hesitate because you're not sure whether he invited AA or finished eating. He suddenly said, I forgot my wallet. And if your colleague says, come on, I'll treat you to a big meal. If it were me, I would definitely go. So we changed our mind, and the conversion rate of transferring album pages increased by 10 points.

In the first and second editions, we ignored this element and made a list of famous enterprises. Later, we thought that users might think this list was clickable, which led to misleading. Then users find that they can't click, and they will be a little emotional, lacking in sharing motivation and registration motivation. So in the third edition, we excluded this interference from visual design.

The key is that in the third edition, we changed our thinking. In the first edition and the second edition, we won the red envelope only after the user invited friends to join the registration. If we let the seed users forward it, we can get the red envelope and force it to be forwarded. Imagine that users will feel that this activity is highly credible and the sharing rate will increase. Maybe he will not only share it with his friends, but also with various groups. So when the user clicks the Join Now button, a floating layer will pop up to remind the user that they can get a red envelope after sharing, and they can also get it after introducing students to join.

Along this line of thought, we successfully achieved our goal and gained a mode of activity. In the process of optimization, pay attention to the data, record the data of each activity, monitor the transformation in real time, and show the data of each trial and error. As you can see in the picture just shared, I have listed all the key data.

Finally, when we pushed it all, we introduced more than 57,000 registered users. Why did the final number of registered users reach nearly 65438+ 10,000? Moreover, the finally introduced registered users only promoted the data of the day, and it was still growing the next day. Because the introduced registered users became seed users, they also began to snowball forward.

When our APP was first launched, we couldn't reach users directly except passively receiving user feedback through some statistical tools, such as major event notification, pulling back silent users, research and so on. Therefore, our product has made a push notification product, and I will call it a job search assistant for the time being, which is similar to the WeChat official account built into the app. I will push some information regularly.

Because our copywriting level is tempered for a long time and we have had good experience. I thought, copywriting alone may not boost the click-through rate of push. Do I have any other gaps to improve my performance? After thinking about it, it suddenly occurred to me that if we know how much users like to push content and then push related content, can we increase the click-through rate again?

My demand has come out: I am eager to get the user's preference for content.

So how can I gain insight? I need to make a trial and error strategy, which is my favorite working method in operation.

Trial and error is the most reliable means of operation and the core of operation post. Trial and error is most afraid of irrationality, so trial and error strategy is very important.

My trial-and-error strategy, look at this table:

The simple explanation is as follows:

Trial and error users' preferences for content at different time points. I aim to sum up the monthly rules. If your business has rules to follow, you can put it in quarters or even weeks.

Click rate of copywriting. If the user's preference for content is regular, there will definitely be a high click-through rate at some point.

The determination of trial and error variables is the key node of the rationality of trial and error strategy. The fields marked orange in the above table are variables.

It should have been arranged in advance. I tried the wrong type of content instead of an article. What is a content type? For example, the channel of portal station, science and technology, entertainment, military affairs, news and so on. I mainly focus on seven categories of content. For example, interview strategy, online application strategy and resume strategy. , are planned in advance, reasonable arrangement every week. Release gradually. For example, I send information summary posts every Monday, push resume strategy on Tuesday, push interview on Wednesday and so on.

The time point is sent at a fixed time every day. For example, I always choose to send it at 8 pm. As for how this time came about, I got it through investigation. I have a micro-signal for the copy test, because I usually answer the questions of recent graduates, so they are very kind to me. I sent a research post asking them when it was appropriate to push information to them, and 80% people said 7-9 pm.

Friends who have made APP notification products know that iOS can't count the data received. In order to be more scientific, we only use the Android client for testing. The click rate of copywriting is more reliable than the click rate.

In other words, who are you trying to make mistakes with? This is very important. I now know that as long as the users of recent graduates are pushing relevant job information, most students will not be too annoyed. Because the job search is just needed, if the job search is successful, most of them will close the notice or uninstall the APP. But I didn't know it at the time, and I found the information pushed very disturbing. So in order to send messages every day to prevent users from complaining, I only choose users with certain characteristics, such as highly active users, that is, users who have logged in and delivered on that day; Or silence users, which can reduce harassment, because if silent users are activated by my push, they will immediately return to the active pool and can't receive my push. Moreover, the reason why silent users are silent may be that there is no demand. If he clicks my tweet, it proves that he likes my content. Just in line with my purpose of trial and error.

As we all know, the title of Android can be customized. Unlike iOS, the title can only be the brand name. Because my purpose is trial and error, not to get the click rate effect, so my title is not customized, just write the brand name. Easier to control.

The most difficult variable to control. The click effect of the title party is the biggest, but it belongs to the abnormal data of trial and error. I must ensure that the level of copywriting is within a range of strength in order to get reasonable business cognition. This involves the trial and error strategy of copywriting. This is more complicated, so I won't talk about it today. In my previous work experience on WeChat and Content Channel, I tried to sum up the strength of copywriting. I know which copy is strong and which copy is weak, because I personally trade: I record data every day, so I can reasonably know the normal click level of each copy. As long as you remember, there are levels of copywriting. In this case, the intensity level of copywriting is 2, and under normal circumstances, the click rate fluctuates by 2 points.

Of course, if you don't have the level of copywriting, you can also compile several articles for each content type, and then try to subdivide the target users, such as users who log in on the same day, and set up a filtering mechanism: each user only receives it once during the trial and error period. So you can push these articles to the same type of users.

Pay attention to data cleaning. For example, sometimes the sending channel is unstable, it can't be sent out, or there are statistical errors, so these abnormal data should be excluded when cleaning up.

In this way, I finalized seven variables that affect my business cognition. Try to get reasonable and scientific advice!

1 year later, trial and error ended, and I began to sum up the rules. I saw through such a table:

Did everyone see it? Content types marked in pink are definitely the most popular content this month. Then, in a certain month next year, if I increase the intensity of users' favorite content, will I be able to boost my performance?

This is my content trial and error strategy. The conclusion is simple and the process is very tangled.

Of course, my trial and error case is a protracted war, because we are a mature product. If your product is in the initial stage or growth stage, you can choose short-term trial and error, as long as you control two points:

First, the goal of trial and error must be clear and unique. Because trial and error is to gain business knowledge, not trial and error for trial and error;

Second, trial and error must find ways to finalize the variables that affect your trial and error conclusion, and strive to have the least impact on business cognition.

This is a case of 20 15, which was mentioned in my article "Talking about Operation with Murong Xue Fei Late at Night", and it is rather rough. Today, I combed my thoughts, told you in an orderly way, and released my optimized source table for Zhang Wei's signal operation. In addition, everyone should pay attention to one point: 20 15, the number of shared collections of WeChat statistics is put together. Unlike now, the statistics of sharing and collection are separated, which makes the method of locating content more scientific.

When I first took over a micro-signal content operation, the daily net increase of fans was negative, so how did I solve this problem? I still work hard on the content. Leaders are anxious to see the results, and I don't have time to make a trial-and-error strategy. I can only find the rules from historical data. Let me briefly describe it:

To clarify my business problem: the daily net increase of followers of WeChat is negative.

There are actually two ways to analyze the problem and solve it:

1, throttling, this is a matter of content selection: what kind of content should I do to meet the needs of users and keep them from running? That is, make a fuss from existing fans to reduce the loss of fans;

2. Racine, if the number of fans keeps decreasing, but I let the number of fans increase more than the number of fans decrease, isn't it positive to increase the number of fans every day? The increased powder number is closely related to the forwarding number, which is generally positive. The more forwarding, the more powder. And forwarding is closely related to content.

So my business problem is actually a content positioning problem. I need to find out what users like about WeChat content. My idea is simple.

As shown below, I made a reading order for this picture.

The number of senders, because it involves business privacy, I omitted it.

This table is the true face of one of my optimized source tables. I have a habit of doing one operation project at a time, ranging from the operation of the whole product to a navigation category, focus map, recommended location and so on. I will work out a table like this to find problems, summarize the rules and then guide the operation behavior.

It took me two days to manually collect all the historical push data of nearly 1 year into this table. What is more tiring is to classify these contents one by one. I roughly divided these contents into 10 categories. Then each article falls into 10 category (classified field in the table), and then this source data table comes out.

1, and the newly added number of followers (fuzzy) field is the number of powder added the next day, which is not accurate data and is for reference only. If your micro signal only sends one content every day, this data will be more accurate.

2. If you don't have a fixed number of historical tweets, you need to roughly calculate the ratio of forwarding to pink, and then attribute the pink to a single graphic according to the number of forwarding of a single article. This is a very heavy project, and you need to sort out the data one by one, and eliminate those abnormal values that are difficult to control.

3. If you don't want to follow the second rule to avoid the trouble of cleaning up the data, you can also make a short-term content trial and error strategy according to the previous rule, and you can send a graphic message every day to solidify this trial and error variable.

Then through perspective analysis, my micro-signal content operation strategy came out, as shown below:

If you look at the column "Determine the content type", you will find that the salary, workplace, inspiration, skills and other contents exceed the average number of powders. Problem solved: I focus on these four types of content. Then study the titles with high reading volume, and try to increase the reading volume of each article by about 10 working days. I successfully turned the daily net number of fans from negative to positive. "

Many people who do WeChat content operations send it when they see that the content read by others is good. In fact, this is not the most efficient. Other fans may have different content needs from yours. The same article can reach 100000+ on other people's micro-signals, but this effect may not be achieved on your micro-signals.

What did I do next when I corrected the daily net growth? After righting, this figure is in a healthy running state. I am not willing to be only an operation manager, so I set my next goal on the speed of powder increase. The operation strategy of adding powder is the next question, which will not be shown here!

Finally, in the fifth case, let's talk about the fundamental problem. What is the value of data in real work? Why should operations do data analysis? I summarize as follows:

1. Find the problem: the problem is the driving factor. We should constantly monitor product performance, analyze data, find out problems that affect business objectives, and eliminate them.

2, reduce costs: john wanamaker said: I wasted half of the advertising expenses, but I don't know which half. Analysis can effectively reduce costs.

3. Decision basis: In an enterprise, the general big boss needs data to make investment decisions, marketing decisions and strategic decisions.

Bottom line: analysis can monitor the effect, find problems, gain insight, control costs and achieve the goal of maximizing business value. So, am I right or wrong? I'm right, but this doesn't solve the real problem in most cases. We need grounding gas. As I mentioned at the beginning, in real work, after years of experience accumulation, the biggest role of data is to "communicate evidence." It's like you committed a crime, denied it, and then the lawyer took evidence.

This is the fourth function of data:

4. Communication and evidence: Analysis can provide evidence in court and make effective decisions on action plans. Operations should monitor and optimize the operating performance of each department, and then provide effective evidence to convince relevant business departments. Then achieve our goal of optimizing products and services.

Sometimes many products, if you experience them with the naked eye or by yourself, will know that there are problems, but the relevant business departments just don't change them and want data to speak. Then we'll try to find evidence.

For example, I have gone through countless times to optimize the conversion rate of the major processes of the recruitment website. I feel really troublesome, but the product hasn't changed. I need proof. There is also a second optimization situation. I didn't talk about the change in the registration process. In fact, the loss of that process is also very high. It is too complicated.

Then at this time, in order to find evidence, you need to find data, and you can use a funnel chart. For example, the following resume delivery funnel diagram.

See, there are too many loopholes in the four-step transfer process. The loss rate of each process node is "unattainable". So I quickly attached the meaning of analysis and guidance to the picture. In view of trade secrets, I won't let it out. In fact, I can see at a glance that it's just a matter of wording and euphemism. Descriptive words such as "very, too" should be used with caution, otherwise the product will get angry.

After giving the product, the product still struggled with the accuracy of the data, and then I thought of another method, the empirical data method came in handy. I want to see what the major processes of competitors are like. I decided to prove this comparison from the user's task load and the time it took to complete the task. What impressed me the most was the summer of 20 14, when hot people were particularly upset. I went there with my bare hands, experienced the three major processes of competing products alone, and recorded the data. The final result is as follows:

In the three major processes, the time for users to complete tasks and the number of actions are significantly higher than those of their competitors, which is not a good phenomenon, and also confirms the reasons for the high turnover rate mentioned above: there are many recruitment platforms, users spend time and energy to submit resumes, and the willingness to switch platforms will be high.

Finally, with the help of funnel chart tools and empirical data, I completed an optimized testimony. Submit the product for review.

Again, the data may not be accurate, but a mistake that can make people gain something is far better than doing nothing. Encourage everyone to be good.