Can data-driven operations really lead to rapid user growth?

As the saying goes, "Good wine needs no bush"; as the saying goes, "Even good wine needs no bush." Later, as the saying goes, it doesn’t matter whether the wine is good or not, as long as you find an accountant with amazing abacus skills, that is, a data scientist, the wine will sell well. This is called using data to drive rapid user growth, or “Growth Hacking” in jargon.

The first time I heard the term "Growth Hacking" was at a big data conference last year (I forgot which conference it was, because now all conferences are called big data conferences). When I was full of negative energy and was about to go on stage to make some discordant remarks, I suddenly found a former Facebook engineer talking about "Growth Hacking" on how to use data analysis to drive user products.

In fact, the engineer’s speech was quite pertinent. He did not promote data as a panacea. He seemed to have specifically emphasized that “only by making a good product can the retention rate be improved.” But later, the story was distorted, and some people in the Internet industry began to believe that data analysis is the magic weapon to turn products into gold. As advertised in street tabloids many years ago: "Acupressure through clothes can increase the size of breasts." As a "coder" who has been dealing with data for so many years, I would feel a thorn in my side if I didn't say anything on this issue. Today we are going to talk about whether data analysis can really drive the rapid growth of user products?

For the Chinese market, where data awareness and methods are between the Jurassic and Cretaceous periods, emphasizing the role of data is generally enlightening. Looking at the data is always more useful than simply praying to the God of Wealth. But then again, data is not the God of Wealth. It doesn’t mean that you can gain eternal life if you believe in him. A sage once said, "It is better to believe in countless things than to believe in numbers." In what scenarios is data operation useful? How can we make it useful? This is a question that the high-quality readers of this official account need to understand clearly.

Three Axes for Data-Driven Operations

There have been many books and lectures recently on using data to drive operational growth. In fact, the principles here are not mysterious. In most data operation scenarios, the methods can be summarized into the following three methods:

1. Establish a user conversion funnel

The so-called user conversion funnel is how your business lures a user step by step. Here are a few examples so you can understand:

1. Advertising: Display -> Click -> Conversion
2. Games : Download -> Activate -> Keep -> Pay
3. Picking up girls: shake -> date -> hold hands -> kiss -> have sex

No matter which of the above businesses, it can be broken down into a series of stages. After each stage, only a portion of users remain. Accurately recording data at every link in the funnel in order to analyze and optimize the throughput rate of each link is the infrastructure of data operations.

2. Find problems with multi-dimensional data reports

A common pain point in data operations is knowing that the pass rate of a certain link in the conversion funnel is low, but not being able to find a way to improve it. A common solution is to break down the data into different dimensions and observe them separately, which can often reveal problems in the product or system. If multiple dimensions can be flexibly combined to observe data, it becomes a data cube. Although the following figure has nothing to do with the funnel data of Internet product operations , the principle is the same.

For example, you find that the click-through rate of your ad is low , and then find out that the click-through rate on the Chrome browser has lowered the overall statistics. In this case, you need to dig deeper into the cause on the Chrome browser. It may turn out that your Flash ad material is directly blocked by Chrome.

This method of using multidimensional data reports to locate and find problems is quite effective. It is actually efficient debugging, but it is still a "supported" strategy.

3. Use A/B testing to guide product evolution

So is there any data-driven “offensive” strategy? Of course there is also the option of developing possible improvement directions for multiple products, putting them online, and letting actual data decide which ones will go up and which will go down. This A/B testing method is often the magic that everyone imagines can optimize a good product without doing anything, and it is also one of the foundations of the "numbers determine the outcome" theory.

When it comes to the A/B system framework, it is not a simple matter. How to establish an experimental framework that combines accuracy and efficiency is worth writing a separate long article, so we will not discuss it here.

The above three tips are very important for operating a product well. However, if you think that you can create great products through data analysis by mastering this kind of data thinking, then forget it.

Problems that data operations cannot solve

How users choose and evaluate a product follows completely different rules in different fields. Simply put, we can divide products into rational products and emotional products. For example, 3C e-commerce is a typical rational product, while clothing e-commerce is a relatively emotional product. Although computational advertising and recommendation systems have similar technology stacks, the former is a rational product, while the latter is much more emotional.

For rational products, data is one of the most critical optimization methods because the problem objectives are stable and easy to quantify. Take advertising products for example. The purpose of advertisers using them is to obtain higher profits (of course, this profit may be long-term or short-term), rather than to obtain spiritual pleasure or pleasure. Therefore, when the input-output ratio of two advertising platforms is very different, customers will not care about which one provides a better user experience, but will choose the one that makes more money without hesitation.

But when it comes to emotional products, it is far from that simple. I remember when WeChat first became popular, a large number of Internet analysts who had switched from non-staff personnel in various industries spoke out and analyzed: Why is WeChat the terminal product of human social interaction ? Why are those who still use QQ obstacles to the wheel of history? But last year, roughly the same group of analysts were discussing: Why are post-90s users shifting from WeChat to mobile QQ? So which is better, QQ or WeChat? This question has different answers at different times and for different users, and it is impossible for us to give a universal quantitative target for this type of mobile IM product. In the operation of emotional products, since it is difficult to give a definite optimization goal, the role that data optimization can play has a ceiling.

Bottlenecks in data-driven operations

So, under what circumstances will data-driven operations encounter obvious bottlenecks? Generally speaking, there are the following aspects:

1. Product innovation direction cannot be obtained through data

A few years ago, there was a very popular gaming company called Zynga. It is said that the boss of Zynga does not encourage innovation, but instead pursues the "take-it-and-use-it" approach, that is, copying other people's game ideas and using his own data operation system to quickly surpass his competitors. What kind of data operation system? To put it simply, it is a lot of A/B testing. Design says: The grass must be green. The product manager said: No, the green and red data have the final say! So, they really divided the traffic into two configurations: red grassland and green grassland. If the data shows that users on red grassland pay more, then they will turn all the grass red and let the botanists go to hell! Relying on such a system, Zynga once dominated the top three game rankings on Facebook for a long time. What happened later? The answer is clear - who knows Zynga now?

There are of course many reasons for Zynga's decline. However, it must be said that the product operation idea of ​​data-only theory is also one of the driving forces: no matter how mature your data testing system is, it can only do trivial tricks like painting the grass red or green. The real innovative direction of the game, such as new modes, new plots, and new designs, cannot be achieved just by having a data operation system, and it is even impossible to judge the superiority through the data system. Take the iPhone as an example. Its brilliant product features such as large-screen interaction, Multi-touch, and App Store all come from Jobs’ belief and insight into the product, rather than the results of market and demand research.

Even for short-term product improvements, relying entirely on A/B testing won’t work. We can pick out plan A with better results based on the data, but if your alternative plans only include X/Y/Z, then A will not come out no matter how hard we try. Someone asked: What if we list all possible product options and let the data choose? Product operation is not like playing mahjong. There is always a number when the cards are combined together, but the truly potential product points require product managers with systematic and innovative thinking. Only by spending the time others spend drinking coffee and going to the toilet on thinking hard can they discover them. Moreover, even if the product points are clearly listed like mahjong tiles, when the product factors and directions become numerous, due to the existence of "dimension curse", we will not be able to accumulate enough statistical data to make decisions within a limited time.

In fact, the most important growth driver of user products has already been mentioned here, which is the insight and creativity of product managers.

2. Long-term user feedback is difficult to judge through data

The Facebook engineer gave a vivid example in his speech: Facebook has insisted on using a rigorous A/B testing framework for many years to decide whether a new feature should be adopted by the online system. But what is the result? He said that in fact, the homepage of the Facebook PC version has not had any major upgrades in three years. In fact, this result itself is thought-provoking: Has Facebook become the terminal product of human social networks, and has it been optimized to the global optimal point and cannot be improved any further?

The conclusion is obviously not like this. I even think that among the many revision plans that were abandoned in those three years, there may have been some versions that were mistakenly killed. The existing A/B testing framework can only observe data performance in a relatively short period of time. Without persistence in product beliefs, it will be difficult to wait for the long-term trends and conclusions to bear fruit.

I have heard a pattern from traditional paper media magazines: generally speaking, the magazine will be criticized in the first few months after a revamp. Then a few months later, the new edition might bring a significant increase in circulation. This can also serve as evidence of short-term data defects.

3. Gaming scenarios cannot be decided by data

Gaming scenarios are very common in Internet products. For example, you may know that there is an Explore & Exploit problem in Internet products, which is to use a part of relatively random traffic to explore unknown space, and use the other part of traffic to make decisions based on statistical optimality. Friends who are familiar with this field know that if we have two E&E strategies, we cannot determine their pros and cons through A/B testing. As for why, I suggest that you sort out the background knowledge of E&E on your own and just treat it as a question to think about!

In addition to E&E, there are many other gaming scenarios, such as mechanism design issues in advertising (that is, how to formulate the rules of the bidding market); and game operation strategies with a certain sociality, which in principle cannot be effectively A/B tested by simply splitting traffic.

Of course, understanding these game-like problems requires in-depth product insights and macro-thinking skills. Many product managers who pursue a rough, fast and aggressive approach would rather pretend that such problems do not exist.

What is the key to rapid user growth?

Although data is an indispensable accountant, it is not the magic weapon for user growth. So, what methods can truly drive rapid user growth for a product? From the successful products in history, the following ideas are worth noting:

1. Make truly excellent products

To put it bluntly, the most important issue for rapid user growth is to do everything possible to brew good wine. A good product is a truly good product, and this is the foundation that all products and operations should pursue. Due to some of the data system problems mentioned above, the intuition and judgment of excellent product managers may not be consistent with short-term data performance. At such a crossroads, it is more reasonable in the long run to adhere to product principles first and data feedback second.

It should be noted that truly excellent products are generally not conceived based on analysis of the existing market. Therefore, there is no ready-made statistical data to support this. The user growth of Pokeman Go, which has been very popular recently, is staggering. However, I don’t think it has established a complete data operation system. Even if it has, it is not the key to its user growth.

The rapid user growth of almost all great products comes from the endogenous driving force of "good products", such as Google , Facebook, and iPhone in the past, and certain taxi-hailing and live-streaming apps in the recent past.

2. Find strategic promotion channels

What are strategic promotion channels? Simply put, they are those channels that offer “cheap prices and sufficient quantities”. The rapid growth in users of many Internet products is often due to the great efforts put into this aspect.

For example, the rise of a certain cross-border e-commerce company has a lot to do with their early strategic purchase of large amounts of traffic from Facebook. And today. Facebook advertising prices have been raised so much that such channel opportunities no longer exist.

In China, many non-BAT mobile Internet products are growing rapidly, and most of them rely on pre-installation as a strategic channel. It is pre-installed by manufacturers, operators and solution providers. Many of the companies that actively spent money on products when pre-installation costs were only one-fifth of today's have grown into large companies today.

Keeping a keen sense for undervalued strategic channels is critical to the rapid growth of your product.

3. Use viral communication methods

There are two methods of viral transmission: one is infection and the other is pyramid selling.

The so-called infection is the induction and bundled installation: through a product that already has a large installation volume, another product is brought to the user's terminal through coercion, inducement or even covert operation, which is often a good way to achieve more quickly, better and cheaper results. Sometimes, this can even work well in ordinary advertising channels. For example, various mobile cleaning or security products often directly threaten users with "your phone's memory is too low" or "your phone is very risky" when they are advertised . Users will feel their anus tighten and install the product immediately.

Everyone is familiar with pyramid schemes. After the emergence of social networks, MLM-style promotion plans are easier to implement. Although the various group-buying models and forwarding lottery models cannot be said to be pyramid schemes in the strict sense, their essence is to mobilize the masses to fight against each other. If such methods are used well, they will definitely have miraculous effects. Anyone who uses them will know it.

4. Build brand awareness

The Internet community has insufficient understanding of the importance of this point. For a product to stand on its own, and to make money even if it remains as weak as Ge Da Ge, core user awareness is crucial.

In the Internet world, people often believe in direct effect communication, but pay insufficient attention to brand building. However, the recent success of companies such as OPPO has made people begin to re-examine the role of brand and user awareness. In my opinion, the user growth brought about by successful brand building can be rapid and healthy.

Having said so much, it is not to say that data-driven operations are not important, but that there are actually more important things to achieve rapid product growth. All Internet practitioners today must master the correct data-driven operation methodology , but also avoid blindly becoming data-worshipping believers. To make a product that is both popular and well-received, although data operations are indispensable, we still have to focus our core energy on the two key points of "making good wine" and "moving it to the entrance of the alley."