6 mistakes to avoid in data analysis

After reading this article, I feel that it is good to summarize some points of data analysis and share it with you. In order to generate real value from data analysis, or to make the business side and management feel real value, a lot of things are actually needed:

1. There must be data, and indeed enough data. is normal data accumulation.
2. Whether the analyst can fully understand the business side's problem, please pay attention to it, not understanding it.
3. Really know the current company resources, and can combine the actual situation of the company when giving suggestions and plans, that is, grounding.


An operations director once told me that an analyst did an analysis for me. To accomplish the operational goals, the most important thing is to increase traffic, because it is too difficult to increase the conversion rate and there are too many things involved. The increase in traffic is to introduce traffic, and then analyze each channel and measure each traffic.


He said: However, I just don't have money, and if I have money, I want you to tell me..., what else do I need to analyze with you!!!


Well, let's talk about it and read this article!


1. Go well Too fast, no time to look back


. People in startups seem to have been under the mantra: "Either die or die." They are so eager to develop products that they often don't have the imagination that users will The specific use details of the product, how the product is used in which scenarios, which parts of the product are used, and the main reasons why users return to use the product again. And these questions are difficult to answer without data.


2. You don’t record enough data


It’s useless to show your team the summarised data. You can't analyze the invisible hand behind changes in data without a breakdown of changes accurate to the day or even the hour. If it is only extensive and intermittent statistics, no one can interpret the impact of various subtle factors on sales or user habits.


At the same time, data storage is getting cheaper. Doing a lot of analysis at the same time is not a high-risk thing, as long as you buy enough space, there is no risk of system breakdown. So logging as much data as possible is never a bad thing.


Don't be afraid of large quantities. For start-ups, big data is actually a relatively rare thing. If you are (luckily) in the early stages of this kind of trouble, Porterfield (this article) recommends using a platform called Hadoop.


3. In fact, your team members often feel like they are blind This silly mistake happens all the time. And if your business builds a data platform that everyone can use for self-help from the beginning, to answer the most important questions in their work, you can avoid the tragedy mentioned above. 4. Storing data in inappropriate places


Many companies think it's enough to throw data at Mixpanel, Kissmetrics, or Google Analytics, but they often ignore which members of the team can actually interpret the data's inner meaning. You need to constantly remind everyone on the team to understand the data more and make more decisions based on the data. Otherwise, your product team will just blindly develop the product and hope to hit the hot spot, and whether it will succeed or fail will be confused.


Example:
One day you decide to attract new users with viral marketing that is common in the market. As you'd expect, the number of users popped up. But at this time you will encounter new confusion: you cannot measure the impact of this marketing method on old users. People can get caught up, sign up as new users, and then get bored and stop using it. You may be overpaying for attracting a bunch of worthless users. And your product team may still be complacent that this product-damaging marketing ploy was a success.








Let's look at a proper example first. Porerfield mentioned that one of his clients has created a data analysis framework that integrates resources from NoSQL, Redshift, Kitnesis and Looker. This framework not only captures and stores its own data on a very high scale, but also withstands millions of monthly hits and lets anyone query the data they want. This system can even allow novice users who do not understand SQL language to truly understand the meaning of data. And in the world of data analytics, basically if you don't know SQL, you're screwed. If you always have to wait for engineers to get the data out, you're getting yourself into trouble. The algorithms or software that engineers build without understanding the requirements are often a torment for users, because their use of data is often no longer on the same level as the former.


You need to keep all your data in one place. This is the most critical principle.


Let's go back to the hypothetical company from earlier. They did viral marketing after viral marketing, but didn't put user activity data in the same framework, so they couldn't analyze how one campaign was linked to another. Nor were they able to make a data analysis comparison across day-to-day operations and during events.


Many companies send data to outsourcers for storage, and then act as hands-off shopkeepers. However, these data are often transformed into other forms in the hands of outsourcers, and a lot of processes are required to convert them back. These data are often data related to your website or product during certain publicity campaigns. Combined with daily operational data, you can dig out which activities contributed to user conversions. The way in which the user journey is analyzed in combination with daily operational data is crucial. But shockingly, despite the critical importance of all operational data at any given time, many companies are shy about capturing and recording them. More than half of the companies Porterfield has seen look at day-to-day operational data separately from activity data. This seriously hinders the company's correct understanding and decision-making.


5. Short-sighted


Any good data analysis framework must meet the needs of long-term use at the beginning of design. Granted, you can always tweak your frame. But the more data is accumulated, the more expensive it is to make adjustments. And often after adjustments are made, you need to record both the old and new systems to ensure that no data is lost.


Therefore, it is best that we design the frame on day one. One of the simple, crude and effective methods is to put all available data on the same extensible platform. There is no need to waste time choosing an optimal solution, just make sure that the platform can hold all the data that may be used in the future, and it can run cross-platform. Generally, such primitive platforms can last at least one to two years.


6. Oversummarizing


While this problem is more common for companies with big data analytics teams, startups are better off avoiding it. Just imagine how many companies only record how many sales per minute on average, rather than how much money is sold per minute? In the past, due to limited computing power, we could only summarize massive data into a few points. But at the moment, these calculations are not a problem at all, and everyone can record operational data accurate to the minute. And these precise records can tell you a ton of information, like why conversion rates are rising or falling.


People are often self-indulgent in making a few beautiful icons or PPTs. These summaries look exciting, but we shouldn't base our decisions on these superficial summaries, because these pretty summaries don't really reflect the essence of the problem. Instead, we should pay more attention to the extreme values ​​(Outliers). (Source: Data Ocean)