How LinkedIn uses large data to make a simple transition to simplicity

For LinkedIn Business Data Analysis Department, large data analysis is not a high, complex and boring work, but a simple, efficient and practical art.

In the big Data age, the Business data analysis department to a company's important significance self-evident. At present, many companies use the data Analysis Department is "analysis on the report" Analysis methods, that is, the daily output is very cumbersome, complex, massive, detailed analysis of the report, but these analysis reports of the understanding and enforceability is not strong. And LinkedIn as a typical data-driven company, in the data analysis, but the opposite, using the "Report on the Analysis" method, to simplify, to the fastest speed in the large data gold mine to explore the most commercial value.

It is reported that since the establishment of the Business data Analysis Department in 2011, the sales revenue of LinkedIn has increased 20 times times, and not only that, the whole company has realized the data-driven automatic and fast business decision. Recently, the titanium media in the United States reporter also exclusive interview with the LinkedIn Business data analysis Department first staff and department director Simon Zhang, for Simon, business data analysis is not a high, complex and boring work, but a simple, efficient and practical art.

Data analysis structure: from pyramid to Diamond to spherical

LinkedIn is no doubt a company based on data, as reporters press time, its users will be more than 340 million people, the huge size of the user generated a huge amount of data, including behavioral data, identity data, social data and content data, etc. The key to LinkedIn's business model is how to tap the user's pain point from these massive data to launch marketable products and services.

The main function of the LinkedIn Business Data Analysis Division, founded on March 21, 2011, is to support other key departments of the company in various decisions through data analysis. Currently, 70 employees in the Business Data analysis department can support more than 4,500 employees across the company.

"Since its inception, every day, sales, operations, customer service, engineering, marketing, products and other departments of the staff will be to our department to ask a variety of questions, such as the user is satisfied with our homepage?" I want to promote a human resources product, which company should I sell? and so on. We started with manual data analysis, but the efficiency was so slow that we began to think about reforming the methods of data analysis. "said Simon.

Like most companies, LinkedIn initially uses a pyramid-shaped data analysis architecture, from bottom to top: Understanding related businesses and products, gathering useful data purposefully, understanding the fundamentals of data analysis tools and how to use them, analyzing data, drawing conclusions and making decisions.

In these steps, the key point of the analysis of different levels of data is the two steps in the middle. "To understand the data analysis tools you use, many analysts are not paying much attention because they think that writing data analysis tools is a traditional IT department thing, but in fact, it is critical that you get a deep understanding of how the analysis tool works and how well you can use it properly, is also the key to distinguishing between good and bad analysts.

Another key point is the data analysis of the process itself, in my years of work experience, the industry agreed that good data analysis is good at simplifying, good data analysts are good at the most simple and clear way to present the core value. Simon told Titanium Media.

That's why, in an era when everyone is talking about big data, LinkedIn's two most important requirements for data analysis are "speed and value".

Only fast enough to form a scale to produce the value of size, and the traditional pyramid-type data analysis architecture allows analysts to spend too much time (85%-95% time) in the middle and lower part of the pyramid, so by the end of 2010 to early 2011, LinkedIn began to think about turning the pyramid into a diamond-shaped structure.

"The main way to become a diamond-shaped structure is to create automated tools that replace the traditional pyramid-level work, to automate the work of all possible elements of the pyramid, especially the lower and middle ones," Simon says. Each diamond-shaped data analysis structure, we will turn it into a pyramid again, and then optimize into a diamond, if each diamond area only half of the original pyramid area, after many conversions and iterations, the entire data analysis efficiency will be greatly enhanced. ”

It is reported that after the pyramid-shaped data analysis structure into a diamond, the LinkedIn Business data Analysis Department to optimize it again, the diamond structure into a spherical structure, forming a closed loop, "Our business Analysis department has developed hundreds of internal staff for closed-loop spherical products, Each product can be a closed loop process from product to data collection to decision making, which means that each spherical product can not only achieve efficient analysis and decision-making, but also form a closed loop, automatic upgrades and iterations.

(Analysis decision: from three months to one minute)

For large data analysis, LinkedIn believes that efficiency is the first rule, and that producing real value in the shortest amount of time is more important than exhaustive analysis. And through the hundreds of of internal-oriented products developed by the Business Data division in recent years, employees in all of the LinkedIn departments can really feel the speed of their work.

Case one: Market and sales Team support product--merlin

For every product salesperson in LinkedIn, when he receives a task to sell a product, he has at least one of the following key questions:

In the traditional manual data analysis mode, for a particular product, the sales staff would like to know the above issues and make a customer use of the sales PPT at least 2 weeks to 2 months, and today, through the LinkedIn market and sales team to support product Merlin, Sales people only need to log on to the system, enter their own name and need to sell the product name and other basic information, Merlin can automatically collect sales staff background data and network data, so as to quickly generate a more accurate sales plan, from the input of basic information to the generation of sales plan only need a minute time, The sales staff can even get the system tailored for his sales ppt.

"The traditional 2 months of research may not be able to get a precise solution, and now it only takes a minute, we have recently put Merlin on the sales staff's mobile phone to support the sales staff, currently more than 3,000 people in LinkedIn are using the Merlin system," "The automated generation of sales solutions makes it easy for us to recruit salespeople without requiring too much training," says Simon. ”

Case TWO: Product testing team Support Products--a/b testing System

For the LinkedIn Test department, in a traditional environment, it takes at least 3 months to complete a test, and now, with A/B testing System, it takes only a minute to get a few key metrics out of each of the 650 pointers in each test, and to suggest improvements. To improve the testing products with the highest efficiency.

"In the traditional environment, more than 10 people 3 months to complete the test work, currently only need a minute, the current A/B testing system can support 2000 internal tests per day, each test tracking more than 650 pointers, through the extraction of the most critical indicators to improve the product advice," "We do not pursue perfection, we only pursue a little bit of progress every day, in fact, as long as we can make progress on the basis of 1% per day, the results of one year's progress will be amazing," Simon said. ”

Case three: Customer service team Support product--voices

For LinkedIn's customer service department, how to measure user satisfaction has always been a problem, as customer service staff can only gather some information from unstructured data such as user messages, but it has not been resolved how to turn loosely-cluttered unstructured data into measurable and improved structured data.

Until the LinkedIn Business data Analysis Department launches the customer service team to support the product--voices, the traditional unstructured data only takes a minute to generate the analysis, for example, if the service team wants to know if the current LinkedIn customer is satisfied with the homepage, It only needs to enter "homepage" in the voices, can instantly obtain the structured visual user satisfaction data. Of course, the instantaneous results of the back must be rigorous and advanced algorithm.

(from writing model to writing robot)

In fact, as there are many more examples, the LinkedIn Business Data Analysis Division has since its inception hundreds of of such products, every day for every LinkedIn staff to improve efficiency and effectiveness of efforts. Not only that, each of the above products can also automatically learn the habits of employees to ensure that the next time they use the same software to respond faster. "For LinkedIn employees, each of our products is personalized for them. ”

The result of automation of scale, of course, is a significant increase in efficiency and effectiveness, which has been reported to have increased by 20 times times since the establishment of the Business data Analysis Department in 2011, as well as a significant increase in the efficiency of all other links.

"We will not be content with this, the main task of our department is to write model, since 2013, our department began to design model robot, which further accelerated the company's data analysis in the automation and efficiency." "said Simon.