What skills does a good data product manager need to master?

1. Very familiar with the company's business and trends.

Therefore, it is necessary to understand the company's business model, strategy, business process, various indicators to be assessed, and the business meaning behind the indicators. I don't know enough about this.

2. Understand data analysis.

Good data PD, even if you don't do data PD, you should be a data analyst. One of the most important tasks of data PD is to turn data analysis into a replicable automation system. Although there are data analysts around you, you should also be clear about the business problems, what data you want to see, or what problems have occurred or will occur in the business when the data appears. This should be in line with the best data analysts.

3. Understand data warehouse and business intelligence.

Behind these two keywords is a huge system. I'm afraid that my short half-year job transfer time is too short, although I can explain the architecture of a business intelligence product to others. Although a number of terms such as summary, drillthrough, measurement, indicator, dimension, slowly changing dimension, hierarchy, attribute and dashboard will be thrown out, multi-level knowledge drillthrough is not supported, and I don't know where to analyze the reasons when I encounter abnormal problems. Fortunately, colleagues with data warehouses can learn more. Well, there is no upper limit.

As a discipline, business intelligence originated in the 1990s. Its starting point is to help users get better decision-making information. The initial motivation of business intelligence is to provide users with self-service information acquisition methods, so that users can get customized reports without relying on IT departments. (quoted from the book information dashboard P4 1). Nowadays, in addition to providing information, business intelligence is more important to lower the threshold for users to obtain data and improve the real-time performance of data. From the direction of lowering the threshold for users to obtain data, we can do many things, such as how to design an information dashboard. How to display data in a more friendly and intuitive way (data visualization)? How do users access offline? How to realize the active sending of early warning data? At this point, it requires little effort.

4. Proficient in data product development process. Data development+product development.

The ultimate goal of data PD is to make data products. Let's take it apart here. First, the data product itself is also a product that can be realized online by users. Since it is a product, the whole set of research and development ideas of the product is not much different from ordinary products. Who are the users, what are their needs, what feature lists are needed to meet the needs, what are the resource evaluation and priority of each feature list, and what is the product life cycle? This is product development. Then it is a data product, which means it has more requirements than ordinary products. Outside the data core, a variety of function lists are needed, such as subscription, search, customization, SMS interface, email interface and so on. But the core of data also needs a data development process.

For example:

Source of data-is it sufficient and stable?

Data PD needs to know enough about the current business processing system construction and the accumulation degree of data sources to judge whether the construction time of data products is appropriate. Inappropriate opportunities lead to repeated work of the project team and the birth of incomplete data products. Data products are used to support monitoring, analysis and decision-making, while the positioning of business processing system is to improve work efficiency and liberate staff. The data collected by the business system may not meet all the analysis requirements. For example, leaders may want to analyze the detailed reasons for the rise of a large number of returns, but at present, the business system does not require users to select reasons or only input options instead of standardization when applying for returns, and the employees responsible for returns can only register reasons in excel instead of inputting them into the system. Therefore, the data that the demander wants to see may not be available, and the data pd is needed to drive the data source to collect in the opposite direction.

The design of analysis model-whether the analysis model is good or not actually determines the success or failure of data products.

In the project, the data analyst of BI can take this responsibility, or it can be led by the data PD, and more will be confirmed by both parties. The content is mainly data analyst, function evaluation and priority, and the project planning and coordination is mainly data PD. Therefore, data PD should know better than data analysts whether the requirements required by data analysts can be realized and what is the commercial value behind them, maintain a smoother cooperative relationship with data development and product development than data analysts, and be able to evaluate feasibility and resources with the help of power.

Sometimes, it's not that we don't have data, but that there are too many data and we don't know how to read them. If you just throw a bunch of data to users, it is hard to imagine how users will interpret it. When we used to do interaction design, there was a popular saying: treat users as fools.

And the data platform, because it may need a certain threshold of use, so it is not realistic to think of it as a fool who doesn't understand the Internet, so we have to think of it as "the user is a fool who doesn't understand data". They may also be able to experience something through a string of data, but if it is the upward trend line of the refund rate, they may experience more-after all, it is intuitive from top to bottom. Then think again, if you add a warning line to this line, they will know when the data is abnormal. Then, imagine what he wants to do when he finds that the data has been rising since July 12. Does he want to know which industry has gone up Will he want to know which channel has gone up? Then, you need to provide industry and channel options or compare with him.

Then he asked the person in charge of this industry, would the person in charge want to know which supplier or type of goods has gone up? So how can we integrate these dimensions and levels and make them very convenient for users to use? The construction of analytical model is very important. It can also be said that the analysis model is the most valuable product of pre-demand analysis. The analysis model should include several points:

Division of themes:

What topics will the whole analysis be divided into? For example, sales may be divided into sales trend and composition analysis, industry ranking, commodity ranking and so on.

Measures and indicators:

Analyze the algorithms and definitions of metrics and indicators involved in the topic (this usually produces a document on the definition and description of indicators and dimensions).

Size:

From which dimensions should we look at these indicators and metrics, such as time and channels, and should these dimensions be screened or compared?

Drilling:

Are these dimensions hierarchical? Do they need to be drilled? For example, channels can be drilled into channel types, industries can be drilled into sub-industries, and commodity categories can be drilled into commodity leaf categories.

Output:

What kind of chart does the analysis need to show?

ETL development of data

The process of data cleaning, conversion and loading occupies more than half of the resources of data product development, and the irregularity of data source will lead to a greater occupation of this resource. For example, the same supplier code, one system is called supplier code, the second system is called supplier code, and the third system is called supplier ID. These three systems are also the company's systems. Although this situation is strange to think of, the reality also exists. Although ETL development is completed by DW development engineers, as a data PD, how can you lack understanding of these tasks, the problems fed back by ETL engineers and the potential risks of the project? Moreover, more often, under the condition that the data is not standardized and unified, the data PD needs to drive the business system to establish data standardization, whether functionally or directly, such as the second person in charge of data entry, and establish a set of entry specifications. These tasks seem to have nothing to do with data PD, so we can say: there is no way, this is a problem with the data source, not our function. However, users have the right to choose whether to use or not to use your data products. If the data provided by data products is not credible, it is undoubtedly self-destructive. Once users don't trust the data, it's hard to keep them. Even if there are many "powerless" excuses, we can't just sit idly by.

Optimization of front-end interaction and experience

Although the content is well defined, how to divide the information level, how to divide the columns and how to design the user's behavior path with so many metrics, indicators, dimensions and drills? These are not important work areas for data analysts. But an interaction designer? In view of the fact that many data product projects may not have interaction designers, data PD should encapsulate the content and design various functions such as information architecture, page layout, charts and so on. Design, and then write detailed functional requirements documents and deliver them to four types of developers: product development, front-end development and data development, and front-end display development.

The function description document of data products, except the product development part, describes the "content", that is, the analysis model. In addition to theme, measurement, dimension, drilling, filtering and output chart type, some contents need to be defined in detail to "sorting method" and other details, and the specific situation is analyzed in detail.

Environment, technology, tools

Maybe to make an ordinary product, you can describe the requirements clearly, confirm the feasibility with the product development engineer and accept the resource evaluation. However, data products are subject to the deployment environment and the selected tools, such as Oracle, IBM's Cogos and SQL Server. Other products don't know. We use Oracle Bone Inscriptions BIEE Company. So as a data PD, do you need to know what functions BIEE can provide? Look at the documents and ask others, and you can't know that you can't. In addition, we should gradually understand BIEE's bad temper, impossible functions and insurmountable difficulties. We also need to continue to understand this and continue to learn.