Why do data analysis plans always fail?

Why do data analysis plans always fail?

Powerful data analysis is the top priority of digital business-it all begins with the practice of intelligent data governance and the emphasis on quality and environment.

Executives mostly talk about the value of ordinary data, but Michele Koch, director of enterprise data intelligence at Navient Solutions, can calculate the actual value of company data. In fact, Koch can use real dollars to calculate the income increase and cost reduction brought by various data elements of the company. Therefore, she understands that there is something wrong with Navient's data, which may damage its bottom line. For example, an error in a key data field in a customer profile may mean that a company cannot handle a loan at the lowest cost.

"Because money is involved here, we need a data quality control panel to track all the actual and potential values we track." Koch said.

In addition, Barbara Deemer, chief data administrator and vice president of finance, said that an early data-related plan within Navient, an asset management and business processing service company based in Wilmington, Delaware, revealed risks. In 2006, the initiative focused on improving the quality of marketing data, and generated a return on investment of $7.2 million, in return for increasing the loan amount and reducing operating expenses.

Koch said that since then, Navient executives have pledged to support a strong data governance plan, which is a key part of a successful analysis. Navient's governance plan includes long-recognized best practices, such as standardizing the definition of data fields and ensuring data cleanliness. It assigns ownership to about 2600 enterprise data elements. Ownership belongs to the business domain originally generated by the data domain, or going to a specific data domain is a more indispensable business domain for the process.

The company has also developed a data quality plan to actively monitor the quality of the site to ensure that high standards are continuously met. The company also launched the Data Governance Committee (2006) and the Analytical Data Governance Committee (20 17) to solve persistent problems or doubts, make decisions throughout the enterprise, and continuously improve data operation and how data supports the company's analytical work.

Koch said, "Data is very important to our business plan and new business opportunities, and we want to focus on improving the data that supports our analysis plan."

According to the data governance of data governance solutions companies Irving and UBM, most executives believe that data governance driven by compliance, customer satisfaction and better decision-making is essential. However, the report found that nearly 40% of the organizations surveyed did not have a separate data governance budget, and about 46% did not have a formal data governance strategy. The survey results are based on 1 18 respondents' questions, including chief information officer, chief technology officer, data center manager, IT personnel and consultants.

In view of these data, experts say that it is not surprising that many enterprise data projects have weaknesses. Here are seven data practices for this kind of problem.

Integrating data together, but not really integrating.

Anne Buff, vice president of communications for the data governance professional organization, said that integration is one of the most important challenges in the field of data and analysis today.

Indeed, many organizations put all their data in one place. But in fact, they don't integrate all the parts from multiple data sources. Therefore, the blll smith from one system has nothing to do with the data generated by other systems (and its complete changes), which brings many incomplete definitions to the business.

* * * Stored data is different from integrated data. You must have a way to match records from different sources. You need to make sure that when all this is combined, it should create a bigger point of view and use something as a connection point.

Different data integration technologies can achieve this, and it is very important to select, implement and execute the right tools to avoid too much manual work or doing the same work repeatedly.

In addition, integration is becoming more and more important because data scientists are looking for patterns in data to gain insights that can bring breakthroughs and competitive advantages.

"But if you can't merge data that has never been merged before, you won't find these patterns." Anne Buff, SAS information business solution manager in Cary, North Carolina, said.

Unrealized business units have unique requirements.

Yes, comprehensive data is essential for a successful analysis program. But some business users may need different versions of data. One form of data cannot meet the needs of everyone in the whole organization.

Instead, we need to consider data supply, that is, to provide data needed for business cases determined by business users or business departments.

Take the different needs of a financial institution as an example. Although some departments may need to integrate data, fraud detection departments may want their data scientists to use unconstrained data so that they can search for red flags. They may want to find someone at the same address and apply for multiple loans by taking advantage of subtle changes in their personal identity information.

"You will see similar data elements, but there are some variables, so you don't want these differences to be too big and make it too clean," explained Anne Buff. On the other hand, she also said that the marketing department of financial institutions hopes to have the correct customer name, address and appropriate target communication.

Only recruit data scientists, not data engineers.

As companies try to go beyond basic business intelligence to predict and standardize analysis, as well as machine learning and artificial intelligence, they need to improve the professionalism of data teams. This makes the status of data scientists concerned, and data engineers are equally important. They need mobile phone data to complete the work of data scientists, but so far, this has received less attention in many enterprises.

Bain company. Lori Sherer, a partner in the company's San Francisco office and a leader in advanced analysis and digital practice, said that this situation has been changing. "We have seen the demand for data engineers increase about twice as much as the demand for data scientists," Scheler said.

The Federal Bureau of Labor Statistics predicts that the demand for data engineers will continue to grow rapidly in the next 65,438+00 years. From 20 16 to 2026, the American economy will add 44,200 jobs, with an average annual salary of135,800 USD.

However, like many key positions, experts say that there are not enough data engineers to meet the demand-which makes IT departments just start to recruit or train recruitment positions. Only recruit data scientists, not data engineers.

Save data in its initial state instead of managing its life cycle.

In the past decade, the cost of storage has dropped dramatically, making it easier to store a large amount of data, which is longer than ever before. Considering the current capacity and speed of data and the increasing demand for data analysis, this seems to be good news.

However, Penny Garbus, co-founder of Eagle Consulting, a consulting firm in Apollo Beach, Florida, said that many people welcome the value of having a lot of data, which is often a good thing. Gabbs said that too many companies have held data for too long. She said: "Not only do you have to pay, but if you store it for more than 10 years, then this information is likely to be out of date." "We encourage people to write a timetable on it."

Garbus said that the deadline for data varies not only by organization, but also by department. The inventory department of a retail company may only need relatively new data, while the marketing department may need years of data to track trends.

If so, it needs to implement the architecture and deliver the data in the right time frame to the right place to ensure that everyone's needs are met and the old data will not damage the timely analysis program.

As Garbus pointed out, just because you have to save (old data) doesn't mean you have to put it in your core environment. You just need to have it.

Bain company. Lori Sherer, a partner in the company's San Francisco office and a leader in the company's advanced analysis and digital practice, said that this situation has been changing. We see that the demand for data engineers is about twice that of data scientists. The Federal Bureau of Labor Statistics predicts that the demand for data engineers will continue to grow rapidly in the next 65,438+00 years. From 20 16 to 2026, the American economy will add 44,200 jobs, with an average annual salary of135,800 USD.

However, like many key positions, experts say that there are not enough data engineers to meet the demand-which makes IT departments just start to recruit or train recruitment positions.

Focus on quantity, not target relevance

Steve Escaravage, senior vice president of Booz Allen Hamilton, an IT consulting firm, said: "We are still building models and using the most effective data for analysis, not the most relevant data."

He said that organizations often mistakenly believe that they should capture and add more and more data sets. They think that "maybe we haven't found anything yet, rather than doubt the correctness of the data"

Considering that many institutions judge fraud by analyzing a large amount of data to find anomalies. In an important activity, leading institutions will also analyze more targeted data sets to produce better results. In this case, they may pay attention to those individuals or institutions that generate certain transactions, which may indicate trouble. Alternatively, when analyzing the patient's results, medical institutions may consider the doctor's shift time when providing patient care.

Escaravage says organizations can start by creating a data wish list. Although this process begins with business, "the mechanism of obtaining information and making it available belongs to the domain of CIO, CTO or chief data officer."

Provide data, but ignore the source of data.

Today, a big theme is "analysis bias", which will distort the results and even lead to wrong conclusions, thus leading to poor business decisions or results. ESCA rure said that in the enterprise analysis program, the problem of deviation exists in many different fields, including how to deal with the data itself.

He said that usually, the work of tracking the source of data is not good enough.

"If you don't know this, it will affect the performance of your model," Escaravage said, pointing out the invisibility of data sources and how data sources make it more difficult to control deviation.

"We have a responsibility to know where the data came from and what happened." There is so much investment in data management, but there should also be a metadata management solution.

Provides data, but does not help users understand the context.

ESCA crude said that it should not only have a powerful metadata management program, it can track the source of data and how it runs in the system, it should provide users with insight into some history and provide context for some results produced through analysis.

"We are very excited about what we can create." We think we have good data, especially data that has not been analyzed, and we can build a new model about the value of these data. "However, although the analysis methods in the past five years are surprising, the results of these technologies are less explained than when business rules are applied after data mining in the past, and the data is easy to explain."

Escaravage explained that the relatively new deep learning model provides insights and actionable suggestions. However, these systems usually cannot provide useful or even critical background for the best decision. For example, it does not provide information about probability and data-based certainty.

ESCA crude said that a better user interface is needed to provide this environment.

"The technical question is how people interact with these models. From the perspective of transparency, it is very important to pay attention to UI/UX. Therefore, if someone sees the recommendation of the artificial intelligence platform, to what extent can they deeply understand the potential data sources, and so on. " He said. "CIOs will have to consider how to build such transparency in their systems."