When should employees be allowed access to data and analytics? (2023)

As business leaders look to get the most out of their analytics investments, democratized data science often seems like a perfect fit. Using analytics software with no-code and low-code tools can help almost anyone with data science skills. At best, this leads to better decision-making and greater autonomy and self-service in data analysis—especially given the dearth of data scientists. Coupled with reduced staff costs (fewer expensive data scientists) and more scalable customization to tailor analytics to specific business needs and circumstances.

However, in all the discussions about whether and how to democratize data science and analytics, one important point has been overlooked. Dialogue must be definedIfDemocratizing data and analytics, even redefining what democratization means.

The full democratization of data science and analytics carries many risks. As Reid Blackman and Tamara Sipes wrote in their articleRecent Articles, Data science is hard, and untrained "experts" can't necessarily solve hard problems, even with good software. Just because it's easy to click a button that produces results doesn't guarantee a good answer—in fact, the answer can have serious flaws that only a trained data scientist would know about.

It's only a matter of time

However, despite these caveats, the democratization of data science will continue, as evidenced by the data science boomsoftware and analysis tools.Thomas Redman and Thomas Davenport are supporters of the development."Citizen Data Scientistand even test the essential data science skills and abilities in each position.

However, the democratization of data science should not be taken to extremes. For businesses to be successful, you don't always need analytics at your fingertips. How many talented people don't get hired just because they don't have "basic data science skills"? This is impractical and very restrictive.

As business leaders seek to democratize data and analytics within their organizations, they should be asking “when” it makes the most sense. It starts with recognizing that not every “citizen” in an organization has the same skill set to be a citizen data scientist. Nick Elprin, CEO and co-founder of Domino Data Labs, which provides companies with data science and machine learning tools, told me in a recent presentation, “Once you start modeling, more complex statistical problems are often hidden on the surface ».

The challenge of democratizing data

Imagine a grocery chain that recently used advanced forecasting methods to adjust its demand plans to avoid overstocking (causing damage) or understocking (causing lost sales). Spoilage and inventory losses are small, but the problem of reducing them is difficult — considering all the variables of demand, seasonality and consumer behavior. The complexity of the problem meant the grocery chain couldn't leave it to citizen data scientists, but could rely on a real, well-trained team of data scientists.

As we discussed with Elprin, data citizenship requires “representative democracy”. Just as American citizens elect politicians to represent them in Congress (ostensibly to act in their best interest on legislative matters), organizations need the right representation from data scientists and analysts to speak on issues that others simply cannot have the necessary expertise.

In short, it knows when and how much data to democratize. I propose the following five criteria:

Consider the skill level of a "citizen":Citizen data scientists will remain in some form. As mentioned earlier, there simply aren't enough data scientists, and using this rare talent to solve every data problem is unsustainable. More importantly, the democratization of data is key to bringing analytical thinking across the enterprise. A well-known example isCoca Cola, which created a digital academy to train managers and team leaders, has produced graduates from the program who have worked on about 20 digital, automation and analytics initiatives across multiple locations in the company's manufacturing operations.

However, when it comes to predictive models and advanced data analytics that can transform the way an organization operates, it's important to consider the skill level of the "citizens". Sophisticated tools in the hands of a data scientist are additive and valuable. the very same tool in the hands of someone who "plays with the data" can lead to mistakes, wrong assumptions, questionable results, and misinterpretation of results and conclusions.

To measure the importance of an issue:The more important the problem is to the company, the more necessary it is to hire experts to analyze the data. For example, creating a simple chart of historical buying trends could probably be done by someone with a dashboard that displays data in a visually appealing format. But strategic decisions that have a significant impact on a company's operations require expertise and proven accuracy. For example, how much an insurance company should charge for a policy is fundamental to its very business model, so it would be unwise to delegate this task to non-experts.

Determine the complexity of the problem:Solving complex problems is beyond the capabilities of the typical citizen data scientist. Consider comparing customer satisfaction scores across customer segments (simple, well-defined metrics and lower risk) versus using deep learning to detect cancer in patients (complex and high risk). This complexity cannot be left to ordinary people to make arbitrary - and possibly wrong - decisions. Democratizing data makes sense when complexity and risk are low.

An example is a Fortune 500 company I work for that relies on data throughout its operations. I ran one a few years agoEducational programmeMore than 4,500 of these managers were divided into small groups, each asked to formulate an important business problem that could be solved in detail. Teams are able to solve simple problems using available software tools, but most problems arise precisely because they are difficult to solve. The important thing is that these managers areNOResponsible for solving these difficult problems, but working with the data science team. Specifically, these 1,000 teams identified at least 1,000 business opportunities and 1,000 ways that analytics could help the company.

Empower people with expertise:If an organization is looking for "targeted" information - that customer in fact, engaging in this type of low-level analysis can be a great way to make some simplified data tools available to those with domain knowledge (i.e. those closest to to the customer). Higher accuracy, for example with demanding and complex problems, requires special knowledge.

The most compelling case of accuracy is when high-risk decisions are made based on a certain threshold. For example, if an aggressive cancer treatment plan with significant side effects is to be implemented and the probability of cancer is greater than 30%, it is important to distinguish between 29.9% and 30.1%. Accuracy is important - especially in medicine, clinical operations, technology operations, and for financial institutions that have to deal with markets and risks, often with very small gains at scale.

Challenge the experts to look for bias:Advanced analytics and artificial intelligence can easily lead to decisions that are perceived as “biased”. This is partly challenging because the purpose of analysis is to differentiate. H. Making decisions based on certain variables. (Send this offer to the older man, but not the younger woman, as we expect them to behave differently in response.) So the big question is when this distinction is actually acceptable or even good—and when it's fundamentally problematic. unfair and dangerous to the company's reputation.

Example consideredGoldman SachsThe company has been accused of discrimination by offering women a lower credit limit than men on the Apple credit card. In response, Goldman Sachs said it did not factor gender into its model, only factors such as creditworthiness and income. However, one could argue that credit history and income are related to gender and that the use of these variables disadvantages women, who have a lower average income and have had fewer opportunities to build credit in the past. When biased outcomes are used, both policymakers and data practitioners need to understand how the data are generated and linked together, and how factors such as differential treatment can be measured. A company should never risk its reputation by letting citizen data scientists decide for themselves whether a model is biased.

The democratization of data has its benefits, but it also brings challenges. Giving everyone the keys doesn't make them experts, and getting the wrong information can have disastrous consequences. New software tools can make data accessible to everyone, but don't confuse this broad access with real experience.