How Non-Technical Teams Can Use Predictive Analytics (Part 2 of 2)
Predictive analytics is not just for data scientists. We explain how business analysts and other non-technical users can get the most out of their organization's data.
- By Shanif Dhanani
- September 22, 2020
Every company has data. However, there's a big gap between raw, unstructured data and real, actionable insights. Predictive analytics is a way to bridge this gap and turn data into knowledge about the future. In the first part of this series, I demystified predictive analytics. In this article, I'll explain how to put it to work by ordinary business analysts and non-technical end users.
Complex programming languages such as Python and R are inaccessible to non-technical teams, but there are other options available for everyone to derive the most value from their data. We'll examine three steps you need to take:
- Know where your data is and what its quality is
- Create a predictive analytics thesis
- Build the model
Step 1: Know Your Data
Before you can start building predictive analytics models, you need to understand your data. Some companies have billions of data points, but you can build valuable models with very small amounts of data as well. In any case, you need to at least know the location and quality of your data.
Every digital tool your company is using creates data, and most have features that let you export this data in a comma-separated values (CSV) or similar file format. Common tools include Google Analytics, Mixpanel, HubSpot, Shopify, Hootsuite, and Amazon FBA. The tools you use largely depend on the industry you're in. Excel and Google Sheets are also ubiquitous and perhaps the most common sources of data for predictive models.
In terms of quality, predictive analytics models require data that is accurate, relevant for the problem you're trying to solve, complete, and up-to-date. If you have a considerable amount of missing data (either individual fields or entire data records), which is easy to spot in an Excel spreadsheet, then you likely can't build a model from it. Further, you can't predict the future based on outdated data. Finally, the data needs to be relevant to your problem. It seems obvious to say that if you want to predict financial fraud, you'll need data of financial transactions, but many users try to make data "stretch" to fit analysis where it's not appropriate.
The more data you have, and the higher quality it is, the better.
Step 2: Create a Predictive Analytics Thesis
Armed with an understanding of your data, you need to form a thesis about how predictive analytics can add value to your organization.
The most successful companies have a highly specific use case that's meaningful to their enterprise and adds value. Finding a "quick win" will also bring positive momentum for future data science projects and help ensure stakeholder satisfaction.
Often, the thesis is relatively simple. For example, a marketing team might want to use predictive analytics to identify customers likely to churn so the organization can decrease churn and increase customer lifetime value. A SaaS company may want to use predictive analytics to surface trends in website visitor data to determine how to increase conversions. An HR team may want to analyze (and then reduce) employee turnover.
Notice that in each of these examples, there's a key performance indicator (KPI), such as churn, conversions, or turnover. If you're like most enterprises, you have a KPI -- which is what you will focus on with predictive analytics.
Step 3: Build the Model
With data and a thesis (and associated KPI) in hand, it's time to build your predictive analytics model. There are many tools non-technical teams can use to upload your dataset, select the KPI column, and create predictions. If you've ever attached a file to an email or uploaded something to Google Drive, you can use these tools.
For example, with Apteo (the company I co-founded), you simply upload your data file and select your KPI. You'll see predictive insights about your KPI, such as what attributes contribute to it. If you're predicting employee attrition, for instance, you may find that attributes such as high overtime and low income contribute the most significantly to attrition.
To make a prediction, you can enter new data, and the software will output a predicted KPI value, all in your browser. You could calculate the odds of attrition for a 37-year-old employee who lives eight miles from the office, has rated employee satisfaction at 6 (on a 10-point scale), and earns $4,500 a month. The data you input to make a prediction should be in the same format as the data you uploaded.
You can also create many predictions at once rather than just on one new data point. You can generate a prediction from the API, allowing you to create predictions in your own app, though this is more technical.
A Final Word
Predictive analytics is for everyone, even non-technical users. By knowing your data, creating a predictive analytics thesis, and building a model, your organization can gain a competitive edge. Historically, predictive analytics was largely done by Fortune 500 companies and those with highly technical teams. Now, anyone can take advantage of it.