Broken Lifecycle

The data science lifecycle and why projects fail

Andrew Engel
The data science lifecycle and why projects fail

Driving Value with AI

Numerous reports and studies have highlighted the importance of AI and machine learning to business. In a report released in 2019 by Accenture, 84% of C-suite executives state that AI is necessary to meet their growth objectives. At the same time, three quarters of those executives struggle with how they build these capabilities to deliver on their promise.

In addressing both this need and challenge, executives often focus on three things. First, companies invest in data warehousing platforms to store and make accessible the data to drive these initiatives. Next, data scientists and data engineers are hired to begin building these solutions. FInally, the organization purchases data science platforms to enable these data scientists and data engineers. Still, organizations fail to derive value from their AI efforts.

Most (over 80%) never make it to production. Those that do can take six to nine months to get to production, if not longer. And even those that make it to production may not produce the return on investment executives expect. A survey of 1,200 firms by ESI Thoughtlab, found that AI investments were returning only 1.3% on investments and 40% of all projects failed to show any positive returns at all. Even the leaders reported a mere 4.3% ROI and break even reportedly took 17 months on average.

The Data Science Lifecycle

Because most companies have such low return on investment and high risk of failure, it is important to get the process correct. It is not simply enough for a company to store the data, hire the data science team and provide simple tools. Businesses must ensure a process is followed that can allow them to identify the most valuable projects, understand the risks and evaluate the likely outcomes. There are numerous data science frameworks that companies can use to understand the process including CRISP-DM and SEMMA, but without a focus on the business problem first, organizations will struggle to see significant ROI.


  1. Define Use Case
  2. Data Acquisition


  1. Data Extraction + Exploration
  2. Data Preparation
  3. Feature Engineering


  1. Model Training
  2. Model Evaluation


  1. Model Deployment
  2. Model Activation

As companies look to make investments in AI, alongside their infrastructure and team investments, they should consider these steps and understand how to implement policies that can help drive value. In next week's blog, we will discuss how to prioritize these steps.

Sign-Up For Your Free 30-Day Trial!