We recently discussed the need to shift the way artificial intelligence (AI) projects are managed in order to make them successful and productive. In that post, we discussed how “agile” software methods are not suitable and can in fact impede the development and application of AI-driven projects. The truth is that the agile methodology simply cannot handle the volume and velocity of data needed for an AI project to be effective. We suggested that the “analytics solutions unified method for data mining,” (ASUM-DM) approach is a better approach to managing these projects.
Another approach that has grown out of DevOps to guide project management in the age of AI is DeepOps. While that may sound like some sort of clandestine military operation, it is in fact a highly appropriate project management toolkit for AI-driven projects. Let’s take a closer look at what it is and how it can impact your projects.
DevOps: The Origin of DeepOps
Traditionally, DevOps approach to project management meant putting together developers (i.e. “Dev”) and operational people (i.e. “Ops) from an organization to solve problems. The idea is to have a team that can interface on a project, bringing complementary skills to the table, and ensuring the project is more successful. What’s more, DevOps dictates a standardized approach to feature development and deployment that involves development, testing, and integration, including a method for version control. Using the this approach, software can typically be deployed faster, has a lower failure rate, has a shorter lead time for fixing bugs, and can recover faster in case of a new version causing a crash.
The problem is that DevOps-driven projects typically focus on software feature deployment, while data science projects require a focus on the creating models of learning based on data; this is what is known as a DeepOps process. That is, projects that are data driven—e.g. process improvement projects, marketing and sales cycles optimizations, etc.—require a more global view in order to succeed. Data science projects use boatloads of data and need to manage and compare different versions of models of that data. Success isn’t the delivery of a feature or product, but rather a data model that can provably work over and over. And that’s where DeepOps and AI-driven approaches to project management excel.
DeepOps: An Approach for the AI Age
DeepOps is ideal for today’s high-speed world because it takes a different approach to project management that allows for the use of AI to model, test, and deploy project milestones at a much faster rate. In this case, the “deep” comes from “deep learning,” a branch of AI. To manage not only the code running the model, but also the different versions created in order to achieve an applicable model, there are specialized tools like DVC. Additionally, the acceptance criteria and a suitable threshold value to test a model as successful has to be defined in the DeepOps process and tweaked iteratively while the cycle is in operation. An example for these criteria may be the accuracy and a threshold value can be defined as a minimum of a 99 percentage of accuracy on the test data set.
What this approach presupposes, however, is that the data used to perform the tests are considered high-quality. That is, in DeepOps, the more data you have and the cleaner it is, the more accurate the tests will be. It can only do this with high-quality data, but that will also shift as projects mature, generating larger quantities of high-quality data. In order for machine learning to have an impact on the speed at which a project achieves milestones, it is necessary for the system to be constantly learning.
Ultimately, the end result is bringing AI-driven software products to the market faster, along with updating them and introducing new features faster without sacrificing stability. Contact us if you’re looking for help with applying a data science approach to your business challenges.