Organizations are increasingly using artificial intelligence (AI) and machine learning (ML) to solve business problems and transform their operations. These technologies are affecting nearly every process across all industries and becoming imperative for competitive edge.
As AI and ML techniques continue to evolve, so do the requirements of the businesses that use them. When choosing an AI/ML strategy, it is important to ensure that the product roadmaps of your prospective AI/ML vendors align with your business’s future objectives in areas such as AI democratization, augmented analytics in BI, and cloud.
[ Need to speak artificial intelligence? Download our Cheat sheet: AI glossary. ]
Here are a few key questions to consider when choosing an AI/ML strategy and solution:
1. How will your solution save my team time?
There are plenty of open source tools and solutions out there that you can cobble together to get the job done for little or no cost. If you're going to shell out real money for a data science platform, you need to see tangible ROI.
Data preparation dominates most data science projects, turning your data scientists into expensive, and often unproductive, data engineers. To counter this, it’s important to choose a platform that makes data acquisition and transformation easy. These data engineering features should be seamlessly integrated into the modeling workflow.
Another area ripe for optimization and automation is model selection. The platform should provide access to a variety of models, open source and proprietary, and recommend the best model for the job.
Feature development is often a time-consuming and error-prone process. Choose a platform that integrates with a variety of "feature stores," like Feast or Uber’s Michelangelo to reduce duplicative work and drive consistency across data science projects and teams.
[ Want best practices for AI workloads? Get the eBook: Top considerations for building a production-ready AI/ML environment. ]
2. What types of AI use cases work best on your platform?
Typically, there are two main categories of use cases for data science: operational and diagnostic. Most machine learning platform vendors excel at one or the other.
Operational use cases
Let’s start with the operational use case. Developing the best algorithm for displaying the optimal set of products on your homepage is an example of an operational data science use case. In such cases, real-time performance, application and data integration, and CI/CD features are critical. For operational use cases, an expert data scientist or team of data scientists may work for weeks on developing and testing an optimal algorithm. Stability, performance, and accuracy are tantamount to success. These solutions are often long-lived and have a direct impact on a consumer.
Diagnostic use cases
In diagnostic use cases, the requirements are very different. Predicting sales for a particular SKU for an upcoming promotion is an example of a diagnostic use case. In this example, a citizen data scientist would benefit from an AutoML platform to help in model training, selection and visualizing results. These use cases typically involve a single data scientist working over the course of a few days at most. The most important features are automation, agility, and results that are "good enough." These use cases are typically one-offs with no customer-facing implications of operational use cases.
Be sure to choose a vendor whose solution best matches your most common use case.
3. How does your platform scale and how does it work in the cloud?
We live in an increasingly cloud-y world, where most of our enterprise data is generated, captured, and stored.
It’s essential that your machine learning platform is cloud-native so you can operate on the data without egress and with elastic scale. However, a cloud-native architecture may not be enough. Many of the public cloud AI/ML solutions operate only in their own cloud, so avoid lock-in by choosing a platform that can work across multiple clouds.
Data platform support
Another major consideration: Where does your data live, and how is it prepared? Choose a data science platform that can operate on a variety of data platforms, including data lakes and data warehouses. Since most of the time in a data science workflow is spent preparing data, the ML platform’s data-wrangling features are critical. This is especially important if your data science team lacks data engineering sophistication, which puts additional pressure on IT to acquire data and thereby limits self-service.
[ Read also: 6 misconceptions about AIOps, explained. ]
4. What level of expertise is required to use your platform?
This question touches on your team’s maturity and skill level when it comes to data integration and machine learning. Are you looking for a solution primarily targeted at the expert data scientist, a citizen data scientist, or both? The answer to this question will determine the style of machine learning platform you can support.
There’s a clear trend in the market for AI/ML democratization, so choosing a platform that can work for both personas is a good bet.
[ Get exercises and approaches that make disparate teams stronger. Read the digital transformation ebook: Transformation Takes Practice. ]
What to read next
Subscribe to our weekly newsletter.
Keep up with the latest advice and insights from CIOs and IT leaders.