🚨 To read the full findings from this research, visit The Machine Learning State of Play 2024 white paper.

Are the challenges of deploying machine learning (ML) overshadowing its true potential in the modern workplace? Through our recent white paper , we spoke to 500+ developers who have experience working with ML systems to gain an understanding of the pain points faced by developers when using ML solutions.

Through this blog, we will explore the complexities of ML deployments, uncover strategies to overcome common challenges, and highlight the pivotal role of open source technology in shaping the future of ML accessibility and success.

What are the challenges of machine learning deployments?

The research conducted as part of our recent white paper found five core challenges associated with machine learning deployments: high project failure rate, time consumption, lack of machine learning expertise, disparity in adoption, and lack of investment.

Top 5 challenges of machine learning

1. High project failure rate

A growing number of respondents from our research highlighted the high rate of project failure as a core challenge for machine learning deployment. This comes with 53% of respondents confessed to abandoning between 1 – 25% of their projects, and almost a quarter of the developers, 24%, have abandoned between 26 – 50% of their projects.

One of the primary reasons for the high rate of project abandonment can be attributed to the complexity and expense involved in setting up and running ML projects. Before even beginning to generate ML insights, developers are required to configure various aspects of a complex infrastructure, including machine resource management, monitoring, and feature extraction. While these steps are crucial, they are also time-consuming and resource-intensive, often leading to a situation where only a small portion of the system consists of usable ML code. Such a scenario can be particularly daunting for organizations that lack the necessary expertise or resources, leading to projects being abandoned midway.

2. Time-consumption

Smaller organizations with limited resources are seeing significant challenges surrounding the amount of time required for the successful implementation of machine learning projects. Our white paper revealed that 48% of developers believe that ML projects demand an excessive amount of time. This perception is even more pronounced in smaller companies, where 61% lack a designated ML team, compared to 57% in larger organizations with such teams in place.

Research from Google revealed further insight into this challenge whereby for every eight hours of ML engineering, an organization might need to invest 24 hours in feature engineering, 32 hours in data engineering, and a staggering 96 hours in infrastructure engineering.

where is time spent in machine learning

This model suggests that a mere 5% of the total time spent is actually dedicated to ML engineering. Such a heavy investment of time and resources can render ML impractical for many, particularly smaller companies, and contributes significantly to the high failure rate of ML projects as teams struggle to allocate the necessary time and resources for successful deployment.

3. Lack of machine learning expertise

From the results shown in our white paper, we learned that 47% of organizations do not have a designated ML team. Whilst participants noted that they were willing to make an effort to build these teams, 65% of survey participants identified the recruitment of more ML-skilled employees as a key factor in easing ML adoption.

47% of organizations do not have a designated ML team

Further research from SAS highlights this issue, with 63% of decision-makers reporting a significant skills shortage in AI and ML. Moreover, it was found that over a third of organizations do not provide any ML-specific training or skills development, exacerbating the problem. This lack of training and development opportunities means that even when organizations are willing to invest in ML, they are hindered by their inability to cultivate or attract the necessary talent. As a result, many ML projects will either not start or fail to reach their full potential due to the absence of skilled professionals who can navigate the complexities of ML technologies and drive these projects to success.

4. Disparity in adoption

When comparing larger organizations with smaller organizations, it was revealed that there is a significant disparity in the adoption rate for machine learning. This disparity is not just a matter of scale but a reflection of deeper underlying challenges that different sizes of organizations face in implementing ML. Our research highlights that while 57% of developers working in large companies (with 500+ employees) reported having a designated ML team, 61% of small companies (1-50 employees) do not have such a team.

To properly democratize access to AI, it is essential to offer solutions that can be deployed quickly and work for an array of developers. This needs to be applied to all organizations, not just the ones that can afford a dedicated team of ML specialists.

5. Lack of investment

According to our research, 65% of organizations report a lack of investment in the tools required for effective ML production. This highlights a gap in the commitment to the foundational aspects of ML deployment, which can be critical for success.

Without the proper tools and training, organizations struggle to initiate, sustain, and successfully complete ML projects. This leads to higher failure rates and a general reluctance to engage in further ML initiatives. For smaller organizations, the challenge is bigger as they often lack the financial flexibility to make substantial investments in both technology and training. This disparity creates a technological divide where only certain organizations can fully take advantage of ML.

How can open source help make machine learning accessible?

Whilst addressing the challenges of ML, open source solutions emerged as a crucial strategy for making ML more accessible and effective. With two-thirds of organizations already leveraging open source tools for ML deployment, the trend underscores a shift towards community-driven resources known for their flexibility, customization, and cost-effectiveness.

Key points from our research found the following points highlighted in reference to the impact of open source on ML:

Aspect Details
Time Efficiency 73.2% of respondents recognize open source solutions as key to speeding up the ML development lifecycle. This is essential for organizations without dedicated ML teams.
Collaboration and Innovation It promotes a community-driven approach for sharing and innovation - platforms such as GitHub are widely used for ML projects.
Flexibility and Customization Open source tools offer adaptability to meet specific business requirements, which will allow organizations to create tailored ML solutions.

What is Civo doing to help?

At Civo, we are actively addressing the challenges in ML deployment by utilizing an approach that centers on accessibility and efficiency. With the use of our offering, Kubeflow as a Service, we are making ML more accessible to a diverse range of organizations, regardless of their size. By providing the essential tools and frameworks, we enable leaders to embark on their ML journey without the burdensome need to set up and manage complex operations and infrastructure.

Our focus is on what truly matters for developers: simplifying processes to enhance time efficiency and productivity. By prioritizing these areas, we aim to make a significant impact on the day-to-day work of developers.

You can find out more about some of our recent work to tackle ML deployment here: