Affordable access to specialized hardware and large historical volumes of data leveraging cloud storage and compute have enabled enterprise ML projects. Similarly, open source software libraries have reduced risk in delivering value with ML systems(2).
Nonetheless, there are additional considerations over the traditional software project lifecycle and even the biggest players in the ML arena have experienced big failures: https://www.lexalytics.com/lexablog/stories-ai-failure-avoid-ai-fails-2020
A well-cited NeurIPS paper catalogues a set of ML system-specific challenges & considerations(3).
Although many organizations can benefit from ML solutions, the cost of retaining talent to maintain these systems becomes prohibitive.
However, emerging practices around maintaining ML infrastructure have found support in powerful libraries like TFX(4).
In some of the high profile cases linked above, failure arose when a system designed in the lab interacted with adversarial users in the wild.
TFX provides tooling to help detect drift and training-serving skew which may arise from such interactions(5).
In other instances, a bias in data collection contributed to undesired results.
Here too, TFX shines in surfacing these relationships with components to assess these biases and even the ability to constrain your model subject to some fairness criteria: https://www.tensorflow.org/tfx/fairness_indicators
Still others cite challenges in managing data infrastructure as a hindrance to widespread adoption of AI projects(6).
TFX is well-documented and comes with powerful templates and sensible defaults to streamline transferring project management and to help reap the benefits of CI/CD.
With a powerful framework supporting the ML project lifecycle, it is easy to stand up a pipeline in days that would otherwise take months to develop.
Ultimately, costly engineering talent can concentrate on product improvements, eliminating vectors for introducing technical debt.
With frameworks like TFX, your living ML solution can continue to evolve with your business.