Machine-learning analytics tools will reduce the complexity and increase the adoption of the Internet of Things (IoT), according to a new report from ABI Research. The firm estimates that revenue from machine-learning data analytics tools and services will hit nearly $20 billion by 2021 as machine-learning-as-a-service (MLaaS) models grow.

Machine learning is the study of algorithms that learn from examples of experience, according to Ryan Martin, senior analyst with ABI. “The challenge of hard-coded rules is that they don’t adapt well in real-world environments. Hard-coded rules can address some stuff but won’t hold up at scale.”

Machine learning, according to Martin, can automate a lot of processes, such as repetitive tasks. And the technology will be key to enabling analytics at the edge of the network, which will make IoT systems more robust and cost efficient.

For example, when analytics are done at the network edge instead of sending all the data to the cloud or to the core of the network to be analyzed, it makes the flow of information more efficient, and that cuts down on network latency, which can reduce costs. “Machine learning is foundational to enabling edge analytics,” Martin says. “It facilitates a more distributed network architecture.”

In addition, by using machine learning, companies can add a human element to technologies and make data the differentiator instead of the technology. Martin says this is why many cloud-infrastructure providers like Amazon, Google, IBM, and Microsoft are investing in machine learning and why some are open-sourcing their deep learning software.

Earlier this year, Amazon released a library called DSSTNE on GitHub under an open-source Apache license. Likewise, Google seeded the open source community with its cloud machine learning library called Tensorflow.

Martin says this fundamental shift toward open-sourcing these machine-learning libraries is a sign that these companies are realizing they need more collaboration and they need to engage the developer community in machine learning in order to get better tools. “There’s been a fundamental shift from proprietary technologies as an advantage to proprietary data,” Martin says. “This is really exciting.”