Audio version of the article
Artificial intelligence (AI) and machine learning technologies have come of age, and have been helping organisations improve operational efficiencies and enable continuous innovation enterprise-wide. However, IDC’s AI Strategies View 2020: Executive Summary (1) notes that only over a quarter of AI/machine learning initiatives globally are in production, suggesting that many enterprises are not fully benefiting from the technology.
In order to succeed with machine learning, Indian businesses need to rapidly implement and scale machine learning models enterprise-wide across potential use cases. The deployment of effective machine learning requires implementing complex, iterative end-to-end workflows from data to models to outcomes. This entire process can be quite tedious and would also create new and unique governance challenges—from ensuring internal and external (governmental) regulations to privacy and security concerns pertaining to end-users or customers whose data may be collected and processed.
The building blocks of operationalising machine learning
Platform selection is a crucial step for operationalising machine learning. The right platform will not only let organisations set up an effective machine learning lifecycle but also help them manage the quality, security and scalability elements of their machine learning model.
With an enterprise data cloud, organisations can gain a holistic view of their data, free up workloads when needed to extend the bandwidth of critical applications, and run analysis from the edge to AI. It also delivers an integrated set of security and governance technologies built on metadata, rendering a single, consistent way to manage and maintain data access and governance policies across all users, analytics, and multi or hybrid cloud.
Once organisations have full control of their data, they can then put in place newer disciplines, like machine learning operations (MLOps
), so that there is better coordination between data scientists and operations professionals and an improved machine learning lifecycle. Borrowing from the core principles of DevOps, MLOps reduces the time to push models into production by decreasing issues and friction between teams to enhance processes like model tracking, versioning, monitoring, and management.
For MLOps to deliver such value, organisations will have to ensure that their machine learning service can provide:
- Model cataloging and lineage capabilities to allow visibility into the entire machine learning lifecycle, which eliminates silos and blindspots for full lifecycle transparency, explainability, and accountability.
- Full end-to-end machine learning lifecycle management that includes everything required to securely deploy machine learning models to production, ensure accuracy, and scale use cases.
- An extensive model monitoring service designed to track and monitor both technical aspects and accuracy of predictions in a repeatable, secure, and scalable way.
- New MLOps features for monitoring machine learning models’ functional and business performance. This includes detecting model performance and drift over time with native storage and access to custom and arbitrary model metrics and measuring and tracking individual prediction accuracy to ensure models are compliant and performing optimally.
- The ability to track, manage, and understand large numbers of machine learning models deployed across the enterprise with model cataloging, full lifecycle lineage, and custom metadata.
- The ability to view the lineage of data tied to the models built and deployed in a single system to help manage and govern the machine learning lifecycle.
- Increased model security for Model REST endpoints, which allows models to be served in a machine learning production environment without compromising security.
These capabilities will enable the repeatable, transparent, and governed approaches necessary for scaling model deployments and machine learning use cases. By ensuring that data behind machine learning models are explainable and interpretable, organisations will gain greater confidence to weave machine learning into their operations while being assured that corporate standards and compliance are met.
For instance, Robi Axiata
leveraged an enterprise data platform together with a data science workbench to build an effective data lake, which provides insights into every customer interaction using applications with enriched data feeds such as campaign management and sales apps. Its data science team activated its AI investment to develop and tune machine learning models using nearly 20TB of accurate data daily to improve business outcomes as well as eliminate operational challenges across multiple domains and use cases. Based on some of the use cases deployed, Robi expects to see a 4 percent improvement in customer churn, an 8 percent rise in upsell recommendations, and over 5 percent improvement in quality acquisition.
Given its many practical applications that drive real business results, machine learning is undoubtedly a game-changer for businesses. The technology can automate processes, uncover new insights, and help enhance or create products and services to deliver a better customer experience.
However, harnessing the full value of machine learning
requires full operational transformation as it calls for the ability to build and develop machine learning models
, as well as deploy and operationalise aspects of those models. As such, Indian organisations should look for an end-to-end platform
that enables agile experimentation and production machine learning workflows
with enterprise-grade governance capabilities built in.
This article has been published from the source link without modifications to the text. Only the headline has been changed.