Machine learning and artificial intelligence are now moving from the realm of research into adoption. Machine learning adoption offers immense benefits which can provide any organization with a competitive edge — if executed well. Technological adoption requires a pragmatic and collaborative approach across the organization driven by agile practices.This also comes the need for trusted data sources, organizational change management, iterative revalidation practices and measuring the business value of the technology insertion.
In part one of this series on machine learning (ML), we defined machine learning, delved further into the various types of machine learning models, and described their common applications. This article focusses on the tactical execution steps and organizational modifications required to make the ML dream a reality.
5 Steps to Machine Learning Implementation
Establishing machine learning within any organization requires planning and collaboration. As with any technology insertion and/or transition, it starts with a vision and moves on to execution followed by continuous monitoring and improvement. The basic steps to building an ML implementation plan are described in five simple steps below: VDOCR — Vision, Data, Organizational alignment, Change management and Revalidation.
1: Establish a Vision
Establishing a vision is perhaps the most important step in implementing a new technology. It is not any different for machine learning. Business and IT must work together to establish a vision and define clear objectives for an ML implementation. The objectives could be as simple as improving the accuracy of the fraud detection system all the way to improving overall operational efficiency — but it needs business and IT alignment and the agreement to work towards a common goal.
Without a clear understanding of what you want to achieve, it’s hard to measure success. You'll find the most common use cases by looking for places that are labor intensive and repetitive such as image classification, tuning/optimizing your data center operations, configuration management, and systems patching/updating. This step also includes establishing key performance indicators to measure the business value of the program.
2: Define Data Requirements
Data is perhaps the single most important element required for the success of a machine learning implementation. Collecting, storing and feeding the system vast amounts of reliable data is the key to improving the accuracy of machine learning algorithms. Data management processes need to be established for:
- Providing an initial set of historical data to train the ML processing system.
- For continuous data insertion to train and improve the accuracy of the model.
Beyond the initial model-training phase, infrastructure will be needed to collect new data from which to learn over time. Data requirements need to be established not only for collecting and storing data but also to ensure that the available data is reliable and secure and is available in a steady stream for continuous improvement.
Related Article: What Data Will You Feed Your Artificial Intelligence?
3: Establish Roles and Responsibilities
Any successful technology implementation requires integration across the organizational landscape that is strategically led by a robust management function, clear establishment of roles and responsibilities and cultural integration. Begin with the creation of integrated solution teams with representatives from IT, marketing, sales, and other required stakeholders that meet regularly during the project to review progress and ensure adequate coordination with their respective groups.
4: Set Up a Change Management Process
Technology insertions often fail due to the lack of adequate change management processes. Change management and training are two of the key aspects of delivery and acceptance of any large-scale modernization effort, and ML implementation is no different in that respect. Change management includes looking at current business processes and re-engineering them based on the updated business model. In addition, training programs that cover mission objectives, product features as well as the newly created business processes are imperative to create collective support and awareness for the mission and its objectives as well as to increase efficiency and use.
Related Article: Change Management: The Key to Successful Digital Transformations
5: Establish Monitoring and Revalidation
Gauging the success of an application and whether it needs changes can be established by measuring its business value. To ensure that ML models remain relevant and ultimately result in business value, they need to be continuously updated, retrained and validated. To achieve this, organizations need to ensure that any ML implementation plan includes the ability to update its criteria based upon evaluated outcomes and to incorporate improved and increasing amounts of data. Also important is to measure how the ML algorithm affects broader business goals.
For example, Amazon is continuously refining its prediction algorithms based on the past purchases of its customers. Similarly, Netflix improves its ability to provide customized content to its consumers based on the content they watch. Moreover, New York Times has even developed an ML system to ascertain the emotions evoked by news articles with the goal of helping advertisers’ places ads more effectively.
Imagine an article that changes its content based on what its consumers wants to read or a movie that changes its story based on the likes and dislikes of its viewers. Sounds eerie! Get ready for it — because it is coming ....
Get Started With Your Machine Learning Strategy
Businesses need to carefully plan and manage technology disruptions and ML is no different in that respect. If you want to get the most out of your business data and automate processes, the time is ripe for creating an ML strategy in your organization. Following the simple VDOCR (Vision, Data, Organization, Change and Revalidation) model will help your organization take its first step towards an ML implementation that considers cultural implications and delivers business value.