Your manager wants to talk about the strategy for implementing machine learning in your organisation? He has probably heard about the benefits of this technology at a recent conference about industry innovation. Up until now, advanced analytics has not been used in your company, no one knew where or how to apply it. But the benefits are increasingly being talked about in IT, so you want to use it too. But how do you organise it when no one has dealt with it before?
Many companies face a similar dilemma. Hearing everywhere that AI and Machine Learning will change their organisations, bring profits or savings, they want to invest in such solutions. They want to become a data-driven enterprise. The beginnings are difficult, the first projects often fail, leading to the belief that Machine Learning does not work. How can this be avoided? For a start, it is worth answering 5 important questions!
Why?
Why do we want to implement machine learning in an organisation? If the only motivation is to have 'artificial intelligence', it is likely that our project will fail. Advanced analytics should support business problem solving, not be introduced just to boast about it.
Why have these technologies not been used before? And why do we want to reach for them now, instead of opting for much simpler solutions? Answers to these questions will help identify the reasons why it is the right time to start implementing Machine Learning in your company.
Answering the question of why you want to use machine learning solutions in a chosen business process will allow analysts and data scientists to thoroughly understand the problem and thus prepare an optimal model.
What?
Understanding what motivates you to introduce advanced data analytics, it is important to gather additional information. It is worth considering where to start and what the biggest business problem is at the moment. What can be automated or improved in the decision-making process? What factors can improve the quality of life for customers or employees? Answers to these questions will allow us to discover the potential of using artificial intelligence.
So where do we start applying AI? Usually, where a key decision point arises and a choice has to be made. A place where a human has to analyse a lot of data and make a decision, where there is often not enough time to do so. It is worth choosing the point in the business process where we will gain the most – improving the quality of the decision or speeding up the process – to ease the burden on employees or customers. For example, chatbots can off-load call centre staff of simple questions, and stocking algorithms can prepare the optimal order for a shop or warehouse.
Who?
In the next stage, it is important to determine who will regularly use the results of the algorithms and who will be responsible for evaluating them. There should be people in the company who are responsible for quality control of the models, overseeing the architecture of the entire analytics system, and selecting appropriate solutions.
In the business area, it is necessary to find people responsible for defining the requirements and expectations for the algorithms. These people should clearly define, for example, what a forecast requiring sale is. They should also include people who will indicate what business metrics they want to use to evaluate the results, such as time savings of consultants on the helpline after the implementation of a chatbot.
In the technical area, people with knowledge of various algorithms and data science methods will be needed. In addition, they will need to be able to implement and maintain models in a production environment, while taking care of good MLOps practices. Often, support from outside the company will be required, as it can be difficult to maintain such specialists internally.
Where?
Now that we understand why the organisation is seeking to use algorithms, who will be evaluating them, and which problems are worth starting with, it is important to consider where to store the results of the models and in which systems to use them. You may need to adapt existing systems or introduce additional components. This will assess whether our technical infrastructure is sufficiently prepared or additional technologies will be needed. In addition, such a process will signal to the IT department that work on changes or development of systems and applications may be required.
When?
The last question provides information on the deadlines for the implementation of machine learning and when the algorithms are expected to deliver results. Every company develops its framework schedule, which is essential for monitoring the progress of the work whether it concerns changes to systems, processes, the implementation of new products, or the use of new technologies. Within this schedule, key milestones are defined that identify main activities and expected results. Then you should designate those responsible for overseeing the implementation, which is another answer to the "who?" question.
Summary
In today's world, many companies think that implementing machine learning is just about collecting data and hiring data scientists to prepare models. This is important, but we need to keep in mind many other aspects that need to be discussed and analysed internally. This will help us avoid rushing into Machine Learning projects, which often end in failure.
