We are often asked “How do I get started?” so we decided to detail the very first stage of a Machine Learning project. Even if Artificial Intelligence and Machine Learning technologies can be applied in such different use cases, there is only one reliable approach to such a project.
Way too many times we’ve heard executives regretting that they haven’t allocated more time to strategize before kicking off with a Machine Learning project. To give it a better shape and understand the goals and objectives could be the difference between failing and succeeding. Even if you only experiment with this technology within your organization, knowing where you want to get with it will avoid unnecessary costs.
Strategy will make the difference
Defining strategic goals is one of the crucial steps that will determine the project’s overall success.
Rather than being overwhelmed by the AI idea, start by looking into the problem that you need to solve within your organization or within your team. You can have the shiniest new technology available, but the results will be worthless if you are solving the wrong problem. With this in mind, you might discover that the problem can be solved with other technologies, not necessarily using Machine Learning. At this point in time, Machine Learning is merely a tool in your toolbox which you will use only when suitable.
Start by asking yourself what problem your product or your organization is facing. Although there are plenty of AI/ML solutions in the marketplace, you need to know what you are looking for because one size does not fit all. If you have a good understanding of the problem, you probably already tested several other solutions. Which is good, because your experience will help optimize the Machine Learning model by formulating a hypothesis.
Proceed with framing the problem to make sure everyone involved in the project is clear about the direction and the objectives.
State the objective in a clear and concise way. What would you like your Machine Learning model to do? For example, you might want to predict how popular a product will be among your customers.
State the outcome that you are going after. It needs to reflect what you care about in your product. Continuing the above example, your outcome could be to have Product Recommendations to only suggest products that are relevant to your customers. Another possible outcome could be marketing cost optimization, since predicting products’ popularity should optimize your targeting and budget allocation.
Define how you will know if your project succeeded or failed. What are the indicative metrics that will evaluate the system’s performance? For example, a metric for product recommendations could be the sales volumes of suggested products predicted by the model. If the volumes increase within a month after suggestions’ launch, then the model was successful. If the volumes decline to compare to a previous period, then the model hasn’t been efficient enough. A metric for marketing cost optimization you could look into the email conversion rate, or into the cost per click of a campaign.
Make sure you know how to measure your metrics and for how long until you make a decision if your model is successful or not.
To summarize, focus on the problem, the use case, and user experience. Let Win1to1 worry about technology, algorithms and everything else.
Subscribe to our newsletter to read about all stages of a Machine Learning project.