In the previous blog article, we talked about what Data Engineering / Data Preparation is and why data is essential for building a good AI Model. In this blog post, we will give you some insights into how an AI Model is actually built, and what is needed for that.
We’ll talk about the 2nd step of the e2e process of building an Artificial Intelligence Model – the Model Creation.
The Model creation can be split up into three parts, which we want to dive deep into now:
In the training phase, we take our prepared and curated data from the former data engineering phase and start training an AI Model. The goal is that our AI model knows what data parameters are important and what results are expected. As an example, in Marketing: We could have input data of App Store Downloads & App Store InPayment Returns, and want to predict the revenue for the next 4 weeks based on that data, using trends & seasonality of the last 2 years. So, our input will be: 2 years of data, to be used to “train” the model; our output: 4 weeks of predicted revenue.
In the optimization phase, we start making the model better. So we try out the trained model on old data. As an example: We predict revenue for the previous 4 weeks using data that has been backdated for 1.75 years, and then see how good it was. We then start optimizing the data, for instance, by changing the important parameters, or other methods the AI Model provides.
Last but not least, we now continually evaluate the model. Each prediction of the model is saved and then compared back to the real results. This gives us information on how well our model is performing generally, and where we will need to make improvements.
By now, the last two chapters should have given you an insight into how, conceptually, an AI Marketing Model can be built, and how most of the AI/Machine Learning experts do it today. The goal of these articles was to give you a better understanding of how things work and to also ensure that you have your data ready for your next AI project
There is one more thing: Data Quality
We want to highlight again, and re-emphasize data quality here. It is very important to understand that, the quality of your data, from the aspects of accuracy of it (I.e. Correct data) to the amount of data you have, is fundamentally important to your model. If you have data with a very low quality, you will end up with a prediction or classification which is very bad. Please keep that in mind, either when you talk to a Machine Learning / AI Specialist, or also in your future journey with AI.