Define the business problem you are trying to solve.

• What results are you expecting from the process?
• What processes are used to solve this problem?
• Do you see AI improving the current process?
• What are the key performance indicators (KPIs) that will help track progress?
• What resources will be needed?
• Consider how to break down the problem into iterative sprints.

Once you have answers, then identify how you can solve the problem using AI.

Ask questions, such as.

• What data is needed to solve the business problem?
• What quantity of data is required?
• Do you have enough data to build a model?
• Do you need more data to extend the existing data?
• How is the data obtained, and where is it stored?
• Can you use pre-trained data?

Consider if your model will operate in real-time to determine if you need to create data pipelines to feed the model.

Consider what form of data is required:

• Structured data in the form of rows and columns.
• Unstructured data, such as images.
• Static data, such as previous sales data.
• Streaming data.

This step is the most time-consuming, with ML engineers spending around 80% of the AI model development time in this stage. A significant amount of time is spent cleaning the data and transforming it into the required format.

Things to consider include:

• Transforming the data into the required format.
• Cleaning the data set of inaccurate and irrelevant data.
• Enhance and augment the data set if the quality is low.

At this step, all the requirements have been collected for the solution modelling to proceed.

ML engineers will define the features of the model, taking the following into account:

• Use the same features for training and testing the model to avoid inaccurate results.
• Consider working with Subject Matter Experts to direct you on what features would be necessary for the model.
• Be wary of using multiple features that might be irrelevant to the model.

Once the features are defined, choose the most suitable algorithm.

While the model is trained and tuned using the training and validation data set, the model will behave differently when used in the real world, which is fine.

The main objective is to minimise the change in model behaviour when it is deployed. Three data sets are used when experiments are carried out: training, validation, and testing.

• If the model performs poorly on the training data, select a better algorithm, increase data quality, or feed more data into the model.
• If the model does not perform well on testing data, the model may not extend the algorithm, and more data needs to be added.

Analyse if the KPIs and the business objective of the model are achieved. If the parameters are not met, consider changing the model or improving the quality and quantity of the data.

Before deployment:

• Ensure to measure and monitor the model performance continuously.
• Define a baseline to measure future iterations of the model.
• Keep iterating the model to improve model performance.

When all the defined parameters are met, deploy the model into the intended infrastructure.