I'm going to start this blog by citing Netflix's example—about it showcasing 'you may like' content snippets immediately after you've finished watching something on it. How does Netflix know your preferences? How does it predict your behavior? These recommendations are possible because Netflix knows about your viewing history.
What is Predictive Modeling?
The premise of this 'knowing' is a process called Predictive Modeling that takes historical/ already available data and feeds it into pre-defined analytical algorithms—known as Machine Learning Models—to predict future outcomes. It leverages several techniques like statistical analysis, data mining and data modeling to build a machine learning model capable of generating accurate predictions based on a given hypothesis.
Predictive modeling can tell organizations who their prime customers are and their user attributes—including age group, gender etc and what these customers need. Based on the output generated from a machine learning model, organizations can make informed decisions about whether to upgrade existing products or create new ones capable of delighting their target audience. The power of predictive modeling can range from detecting spam emails in your Gmail inbox, to predicting stock market trends based on past performance.
Dynamics 365 and Predictive Modeling
Building a predictive model is a computational process that requires a lot of preliminary research. When done incorrectly, it can prove to be counterproductive. For instance, imagine a food delivery app recommending a double beef cheeseburger to its vegan customers!
Azure Machine Learning prevents such instances from happening by providing you with several pre-configured machine learning models suitable for catering to different business needs. Microsoft Dynamics 365 Finance and Supply Chain Management can help you identify and extract relevant data, create machine learning models using Azure Machine Learning, and test the performance of each model. Let's learn how to configure it:
Step 1: Define the business problem
Clearly defining a business problem will allow you to accurately test and measure the success of your model. To correctly define it, you may also need inputs from both internal and external customers. For example, a manufacturing company may want to predict the occurrence of non-conformances during the manufacturing process. The company has a set number of operations to build products. But occasionally, an item does not meet the set standards and gets discarded. The management would like to understand the reason for this defect and the likelihood of the next non-conformance. Once you define a problem, you can intuitively decide what the solution should look like and proceed accordingly.
Step 2: Collect the data
Once the business problem is defined, you need to collect the data relevant to it. Since the overall quality of the data indicates the quality of the model, it is essential that the data is up-to-date and accurate. You can also extract this data using Microsoft Dynamics 365 because it can connect an external repository to your ERP solution. So if you have data in Dynamics 365, you can configure the data for extraction without having to write too much code. Here's a dropdown of how to do so:
- Click on Configure Entity export to database
- In the Data Projects dialog, click on the Export Projects option and name your project
- Identify source data in the Microsoft Dynamics 365 Finance and Supply Chain Management user interface
- Identify the entities and fields that contain the source data
- Create data management export projects in Microsoft Dynamics 365 Finance and Supply Chain Management
Step 3: Organize the data
Data preparation is a key step to take before you can feed the data into your prediction model. You need to clean the data and remove any outliers that can skew the overall result. A well-selected dataset ensures that the predictive model's performance is as accurate as possible. Once you extract the source data, use the following steps to organize the dataset to train the machine learning model:
- Set up the Bring Your Own Device (BYOD) functionality with Azure SQL
- Configure data migration exports from Microsoft Dynamics 365 Finance and Supply Chain Management to BYOD
- Finalize dataset preparation for model training
Step 4: Train the machine learning model
The next step is to train the data against different machine learning models available in the Azure Machine Learning Studio to choose the best model for the dataset. For this, you need to:
- Ingest the training dataset to the Azure Machine Learning Studio
- Identify the target column which is already present in the dataset
- Select/ create the training cluster which provides resources to run the model in Azure Cloud
- Segregate your dataset based on the machine learning problem: Classification, regression, time series, etc
- Define the primary metric to rank the different models: Accuracy, precision score, etc
After selecting all the parameters, you can train all the models and identify the one that performs the best based on the primary metrics chosen. A key element to note here is that the best performing model may not be 100% accurate, and so we should always leave a small margin for error.
Step 5: Publish the model as an Azure service
Once you have chosen the best-performing model, you can deploy it as a service so that different engineering solutions can use it at their end. To deploy the model as a service:
- Click on the Deploy button
- Select compute type and authentication options
- Get the REST endpoint and test it to check whether the service is responding correctly or not
Step 5: Run machine learning service in production
In the last stage, you can create an Azure-hosted functionality to extract new production data from Microsoft Dynamics 365 Finance and Supply Chain Management and feed it to the aforementioned machine learning service to predict accurately. For this, you need to:
- Prepare data for machine learning service consumption in Azure SQL
- Call the machine learning service and populate the Azure SQL repository with the predicted data. You can also use Spark for Big Data/ diverse data sources
- Prepare dataset for Power BI presentation
- Create Power BI dashboards/ reports to ingest data from the Azure SQL repository and present the predicted results to end users
- Embed Power BI reports in Microsoft Dynamics 365 Finance and Supply Chain Management client
What you've learned and where to go from here
So far, you have learned how to create and deploy a prediction model using Azure Machine Learning and Microsoft Dynamics 365 Finance and Supply Chain Management. The next step is to integrate this prediction model into your production services so that you can leverage the benefits of real-time forecasting in your daily operations. A predictive understanding of your business can help shape your strategies and channel your efforts in the right direction.
With Microsoft Dynamics 365 Finance and Supply Chain Management, Azure Machine Learning, and predictive modeling, you can improve customer retention and satisfaction and empower your employees to proactively optimize operational efficiency and speed.
At Columbus, we have helped clients create winning organizations. We have industry experts with years of experience who specialize in digitally transforming companies by deploying data and analytics-oriented solutions.
