More than a year ago, we started to develop AI projects with our customers. Some projects start with consultation, research and evaluation. Some of them we built up PoC (proof-of-concept), and some have already entered the development and implementation stage.
Today, "implementing AI" seems fashionable and trendy for a company, but we found that many companies have not yet fully grasped the life cycle of AI and ML (Machine Learning) projects.
First, many companies don't realize that they need to spend time on data cleaning. The AI and ML learning process requires high-quality data. "Garbage in, garbage out" still holds. Although some companies have implemented CRM or ERP systems for a long time and are able to mine data from past records for machine learning, we still need to spend a lot of time to determine meaningful machine learning and test data. (For example, we may need to filter out some abnormal, inappropriate or duplicate data to optimize the AI efficiency, effectiveness and accuracy).
Another major problem that typical companies face is the lack of their own business operations data. The business owner or management team may know that the application of AI and ML is already feasible, and the threshold and cost of use are also affordable. However, one of the main stages of using AI is to go through a machine learning training process: capturing and identifying business value from historical data and records to derive algorithms that can predict the best business decisions. The data cannot be other company data, they must come from their own business activities.
Most companies should have good accounting data records, but this is not enough. For example, if we want AI to boost turnover, we have at least data from customer profiles and sales order details.
In fact, many companies may only have a few similar Excel files and worksheets. The data in these worksheets come from different departments or business units, and there was no strict integration in the past. They are not enough for machine learning, because as data analysts we still need to spend time to understand how the data is collected and correlated before we can select the appropriate data set for machine learning training and testing. Data from an ERP system is relatively better for machine learning because data relationships have been inherited when it is entered into the system.
If the customer uses an ERP system provided by us, it becomes easier to start an AI project. Our ERP system is built with Python, which is one of the most common computer languages for AI and ML applications. We only need to add a few lines of codes and use of AWS or Google Cloud API to activate the AI function in our customer's ERP.
Equip your ERP system with AI capabilities
1 May, 2021
by
Laps Solutions Limited
Join us and make your company and business a better.
1,000+ accounts work with Laps Solutions and grow their businesses.