For me, the most interesting aspect of the project, was the high variability of the market. Price of steel has many factors to take into consideration
The client is a European leader in construction material trading. They aim to offer their customers competitive prices, fast order completion & delivery, and tailor-made additional services.
The client operates on an unpredictable market where prices of construction material, currency rates, and demand for the offered products constantly change and need to be monitored and benchmarked. These factors also impact the delivery time of the offered products. Due to this, the purchase process of the materials, as well as ensuring impeccable customer experience and timely delivery of goods, becomes a tough task.
These circumstances limited our client’s capacity to secure a reasonable profit margin while offering customers the goods at a competitive price delivered on time.
Building a model that would predict the total amount the customer will most likely pay for their order based on the ordered resources, the volume of the order, and customer data.
Making the sourced data available and easy to use for the sales team helping them to provide calculations to their customers.
In addition, multiple changing market factors influencing the prices of crude resources made the process of price evaluation difficult for the company’s employees.
Issues with communication flow, analytics, sales dynamics (the number of new customers, up-sells, and cross-sells) have resulted in company analysts’ wrong decisions about new product prices. These decisions were often entirely based on employees’ intuition.
The lack of information about multiple market factors made offer customization difficult and resulted in a decrease in sales volume, a low profit margin, and sometimes unprofitability of certain projects.
Before the client approached us, they already felt that AI could probably help solve the complex problem with offer customization, but they weren’t sure where to start.
They were lacking the know-how and technical skills needed to adopt AI and needed help with the selection of a predictive model to start with.
The team had a general idea of what they wanted to achieve, but they weren’t able to specifically state how they would imagine the solution to their problem, which is very common – and natural – at such early stages of preparation for AI adoption.
To help the client, we implemented our unique AI Sprint approach that is a rapid prototyping of an AI model. It aims at verifying if and how AI is applicable to solve a given business problem. It includes a 2-day workshop and prototype development and takes less than 1 month.
We started with the AI Sprint, during which we aimed to find the most time-efficient solution that would assist employees in making decisions based on historical data (purchase history, financial data, delivery time) in correlation to changing market data. This was important to ensure the client offers customers a sufficient amount of the product in a given time and can increase profit margin together with revenue.
During the AI Sprint, we learned that the prices of products the client offers rely on multiple factors, including listings on the stock exchange.
We mapped out the whole sales process in the client’s company and noted what the prices depend on. Having identified what data was needed for the models, we made a list of the data the client collected as well as the data we had to obtain – such as information about competitors or prices of resources. This third-party data would be made available with the use of robotic process automation, automatically sourcing up-to-date information.
After the workshop, the client’s team and our data scientists collaborated over the tasks to create a proof-of-concept – exporting data in CSV format, building the flow and the predictive model as well as a Robotic Process Automation robot. Our team also prepared recommendations for the client for further use and development of the model.
The outcome of the workshop was an offer/order price estimation model. This model predicts the total amount the customer will most likely pay for their order basing on the ordered resources, the volume of the order, and customer data.
The developed predictive model analyzes available historical data and up-to-date records to provide accurate estimates of the order value. External data is sourced by an RPA robot and added to the predicted estimate to provide insight for the sales staff.
The PoC was created and reached a certain accuracy within two weeks, proving the initial hypothesis of using AI to predict the total amount the customer will most likely pay for the order. With that, we were able to create a plan for continuing the work. After another 3-month of development, our models reached the expected accuracy.
The development of the PoC took two weeks and it helped validate the initial hypothesis about the factors influencing price fluctuation on the market. The result PoC delivered assured the client that AI can be useful in solving his business case and that it’s worth continuing the development. With the PoC the client received a model providing estimates of the order value and delivering the estimates to the sales team. The process of getting an estimate includes exporting the order as a CSV file, sending the file to a specified email address – the email then triggers a robot that retrieves the missing data from the database, starts the predictive model, and gets the estimate. The estimate is then sent as an email to the sales team.
The solution will support the client’s sales team in their everyday work, making it easier to provide calculations for the customers. The end customers will get more accurate estimates, and they won’t have to plan for a big margin of error between the due amount estimated on the day of placing the order and the actual one that they have to pay upon delivery.
Top Artificial Intelligence Companies 2023Top AI Companies 2023Top Web Developers 2023Top Web Developers 2023