450x faster way to pick a supplier
Problem
Estimate costs per each order from 5000 suppliers
To deliver a proposal to a client, a manufacturing produce-to-order company needs first accurately to estimate the costs of inputs. The company works with approximately 5 thousand suppliers and carries tens of thousands of parts (articles) in its systems which it uses to manufacture an extensive range of its products.
Manual process that employs four people
Weekly, the purchasing department had to resolve hundreds of cases. There were four people dedicated to costing inputs. An estimation took, on average, 15 minutes, and although the staff was experienced, there was a degree of error in the estimations and some volatility in outcomes.
Solution
- Two neural networks
- Fully no-code solution
- Integrated into existing ERP
Several distinct tasks to automate
The system that would help the client with automating the work of the experienced staff would have do 3 different predictions:
- Identify potential suppliers,
- Predict the price at which these suppliers would deliver the part,
- Predict the delivery dates for the potential suppliers.
Neural networks to predict prices and delivery dates
Thanks to a long history of purchasing, the company had a database of 120 thousand records with all the information needed to train two separate prediction neural networks. The recommended supplier was chosen based on the ideal combination of price and delivery date per each proposal that needed to be submitted.
Outcome
- Outcome: Removed cost rise pressure
- Data: Internal
- Development time: Short (<1 month)
The solution has reduced a task that would take an average of 15 minutes to 2 seconds and allowed the company to dedicate human skills to other activities. This highly manual process became an efficient AI-assisted activity.
