Why and how to use AI for real time dynamic resource allocation

Organizations across all industries face unprecedent challenges: customer demand volatility, shortness of labor, rising costs of energy, etc. Being able to respond in real-time to uncertain events and operate more efficiently is therefore a critical capability. That’s why a large manufacturing company selected Apgar Advanced Analytics solution to automate Job Scheduling in a large US plant and to reduce labor costs, raw materials loss, energy consumption, and inventory costs.

How to apply AI to resource allocation? .

A Job Scheduling use case in manufacturing

Apgar used its RILP platform to train and deploy an artificial neural network (ANN) able to
recommend job scheduling actions, meet customer demand and optimize operation costs.
RILP takes advantage of reinforcement learning (RL), discrete event simulation, and
historical data for training. After iteratively tunning the RL parameters we got model that beat
the performance of the human scheduler heuristics, used as a benchmark.

“Being able to create an AI agent that can operate a close simulation of an manufacturing plant with 30% lower costs than the human scheduler heuristics was an extraordinary achievement and showed the potential that APGAR AI solutions have to create value in Operations.”

Mário Duarte Head of Advanced Analytics, AI Driven Operations Expert

Apgar Advanced Analytics follows an agile and value-oriented methodology.

Business Case Discovery

Apgar Advanced Analytics starts its customer engagements with a business
case discovery provided free-of-charge. During this phase business challenges and the
operational systems/processes to be operated in an optimal way are discussed and detailed,
and historical data is also analyzed. If the business problem is akin to be solved using
Apgar RILP solution a Proof-of-Value is proposed and the success criteria defined.

Proof-of-Value (PoV)

The PoV entails the configuration of a simulation of the process, the
preprocessing of data and the implementation of the heuristic rules used as benchmark.
Once an AI agent performs better than the benchmark, we are ready to evaluate it against
the current way of working in a real environment. To lower the risk, we can provide the PoV
at a reduce fixed price and include a success fee due only if the success criteria is met.