When discussing artificial intelligence with people a number of misconceptions are commonly expressed. Some examples of common misconceptions are; artificial intelligence is to make computers interact like a human, artificial intelligence is something that is not available and is science fiction, artificial intelligence is something only companies like Google can do.
There is a lot of misconception, telling people about artificial intelligence, telling people about neural networks, deep learning and self learning algorithms is commonly received with a lot of skeptics. A commonly heard response is that it is too complex to develop or the business problem cannot be resolved with artificial intelligence.
It is a true fact that developing artificial intelligence solutions, deep learning solutions and self learning algorithms can be complex and often is. However, you do not have to be Google, as stated by Google itself as one of the lessons learned by the Google development teams. One of the takeaways from Google is however, you need a lot of data. And this is a very true statement. When developing, as an example, a solution based upon self learning algorithms you will need a lot of training data and a lot of actual data.
An example of a business problem which can be resolved by using more advanced solutions like self learning algorithms is supply chain planning and workforce planning. A large portion of companies who do have supply chain operations as part of their business and have the associated workforce planning in warehouses and logistical centers do planning of operations and workforce based upon rule based engines or based upon human intervention. Supply chain planners and warehouse operators spend a large portion of their time on planning. Given a large dataset with historical data and a self learning algorithm you could optimize operations fully automated in a fashion that would not be possible with a human operator or a rule based solution. 

The above shown diagram shows a commonly implemented method used within companies to plan operational tasks in supply chain and warehouses.

  • Work orders for incoming goods, outgoing goods and internal move orders are entered into a central ERP system (in this example Oracle eBS).
  • Warehouse managers and planners do query the workload in the eRP system as shown in bullet (A).
  • Warehouse managers and planners plan, based upon the available workforce and amount of work, the number of workers needed for that period of time and plan which worker will execute which tasks (in which fashion) as shown in bullet (B).
  • Warehouse managers and planners plan instruct the workforce about the tasks assigned to each individual, the order in which they need to be executed and in which fashion, as shown in bullet (C).
  • Upon completion of a task, or upon completion of the workday, the tasks are registered in the central ERP system as completed.

The above way of working is a good way of executing the operations in situations where the planners and warehouse managers are actively involved in day to day operations and are part of the workforce itself. Small companies with a limited number of employees working on supply chain operations will find that this is the most suitable solution for their situation. As soon as a company grows above a certain number of employees in supply chain operations and the number of work orders is above a certain threshold this model will not provide you the most optimal planning and will make operations slower and more expensive than needed.
By implementing a solution based upon self learning algorithms you will be able to create a solution in which the perfect balance and planning for your workforce and the work that needs to be carried out. Planning of tasks over available resources, planning of the number of employees needed on a specific date and arranging tasks in such a way that they can be carried out more quickly and in a more efficient manner will be the result of implementing a solution based upon self learning algorithms. The basis used for the training of an artificial intelligent solution will be based upon the historical data available from the period that planning was done manually.

A self learning algorithm will initially learn on basis of historical data, based upon this data a planning will be made. Every planning and the achieved results of this planning will be added to the dataset used to optimize the next planning. As an example, if the algorithm decided that a certain load of incoming goods should be unloaded by two persons and the result is that the task is completed within 30 minutes the algorithm might try another solution in the next occurrence. Within the next occurrence it might assign the task to three persons and measure the result, a next occurrence might involve only a single person. Given the results in time and other factors (labor cost, costs of other tasks pending, priority planning) the algorithm will decide the optimal number of resources to assign to a specific task.
A self learning algorithm based solution can also involve the order in which tasks will be carried out, the optimal way in which the warehouse can be stocked and which goods can be placed best at which location as well as in which order trucks should be loaded and unloaded. All to come to an optimal result in a way which for human operators is as good as impossible due to the number of input parameters that can be given to a self learning algorithm based solution. For a human operator it might make sense to take certain decisions while based upon a calculated approach a non-obvious approach might result in the optimal solution and planning.
An even more interesting result of self learning algorithms can be optimizing the location of items in your warehouse to optimize the efficiency of your workers, resolve queuing problems, lower the cost of storage space per item and the handling cost per item.
For all solutions a large set of data is required and depending on the number of logistical actions per day the amount of number crunching can be enormous. The benefits and savings can be equally enormous. Capgemini has developed a number of blueprints based upon Oracle technology in combination with non-oracle technology which is ERP and warehouse management system independent. Optimization blueprints are build based upon Oracle Exadata and Oracle Private Cloud appliances in combination with Oracle R enterprise solutions and/or custom build self learning algorithms written in Python