We have now almost completed the next generation in Can Do's resource management tool.
The previous AI, which indicates a recommended action when people are overloaded, gets another model in addition. This model suggests concrete alternative resources. The new model is far more adaptive than the current AI.
What kind of AI is this?
The previous AI that determines recommended actions is a rule-based system with a very weak learning domain.
The new AI, on the other hand, is fully capable of learning and belongs more to the machine learning (ML) category. More precisely, this AI works according to the IRL (Inverse Reinforcement Learning) method and learns from the user.
What is the problem with resource planning tools?
One solution approach for project managers in the case of an overloaded resource is to schedule another person to do the work. However, this person must meet several criteria to be eligible. These are, for example, sufficient availability and - most importantly - the necessary skills to do the work instead of another person.
Can Do is also a skill management software and all resources can have skills. It is also possible for the project manager to specify which of his many skills the employee performs when assigning the resource. However, this is simply too much work for project managers.
Can Do has to find out which person can replace another person in case of overload.
The illustration shows that the AI has found 2 overloads in the whole project, where the "old" AI advises to "intervene immediately". The illustration is completely redesigned and looks completely like this:
How does the AI determine the alternative resource?
Within the first layers of the AI, we have already firmly defined the first learning models. This is to prevent the AI from trying all of a company's resources when searching for alternatives. In installations with 4,500 users, this would take far too long and the results would be questionable. Therefore, the AI assumes, for example, that people in the same department may also have similar skills. There are other assumptions that help the AI search more specifically
What then is the concrete proposal?
The AI has identified an alternative person and simulated their availability. This means that if the alternative person comes into question and is used, the risk situation is better than it currently is. The suggestion is therefore an alternative person of whom the AI assumes that this person can also perform the work and is sufficiently available.
What is the procedure for the project planner
The lower layers in Can Do identify and assess the overload risk. They make a recommendation such as "intervene". The project planner can now have an alternative suggested for this particular risk. He can accept this (replace) and have the current risk resource in this work package replaced by the new resource. Work that has already been done is taken into account.
If he rejects the proposal of the AI, another resource is suggested to him. This can be just as suitable, but not better. A maximum of 5 resources are suggested. If all suggestions are rejected by the project manager, the AI is at the end of its possibilities and can no longer help here.
The button below "Ask AI" triggers the new AI, the AI now starts to find alternative people and also simulates if the capacity is sufficient. If a resource is found, it will be displayed. Here is the whole process in a short video:
How is AI evolving?
The assessment of whether a person can take over the work of another person is best made by humans. So when the project manager accepts or rejects a proposal, the AI learns from him. Over time, the AI gets to know the people in the company better and better, so to speak, and makes better and better suggestions.
What are the benefits for users?
This list is extremely long, here is just an off-the-cuff list:
- Time savings: the project manager does not have to research what resources are available himself
- Knowledge: Project managers don't have an overview of the capability of all resources in the company, certainly not if the company is global. The AI gets this knowledge and suggests people that the project manager would never have thought of.
- Quality: It becomes much easier for the project manager to resolve resource overloads and thus stabilize the project
- More even workload: Typical resources that are always overloaded can be relieved.
Easier introduction for new project managers: This group of people does not yet know the resources of a company well, but the AI does.
- Further analyses: For example, key resources can be identified.
Does the AI work?
We don't know for sure yet. The first internal training at Can Do was promising. In July 2023, the beta version will go to three selected cloud customers. There, the AI will then be trained by the customers on a company-specific basis.
Ultimately, we will survey users to see if the suggestions make sense and save you a lot of time.
As soon as the first round of training is completed in the summer, I will report the results here and then also communicate a concrete release date.