The purpose of the control phase is to make sure that the problem does not return.
For that you have to solve technical and non-technical problems.
Let’s start with the non-technical problems:
- Train your employees.
It is of utmost importance that you train all relevant employees. Train the people who are going to use the A.I. solution and train those who are going to take care of the A.I. solution. Make sure before your outsourcing partner leaves the house that the employees will be trained. Trust me. The real questions arise when your product is released on the employees or customers. Plan this additional time and budget ahead. If you start training your employees while your outsourcing partner is maybe already on another project, you will never get as good and fast feedback and support as if you had it slotted it in in the overall project plan
- Document everything
Actually also a no-brainer, but mostly overlooked. You will need documentation for people being trained (so that also new employees in 6 weeks will be able to be trained) and also the complete A.I. solution needs to be documented. It is a complex thing which you are building. You do not want to find out how badly your solution was documented only in the moment where it stops working.
- Develop monitoring plan
Just imagine you implemented an A.I. solution in your system and it runs well in the beginning. But as time goes by, the data/ customers/ products / inputs you are using now do not fit the algorithm any more. The result will be that your formerly artificial intelligence product will not make intelligent decisions anymore. So discuss with your outsourcing partner in advance to set up a monitoring system. Let them come up with key performance indicators for your A.I. solution. You should have a dashboard where you see how reliable your new algorithm is working. Once you have worked out a set of KPIs and also found a way to display the values in your monitoring dashboard, have them help you set it up and train an in-house employee on actually monitoring it. This again sounds trivial, but many of us have seen monitoring systems just not being used in companies as nobody felt responsible looking at it. Furthermore discuss with your outsourcing partner and with in-house employees, what exactly will happen, once your monitoring system will sound alarm. Who will be able to retrain your algorithms? I suggest to have a service contract with your outsourcing partner which will regulate all these questions. But also come up with a plan, if your outsourcing partner should not be able to do it. Either they just might not have the capacities to help you at that time or they might have gone out of business in the future. As a good business person, you plan ahead.
- Develop implementation plan
Many people forget that their precious prototype also needs to be implemented in their systems and processes. If your outsourcing partner develops the prototype in Google’e AI environment, but you are using AWS? What would you do? Talk about implementation also at the beginning of the project to allocate enough time and ressources for these questions. Furthermore make sure that the data-flow is properly aggregated, so that the new algorithms are set up automatically with the right formatted data.
As you see, there is a lot to think of. Please make sure that all future collaborations with your outsourcing partner are being written down in a service level agreement so that everyone knows, what to do.
Overall over the course of the last two weeks, we discussed how process optimization can bridge the gap between project management, agile development and A.I.. And I hope that these short texts could give you a primary idea, how to handle your first A.I. projects.