Artificial Intelligence, ITSM and What LEAN Has to Do with Them

On the way to the correct solution, or preparation

So, accuracy... We followed the Kano principle and identified what we could make with the most reasonable effect: classification matrices, a decision to set the class via searching for key words in the description etc. And hallelujah!

70% of the scope of requests were closed by the robot! Everyone was happy: ’We are cool, we are gods...’. We really implemented it and have enjoyed a normal life with it for several years. But the time moves on, and the loss is still there. Now, we want both classification and human accuracy. We begin to solve the problem of closing the remaining scope of requests. Remember that this is approximately 30%.

So, the main problems are:

1. There are requests of direct users with no description structure.
2. Description of new types of requests takes time.
3. Requests similar to others in the classifier go to the wrong team...

Is it becoming clearer what the point of our story is? So, the time goes by, and LEAN does not bear the costs...

The core of the problem

What is a request? It is a text that needs to be processed. Then, its class should be defined based on the processing results. For example, the use of the above-mentioned classifier by certain phrases and words requires a rather long preparation of the classification matrix and its constant updating.

We began to think about what was to be done. The team realised that it could not solve this problem. Then, we addressed our colleagues from the optimisation department. We have a team such as the one at Toyota factories. They assist all the company in process optimisation: they search, they dig and delve etc.

’You can look endlessly at the three things: how fire burns, how water flows and how other work...’, an unknown author.

We began to storm the hill using the brainstorm. Brainstorm is a very useful tool, and the 5W method turns the storm into the hurricane! Here are the solutions we found:

Our initial problems:

1. The problem of accuracy, or rather the technological weakness of the existing solution. There is no way to improve it.
2. The problem of the cost of support—someone has to update the classification matrix constantly, to monitor deviations.

Suggestions for a solution:

1. The machine should be able to make a decision on the quality of the proposed option.
2. The solution should learn with minimal cost.
3. Apart from that, the cost of the solution support is the same as it was for the previous solution.

We began to go over the options.

When selecting technology, one can think of the statistical analysis with BI elements. But it was expensive, and we didn’t really need a monster with ERP elements there. Our problems were similar to the tasks solved by artificial intelligence and machine-learning engines. Well, our optimisation department was confident of success and invited the guys from the digital solutions department to the next meeting.

Problem solution

Data architects and data engineers spent a couple of weeks on going through a considerable number of frameworks and presented a solution—the first assessment and model.

A month later, we aligned our ITSM with ’Artificial Intelligence’ and completed testing.

What is the result? We don’t need request coordinators at all since the robot processes 99% of all incidents. The remaining 10–15 daily incidents do not create a negative feeling of a routine. The team is satisfied: without distraction from the main tasks, the employees freed themselves from the routine, simply stating that this ’archaic tool’ was outdated and took them off their work.

Conclusion

Constant monitoring of your processes together with your team is invaluable. It allows you to find unnecessary costs and work them off as well as to understand the need to use new technology. Solving the tasks of eliminating routine problems, even the smallest ones, we really create value. And this is the value not only for our customer but also for our employees and our company.