Expectations from artificial intelligence are now truly high. And often overheated. But let's look at one practical example of whether he is omnipotent? Or is this not a panacea?
Now most companies are looking for ways to reduce costs? Including IT. One of the areas is the optimization of IT departments that provide technical support. And at the same time, improve the quality of this technical support. In general, one of the ideas that is spinning in the heads of managers is to introduce artificial intelligence to the zero technical support line, so that all incoming requests are correctly classified and automatically sent to the right specialist. In general, so that there is no need for a zero line.
But how feasible is this idea?
In short, AI can be used as a chatbot to correctly classify a request or solve a typical user problem. That is, to reduce the number of requests to technical support (in practice by 30-50%) or to correctly “link” the request not with checkboxes in your personal account, but in the format of leading questions to the user.
Let's now look at the situation in more detail.
Let's take a typical corporation, with a developed IT infrastructure and its own IT technical support. 95% of requests to this technical support will be via email. Yes, yes, no more than 5-10% of users will use self-service portals and personal accounts. After all, it’s one thing to simply write an email, and another to go to the portal, log in, select the type of question, write, tick... In general, man is a lazy creature and uses the most convenient tool, avoiding unnecessary gestures.
At the same time, 80% of users write messages as if they are paying for each character from their salary or have just learned to speak. A typical request looks like this: “1C does not work.” Train everyone to write a request using the 5W1H methodology - a utopia, the same as getting people to use a self-service portal.
But let's imagine that people have learned to write requests and everything is completed and completed on time. That is, our technical support specialists are simply exemplary employees. So, in order to sort tasks by letter from email, you will need to work a lot, methodically and expensively. Approximately 2-3 years and more than 20 million rubles: you need to clean, normalize and mark up the database in ITIL, show examples of distribution and assignment of tasks on 2-10 thousand requests, and then maybe the neural network will be able to normally understand the request from 1 letter. But we have little faith in such a scenario.
Another huge limitation is that each company has its own zoo of IT systems, its own organizational structure and distribution of powers. In addition, in most companies there is no clearly distributed area of responsibility. And if there is, then it turns into bureaucracy, when the issue is resolved partially and not comprehensively (to solve 1 user scenario, for example, issuing a new PC, you need to write 5-10 requests).
And as a result, there are no out-of-the-box solutions for sorting incoming writing tasks and never will be.
And as a result of training, the AI will most likely begin to understand from which user which problem may come, and in general it may turn out that 90% of requests are typical, which should not even reach technical support. But for this you do not need to develop and implement AI for recognition, but you need to do analysis, optimize scripts, typical problems and optimize the staffing structure.
As part of this approach, a chatbot with AI and a knowledge base about what IT solutions the company has, with user instructions on board and a description of the organizational structure (who is responsible for what) will help well. That is, an AI bot that either solves a typical problem, or normally describes and binds a task for technical support.
We recommend studying the case from CROC 2020, description here .
The key thing to decide is how to build interaction with the user, through which channel? This is a complex issue from an information security point of view. Typical corporate security departments are now closed with an iron curtain - everything is inside, nothing from the outside. Although it doesn’t work well, they won’t allow any mobile applications, especially bots in a telegram; access via a link, too, you will need to log in from the corporate network (which is also a quest). One more question - how will the system determine who exactly is communicating with it? Moving your entire infrastructure to domain accounts won't be quick. This is a painful path. In general, the solution to this quest is a good exercise for the mind.
The second critical task is preparing the IT knowledge base. It is necessary to describe all IT systems, collect all typical problems and train the chatbot on them, as well as accompany the chatbot in the first year.
Summary
In general, AI can be a great tool that will reduce the burden on technical support and improve the quality of service. But this is definitely not a magic pill. You still need to build processes and restore order; AI will not do this for managers. That is, we again come to a systematic approach : building a client path and typical scenarios, describing the organizational structure with the distribution of areas of responsibility, debugging processes and communications, etc.
