⬆️ ⬇️

How does Yandex.Mail support service learn how to respond within three hours

We are trying to make Yandex.Mail such that everything in it is immediately obvious to everyone and no help is needed. Alas, this is not always possible, so we have a support service. I think that Habré is widely understood that it is very difficult to organize a support service so that it can quickly respond to all users of such a huge service as Yandex. Mail, and at the same time did not send out semi-automated replies, but helped essentially and deal with the problems in the service itself.







Last time, we talked about how we improved the work of the Yandex.Mail support service by reducing the response time of users to one day. And today, when Habrahabr moved all the hubs about project management to Megamind, we decided to tell you how to reorganize support work so that the average response time for people decreased to three hours.



Do not be afraid of their desires



More than a year ago, when we just started changing the work of the support service, the average response time was more than three days. Then the answers during the day seemed to us the ultimate dream. But as soon as we achieved this, I wanted more.

')

So how is the task that has not yet been solved (in our subjective opinion), many online stores and even paid services, of which we are users, turned out to be within our reach?



Problem search



In order to find out on the shore what and how influences the response time, we unloaded all events from our administrative panel for a full month by the minute. Any letter from the user takes time from a support service specialist, even if it just had to be moved to the next queue. Therefore, it is important to consider not only the arrival of the letter and the response to it, but also to moving to another queue. Closing without an answer takes at least a few seconds - the letter should be read. We transmitted this data to analysts and asked them two questions:

  1. Can the support service of the current composition to process letters in real time?
  2. What exactly gives the greatest increase in response time?


Analysts told us the following:

  1. The number of outgoing letters from us for a period of time is 96% of the number of incoming in the same period. This means that the group copes with the load and theoretically could handle most of the tickets in real time.
  2. If we take into account only tickets that are closed in less than two days, then the average time it takes is 12.3 hours. For different days, this figure ranges from 8 to 16 hours. Despite the fact that there are 6% of them, in the process of working on which difficulties arise, we still brought out the first metric - the average response time should be reduced by 9.3 hours.
  3. 53% of all letters that were processed by a support desk employee during the eight-hour workday came on the same day, and 47% remained from yesterday.
  4. The number of open and closed tickets per day usually varies within 25% in either direction. This difference is also large enough to provoke delayed response times.






The matter of technology



The main thing that we understood: the main delay in the response was caused by the old tickets accumulated in the queue. So we formulated the first task: one-time reduce the level of open tickets, while maintaining the average load. To solve it, we didn’t even need to change the work schedule or hire new employees. We just organized a couple of subbotniki and dismantled all the hanging problems.

This change alone gave an increase in the average response rate at 6 o'clock (12.3 - 6 = 6.6). That is, to our goal at 3 o'clock we needed to cut the response time by another half.



It remained to synchronize the work schedule of the support service staff, since the floating schedule on weekdays gave us about 3-4 hours to an average time delay.







As can be seen on the graph, it was necessary to pay attention to the periods from 7 to 11 and from 19 to 22 hours. In this case, the time from 13 to 16, on the contrary, can be unloaded, which we did.



The new metric was an empty queue at 22-00 daily and a new work schedule. The effect is observed on the graph:







And the winner is ...



Only organizational changes have allowed us to grow in response time more than doubled - from 30 to 75 percent of responses within three hours. The number of letters per day increased by only 10 percent, thus our fear of a significant increase in letters due to quick replies turned out to be in vain. To disassemble the remaining 25%, you need to enter two shifts, since most of them are questions that came at night Moscow time. Now we answer them within 8 -12 hours. But we will tell about it next time.



We act on the lead



80% of the letters that come to the support service are typical questions that have standard answers: the user simply did not find any setting or wants to find out how some service works. The robot can also answer such questions.



To solve this problem, we made a feedback form in the form of a pop-up with a noticeable search line for questions and answers, a web screenshot and the ability to immediately attach the problem letter as an attachment.















In the measurement of benefits, we were compared with the usual feedback form. The key metric of the project was the conversion from those who came to the form to those who wrote to the support.



The final figures were as follows:



From those who wrote to the support service:

Those who have not written:



The new form with the search turned out to be 33% more efficient than the old one.

It also became clear that people, contrary to popular opinion, can not only write in support, but also use the search themselves, read and find answers - it’s enough to give them a handy tool.



Results



Yandex.Mail is a free mail service with one of the fastest support services even among paid services. 75% of our users receive answers to their questions within three hours, and 90% - within 24 hours.







We managed not to grow in the number of letters and at the same time to powerfully improve the response time by the current team - without any major changes in the work schedule. And on this we do not stop. Now we plan to learn how to answer important questions within an hour, start automatically checking the quality of the answer, learn how to count the user's karma on the service in order to more accurately determine the level of his knowledge and skills.

Source: https://habr.com/ru/post/245637/



All Articles