We aim to hire smart people, and let them be smart. This means we try to offer sensible guidelines that will help, but avoid “scripts” or rigidity. Speak in your natural voice as you would to a peer at a conference. Obviously avoid unprofessional language, but you’ll want to match the customer’s tone. There is often a desire to “unify” by everyone speaking in a robotic tone:
“Thank you for contacting support. We can help you with this. It looks like you are asking for help with resetting your password…”
This dehumanizes us and we lose our best asset: Support from a human. When you speak more naturally it anchors that we are also real people and not “support minds as a service:”
“Ah, sorry to hear that you lost your password. I’ve issued a password reset and you are good to go. In the future, you can use this link:
Let us know if there is anything else we can help with. “
At GitLab, we consider the elements of our responses carefully. If you find yourself wanting to use canned responses, or are saying the same things over and over, it's probably an opportunity to improve our process. That is, instead of creating a text expander asking for logs, take a step back. Is there something earlier in the experience of opening a support ticket that we could do to reduce the need for repetitive text? There are times when it's appropriate to use formal language and canned replies, but those will be rare. Whenever we can we push towards empathy and humanity and automate and preload the process.
Consider this spectrum:
Request requires -- v Pure process |--------------------|----------------------| Higher level thinking Robotic Support Voice |--------------------|----------------------| Empathy and Humanity Tone is -- ^
Be empowered: at GitLab Support we want humans with agency, not agents. If something feels broken, ask. If something feels inefficient, fix it. Everyone can and should contribute.
When it comes to actually answering tickets, the sandwich method is a great 3 point guideline that will help you elevate your responses. A great customer reply will contain the following 3 things:
For example, a customer might ask:
“ My GitLab server appears to be slowing down. Can you help me?”
An okay response is:
“ Please send us over your production logs and we can use that to troubleshoot some more. “
Notice we asked for what we needed, and we’ll be able to help. Let’s make it great using the sandwich method:
“It will be helpful to get as many logs as possible during the slowness to help us isolate the problem. You can find them in /var/log/gitlab (This is our ask)
Usually when we see slowness, it’s isolated to a specific part of the application. Can you help us narrow down the issue by outlining when you see things slow down? (This is our premise for them to reinforce our expertise.)
Once you send these over and help us understand how you are getting to the slow state we’ll be happy to help you dive in some more.” (This is us reassuring them we’ll help.)
We’ve asked for what we need early. Hit them with it, so they can start thinking on it and if they stop reading right there, they didn’t miss the ask. We’ve given a hypothesis which is something for them to chew on and understand our vantage. We don’t want to serve our customers we want to partner with them. This is one way for them to see us as a peer vs. “support minds as a service.”
We then make sure to let them know we are still here and will be when they come back.
There will be times where you might need to add more or even apologize for something, but this method should be applicable to the majority of tickets and help us deliver excellence.
We can think of working through tickets as alternating between two modes of operation: The Characterization Mode (CM) and the Hypothesis Testing mode (HT).
In the characterization mode, we're working to establish basic facts about what the user is trying to do, what is effectively happening, and information about context and state that might be relevant. We can construe a ticket as a puzzle as a useful metaphor, and look to elaborate its contradictions. This also might serve as a baseline for reproducing steps and for a potential bug report.
We can be transparent that we're working to characterize the user's issue. When in this operation mode, we can ask:
The second mode of operation is the Hypothesis Testing mode. This is the creative step where we can behave like scientists and theorize about what might be going on the user's side.
We also can be transparent that we're in the hypothesis testing mode. When doing so, we can clarify:
It is interesting to note that the hypothesis test feeds back into the characterization mode, as we're establishing new facts about the user's scenario with the test.
You can come up with multiple theories and corresponding tests in a single response. In fact, doing so might help explicit the structure of the operation modes described above. The facts established in the characterization step would be common to all theories. The possibilities described in a single theory, however, might not apply to others, so it is important to keep them separate.
The idea for this was taken from Jeff Anderson's talk: https://www.youtube.com/watch?v=DK1ew1HpmeY&t=127s
"Ticket deflection" can sound like a way to weasel out of work, but in reality it's about improving the customer experience. Customers don't want to write into support. They'd much rather not have the problem in the first place. Fail that, they'd like to solve the problem themselves. If they can't do that, then they want a technically skilled individual to help them solve their problem.
There are 4 primary tools of ticket deflection:
Having an excellent product is the first line of deflection - a product that is free of defects and works as expected will reduce the number of support cases organically.
Support plays a vital role in surfacing issues that our users encounter while using GitLab by:
The Statement of Support describes the areas covered by Support and the areas we can't promise to cover. This is a tool to set expecations for customers as well as to help the support team to make sure we're supporting what we're experts in. You can read more about the philosophy behind it in the blog post where we introduced the Statement of Support.
As a member of the Support Team at GitLab you should be:
By using a docs-first approach to answering, we can ensure that the documentation remains a highly useful single source of truth, and that customers are more aware of where to find content on their own.
Always respond with a link to the docs. If docs content is missing, create it and link the customer to the MR. If you are working on a ticket that is about to breach, you can clear the breach with a response, and immediately follow up with the MR. Remember: go slow, to go fast.
The best way to improve the customer experience is by being knowledgeable about our product. You should coordinate with your manager to develop an intentional learning plan that builds on your strengths or broadens your knowledge. You should also freely ask questions, pair with others and demonstrate a posture of vulnerability that invites others to follow.
Whatever you learn, make sure that you're constantly bubbling it up and broadcasting it: