Skip to main content

I am finding that a lot of times the “Extremely negative message” Lifecycle event is not what we would consider “Extremely Negative”. In many cases it may have negative sentiment, but it is not in the Extreme Category. What is the best way to calibrate this? I’ve been going through and trying to delete those that it has incorrectly classified, but I’m not sure that is helping to train the model. Any tips for how to improve this? 
 

We have some cases where we have messages in foreign languages being translated via Chat GPT and it is flagging almost every message as extremely negative, even though in most cases they are simply reporting issues or asking for status updates as part of routine business. As you can imagine, a view like this is alarming for our account owners to see, and not accurately representing the customer relationship. 
 

 

Two questions:
 

By deleting, you mean clicking the trash can next to the lifecycle event, right? 

Are these Extremely Negative Messages coming from a ticketing system, by any chance?

We have this exact situation happen and made a couple of adjustments in our settings to make this a little better, but I don’t want to muddy your waters.


Deleting - Yes, clicking the trash can by the Lifecycle Event. 

Some of them are attached to tickets, but most are from email threads, some of them discussing tickets, but not a part of the ticket thread. 


Ok, we had this issue as well. 

One thing we did that made a difference was to go into our Salesforce (this is what we use for tickets) configuration and select these options:

 

 

 

Another thing I’d recommend is once you’ve trashed the “Extremely Negative Message” to change the red unhappy face emoji to the yellow neutral emoji. I’m trying to get buy-in from all my users now, and it can be hard, but in situations like this, it truly does take a village.

 



 


We are configured in the same way, and have been making adjustments to the sentiment, but it seems the Extremely Negative Message Lifecycle Event has very thin skin. 

Can anyone from Gainsight share how that lifecycle trigger is set up? What terms it is looking at? I’m hearing I’m not the only one that is experiencing frustration with it. We just deleted all of these events and turned it off because it is pulling in so much noise that isn’t applicable. 


@MKorzun  You are not alone and I have a data science project that I am teeing up for engineering to review our Extremely Neg, and all of our other “Risk” events to optimize them.

We already have work underway right now for Sentiment that is leveraging some newer full LLM capabilities to improve its accuracy so you should see some improvements in the very short term already from that work. But we strongly believe that our core use case is Risk Identification (outside of all the other great things we do) and so a project to refine our Risk events and improve them is a high priority for us right now.

In that project, one change we are also considering and I would love the communities feedback on is a shift in the name of some of our Lifecycle events.  Changing ‘Negative Sentiment’ (which currently does not show up often) to “Friction” and ‘Extremely Negative Sentiment’ to “Escalation”.  Ultimately, I view Extremely Negative Sentiment as a little too broad and too much of a catch all at this time… when everything is extremely negative… nothing really is.

Out of the Box, our risk events are currently: Negative Sentiment, Extremely Negative Sentiment, Churn Risk and Churn Notification. (you might bundle account personnel changes in there too)

I’d love to hear all of your thoughts and what you would prefer in both the naming and the behavior of different “Risk” type events.


@bradybluhm When we rolled out Gainsight to all our CSMs, our biggest hurdle was how alarmist Staircase AI seemed. The terms “extremely negative message” were off-putting to the team, and the idea that they needed to change the emojis to continually teach the AI didn’t go over well at first. Most of the “extremely negative messages” in our case were from Salesforce tickets (cases) and many of them were simple requests for help. A really memorable/funny one was when Staircase flagged someone unable to meet for lunch (during a conference) as an extremely negative message. 

Speaking for my end users, your proposed changes would be a big step forward. I might even suggest something like “Possible Friction” instead of “Friction” - this gives less of a 3-alarm fire feel, more like a gentle “hey, you may want to look into this”. 

I’ve never even seen “Negative Message” - and can’t find it in my list of Lifecycle Events, so I can’t speak to that. 

Another you might consider adding is “Customer Advocacy” - I realize this generally falls under “Highly positive message” but when I look at operationalizing this for my Product and/or Marketing team this would allow them to wade through even less stuff.

I really appreciate your engagement with all of us here in this corner of the Community and can’t wait to see how this product evolves.