May’s Camp Optimization came upon us and we found ourselves at Show Bar, right down the hall from the Roboboogie head office in SE Portland. This month, we were graced by a few new friends from FullStory, including Sr. Director of Product Management Agata Bugaj, who came to talk about the future of AI in online interactions through her topic, “The Path to CX Omniscience—the Power (and Potential Pitfalls) in Predicting User Frustrations.”
At FullStory, Agata spends a lot of time working on customer interaction. She opened her talk with a story about looking for a gift for her son on a well known big-box brand’s website and finding herself frustrated as she couldn’t find what she was looking for. She realized that in a store, she would be able to find an employee to come to her aid. However, her online frustration couldn’t be resolved so easily. But if it could, would it be… creepy?
Through technology like what FullStory offers, Agata explained how her woes could be identified, addressed, and resolved.
How do you know when users are frustrated? There are only a few paths customers can take online to report issues. (Think of it as the digital version of “I’d like to speak to the manager.”)
The pathways include:
- Support tickets: A support ticket can come in several forms, either through email, “contact us” page, or a direct pipeline. But, as Bugaj explained, we’ve already put the customer through enough pain, why put the burden on them to express their frustration? This isn’t an efficient way to connect.
- Social media: This one elicited some chuckles from the crowd. Think about the times you’ve put a brand on blast for a bad experience. As Agata said, “when our customers are frustrated, we want to know, but we shouldn’t find out at the same time as the rest of the world.” Users know this is effective, although not the quickest or most fun.
- Analytics: This is a great tool but it comes with downsides. Pointing out the tendency for false-positives and difficulty isolating an event, analytics can be difficult to identify specific pain points.
Online, user interaction is interpreted differently than in-person interactions. There are a number of signals that can be identified and learned from—this is something that the FullStory tool has been focused around and utilizes AI to learn from. FullStory has spent years developing and learning from these frustration signals in order to improve their tools and respond faster to customer issues. Frustration signals fall into these categories:
- Rage Clicks: We’ve all done it. We’re in the shopping cart, fill out all our information, and go to click the purchase button to find that nothing happens. You hit it again and again. Finally you’re slamming on your mouse like your life depends on it and still nothing happens before you finally quit.
- Dead Clicks: Dead clicks indicate users are using your site in a way that wasn’t intended. Clicking on an image or text with no links indicates that your customer is intending to use your site in a certain way, but it’s not working and they’re getting frustrated.
- Error Clicks: Often associated with broken links or errors in site functionality.
When FullStory launched a new UI, they noticed a spike in rage clicks. By watching these interactions, they could see that customers were looking for more information that wasn’t there—clicking on keywords that didn’t have links to any more content. FullStory learned in real time that their website needed to add more detailed information, and they were able to act immediately and witness an almost immediate drop in rage clicks.
Interpreting Frustration Signals
Context is everything. The advantage of tracking user behavior is identifying trends, but seeing is believing. Watching user experiences can help verify frustration, and FullStory’s capability to view user session replays helps prevent things like the dreaded false positive.
Another important step is quantifying the impact. How big is the issue? Is it worth paying attention to? Taking these items into consideration will help prioritize any problems as they arise. Another tactic is to set up alerts. Say you have a small frustration on your website where customers occasionally get stuck, but it’s only 2-3% of your audience. By setting up an alert, it can help identify if a small frustration becomes a bigger one, say closer to 10%, which can help you head off potential issues.
With machine learning, these issues will become fewer and fewer, as the technologies will be able to get out ahead of potential pain points and solve problems proactively.
Identifying and avoiding pitfalls is key to ensuring program success. Here are a few to look out for:
- Only looking at one facet: Narrowing in too far can lead to a biased interpretation. By looking at the context surrounding an issue, you can better understand a customer’s frustration and make more informed decisions on optimizing your website.
- Alerting on the wrong signals: Being intentional about what you pay attention to is important as well—do you care about when 1,000 customers hit this frustration, or 10% of traffic? Be sure to keep an eye on your numbers and KPIs.
- Being creepy instead of helpful: Perhaps most importantly, avoid being creepy instead of being helpful. As AI becomes more efficient, people will likely get used to it as it only becomes handier and more helpful. During the rollout of Facebook’s newsfeed, for example, people found it creepy and stalker-ish. Now it’s a part of everyday life. You should ask “is this the best way to achieve my goals?” If you’re not careful, you can scare away even the most dedicated of your customers.
Big Thanks to FullStory
Thanks Agata and the FullStory team for joining us this month, and to everyone who came out to May’s Camp Optimization. Mark your calendars for the next event on Thursday, July 25, for a Digital Experimentation Strategy Hack with the Technology Association of Oregon. Drinks are on us!