For some reason, there are still companies out there that mistakenly think that Twitter is useful only for interrupting people’s conversations with a bit of marketing.

The reality, however, is that technology today also allows brands to provide customer service over these platforms, and even mine them for business intelligence data. The advantage of this is borne out empirically – customers that have a positive interaction with a brand on Twitter are up to 20% more likely to make a purchase from that brand. 

Although much of the actual content which is generated and shared on Twitter is text, the number of likes, number of shares, which hashtags are used and the number of replies or comments are time series data ripe for analytics, especially since Twitter is continuously updating.

Whenever you have real-time data, you can plot it on a time series to find the signals in the outliers of those metrics. Since social media is coupled to the real world, social media stats are often connected to other time series data that you’re already collecting and analyzing. In other words, companies can correlate both business and Twitter anomaly detection in order to pinpoint both platform errors and anomalies — improving their ability to respond in real-time.

Outlier identification and damage control

Reactions on Twitter can be very quick, especially during backlashes against brand messaging faux pas – there are so many reasons why a viral social media post can cause headaches for a business.

One classic example comes from the Twitter account of none other than social media platform Snapchat.

Back in 2018, Snapchat users began experiencing an issue. Their Snap Streaks — a metric pointing out how long they’d been in continuous contact with other users — had disappeared. The Snapchat team created a few fixes for the issue, and the problem was seemingly solved.

Their Twitter fail, however, occurred when they set their support bot to automatically reply to users complaining about their disappearing Streaks. Every time a user made a tweet that included the keywords “lost streak,” the bot would make the following reply.

[here is the embed code for the reply tweet: <blockquote class=”twitter-tweet”><p lang=”en” dir=”ltr”>Thanks for reaching out! We&#39;re happy to help. Please go to: <a href=”https://t.co/pLsBLYE56r”>https://t.co/pLsBLYE56r</a> and select ‘My Snapstreaks disappeared’ </p>&mdash; Snapchat Support (@snapchatsupport) <a href=”https://twitter.com/snapchatsupport/status/1017125218324803584?ref_src=twsrc%5Etfw”>July 11, 2018</a></blockquote> <script async src=”https://platform.twitter.com/widgets.js” charset=”utf-8″></script>

Twitter users quickly caught on, and hilarity ensued, with hundreds of users piling on to make fun of the company.

If Snapchat had been monitoring the activity of its Twitter account, it would have seen an abnormal spike in complaints about a disappearing Snap Streak. Drilling down a little deeper, they would have seen users making fun of its auto-replying support bot and turned off the function before it went fully viral.

Unfortunately, not all incidents like this can be prevented, but running anomaly detection on your Twitter metrics enables your organization to react quickly and better control your brand’s image. When you provide outstanding support in the wake of an incident, you have the opportunity to snatch victory from the jaws of defeat.

Outlier detection and revenue opportunities

Applying outlier detection to Twitter metrics can do wonders for both the brand management front, and also alert you to new opportunities to make more revenue.

Consider a company monitoring mentions of its brand. There are several metrics being monitored in real time since the data coming in is segmented by geographic region. The company could have run out of one of their hot products in one regional warehouse. With anomaly detection, the BI team almost immediately learned that a celebrity had endorsed the company. By knowing about the endorsement in real time, the company was able to replenish inventory quickly, and take other proactive revenue enhancement actions like changing the product pricing, and even bundling the hot product with some of their slower moving offerings.

With positive buzz, you probably see a sharp increase in unique visitors on your company’s web page and online store, which could most likely turn into a sharp uptick in orders as well. Yet you probably don’t know why this is happening or just how to leverage this into an opportunity. An outlier detection algorithm would be able to spot all of these.

However, spotting outliers simply isn’t enough. Anodot’s AI analytics system correlates these outliers the instant they occur, points to significant anomalies, and allows you to extract actionable insights in real time, rather than otherwise overlooking seemingly meaningless outstanding data points.

Real-time is just in time

Outlier identification needs to happen in real time, as does the correlation of the detected anomalies, because opportunities can open and close very quickly on social media and your analysts just don’t have the time to wade through alert storms. Twitter runs at high-velocity, thus requiring real-time response to changes in mind-share and the emergence of new groups of receptive audiences for your marketing efforts.

Topics: Machine LearningArtificial Intelligence
Subscribe to our newsletter and get the latest news on AI analytics

Written by Ira Cohen

Ira Cohen is chief data scientist and co-founder of Anodot, where he develops real-time multivariate anomaly detection algorithms designed to oversee millions of time series signals. He holds a PhD in machine learning from the University of Illinois at Urbana-Champaign and has more than 12 years of industry experience.

This website stores cookies on your computer. These are used to improve your website experience and provide more personalized services. Learn more