Use Einstein Social Insights
After completing this unit, you’ll be able to:
- Define use cases for Einstein features in Social Studio.
- Identify ways to improve sentiment analysis.
Before You Start
Before you start this module, be sure to take the Social Studio Basics and Einstein for Marketing Basics modules to learn key concepts about Social Studio and Einstein. The work you do here builds on the concepts and work you do in those badges. Before trying to use these features at your organization, talk to your account manager about requirements for using Einstein within Marketing Cloud.
AI-Powered Social Media
Social media marketing has become a vital part of an organization's digital marketing practice. But managing a social media presence today can be daunting. The high volume of social content makes it impossible to keep up with everything that people say about both your brand and your competitors. And that’s assuming that the content is just text—how do you accurately process GIFs or memes directed to you? Toss in the management of multiple social accounts and communities spanning your business, and sprinkle in the need to manage organic reach, and you’re facing a monumental task.
Luckily, Social Studio in Marketing Cloud can help. Social Studio enables you to maintain a holistic view of your social media reputation and conversations, uncover actionable data, identify influencers, generate leads, and automate tasks to free your marketing team and increase ROI. In this module, you discover how Einstein—artificial intelligence (AI) that’s built right into Salesforce—and automation enable you to scale your social media marketing efforts.
Let’s see how one company uses the AI-powered capabilities inside Marketing Cloud Social Studio to drive its social media marketing.
Track Your Sentiment Analysis
Meet Paulo, marketing specialist for Northern Trail Outfitters.
Paulo has started using sentiment analysis in Einstein Social Insights to understand and track the tone of customer posts and conversations in social media and to prioritize his responses. Sentiment analysis examines the tone and sentiment of conversations using a scoring algorithm, then scores and rates the language as positive, neutral, or negative. Paulo can train and refine this Sentiment Model to provide more accuracy as he goes along.
Here’s how Paulo works with it.
In Social Studio, Paulo clicks Manage Model to get started and add some basic information about keywords and phrases. Then he rates them as positive, neutral, or negative. As more samples come in, he can assign numerical values to them as well.
The Sentiment Model shows Paulo two main statistics.
- Sentiment Precision—This value measures the Sentiment Model’s accuracy in determining the correct sentiment. For example, if the Sentiment Model matches 15 posts out of a 100-post sample, the precision is 85 percent. This number may seem unusual, but in this case, a lower number value represents better precision.
- Sentiment Recall—This value measures how accurately the Sentiment Model scores neutral posts. Say a sample contains 100 posts, and 60 of the posts are either positive or negative in sentiment and 40 are neutral. If the machine learning algorithm classifies 50 of them as positive or negative and 50 of them as neutral, the algorithm has incorrectly labeled 10 posts as neutral and the Sentiment Recall is 50 out of 60—or 83 percent.
The Sentiment Model uses these scores to calculate an overall percentage called Sentiment Agreement.
We recommend getting your Sentiment Agreement score to 79 percent. How do you do that? You train the machine!
For example, this image shows Paulo's initial attempts to tune the Sentiment Model in Social Studio to better characterize the language people use in conversations about the Northern Trail Outfitters brand.
Paulo uses keyword scoring to train the Sentiment Model to adjust scoring of incoming posts that use specific phrases like “It’s fine” and “Love it!” People can use these keywords in various contexts, and the same phrase can have a wildly different meaning from one user to another. So Paulo works with the Sentiment Model to help parse out those differences.
Paulo adds phrases and scores existing phrases with number values to let the Sentiment Model know how he interprets those phrases, and Einstein takes it from there. He can agree with Einstein’s sentiment scores for phrases, disagree with them, or note that Einstein’s evaluations don’t make enough of a difference to change the results.
Review Social Information with Analyze Dashboards, Workbenches, and Reports
Social Studio provides sentiment analysis, scores the reach and effectiveness of social media influencers, and automatically detects and sifts out spam sites and posts in social channels, among other services. Einstein powers a lot of this information.
Paulo wants to know how he can collect all that information and put it in a format he can analyze.
Social Studio provides three ways to view this material for better evaluation.
- Dashboards show a summary of key factors for time periods between 1 and 90 days. These dashboards contain cards with the relevant information you need.
- Workbenches contain datasets with 1 to 93 days worth of data, providing you information on your social account conversations and segmentation types, such as keywords, sentiment, and more.
- Reporting enables you to pull reports from the Analyze dashboards on a daily, weekly, monthly, or quarterly basis to ensure you have the most relevant and up-to-date dataset. Social Studio formats the reports in PDF for sharing or printing.
What to Do with This Information
All of these tools provide you with great data, and you can configure the tools to provide even more relevant data as you go along. But Paulo knows that data without a plan to use it is just a bunch of information.
When it comes to getting the most out of your data, keep this guidance in mind.
- Go where you’re being talked about. Know your audience and analyze data from where real conversations are taking place, not just the major social media platforms everybody knows.
- Know who is talking about you and how important their influence is. Study what they’re talking about and leave random mentions for later.
- Always account for misspellings or more general references, not just your specific mentions. Take a look at your entire industry, and incorporate some common typos or errors to make sure you get a fuller picture of the conversation.
- Work with your sales team to pass along potential sales. You’re all on the same team, right?
- Make sure your analytics include sentiment analysis, active networks, total mentions, and pain points. Track these important data types over time to get an accurate picture of how you’re performing.
- Focus on the customer, not the channel. They’re sharing a message, and how it gets to you shouldn’t matter more than the content itself.
- Keep humans involved in the process. Einstein makes a lot of this work efficient and manageable, but human insight makes your data more useful. Rely on automation for the easy stuff, and then take the wheel from there.
In this unit, you learned how Einstein Social Insights helps you understand and respond to customer sentiment in social conversations at scale. In the next unit, we look at Einstein Vision for recognizing social images of your brand.
Rights of ALBERT EINSTEIN are used with permission of The Hebrew University of Jerusalem. Represented exclusively by Greenlight.