Unmetric Unveils Prediction Engine For Social Posts
Social intelligence platform inspired by Asimov’s “psychohistory” attempts to forecast engagement level of text-based posts.
Unmetric’s purpose in life has been analyzing past social data and figuring out what your competitors are doing, so you can adapt your strategy.
Today, for the first time, the New York City-based social media intelligence platform is looking to the future with a new Predict product on its platform. The new offering estimates how much user engagement your brand’s social post will generate — and ways to tweak it to get more.
“We’ve never done anything predictive” before, CEO Lux Narayan told me. “Everything so far was like a rear-view mirror.”
He pointed to Isaac Asimov’s Foundation series of novels as an inspiration. In those classics of science fiction, “psychohistory” plays a starring role as the ability to make predictions about large social groups based on past behavior. In one form or another, that’s the same idea behind the current wave of predictive analytics and marketing.
To develop the prediction engine vertical by vertical, Unmetric first took about 6,000 Facebook brand-generated posts that related to automobiles. Analysts tagged 4,500 of them for such characteristics as structure, tone, intent and context, creating what Narayan called “a fantastically rich data set.”
The 4,500 were then processed via machine learning, in an effort to determine the hidden clues that forecast the highest engagement levels for sharing, commenting and likes.
Based on relationships discovered in this “training set” between content elements and engagement performance, the engine then tackled the unsorted remaining 1,500 posts in an effort to “predict” their engagement. Since they had been posted, of course, the predictions could be measured against their actual engagement.
At first, Narayan said, the error rate was high, about 45 percent. By fine-tuning the engine, it has been brought down to what he considers a more acceptable 14 percent.
To use Predict, a marketer submits a post and answers multiple-choice questions, like whether the brand’s post is part of a campaign and what kind of campaign, if any.
The analysis is currently text-based, so the qualities of any image in a post are described by answers given by the marketer. Predict offers a visual indicator of how much user engagement the post is likely to generate and provides up to four tips on ways to improve the post’s chances, such as in the following screen:
At some point, Narayan said, Unmetric expects to utilize a third-party machine vision solution to scan the image directly, but for now, the system’s understanding of the accompanying image is entirely derived from the marketer’s answers.
A user can also manually enter social posts from a competitor, so that a brand’s own posts can benefit from the kind of competitive social intelligence that Unmetric specializes in.
At launch, Predict is available only for Facebook posts, although Narayan said Twitter is coming soon, with support for Pinterest and Instagram “in about four months.”
Unmetric says it doesn’t yet have field results on how effective Predict is. Narayan acknowledged that predictive tools for social posts exist on platforms like Adobe’s Marketing Cloud but added that those are for a brand’s own posts and are not designed for analysis of what works in competitors’ posts.
There are also prediction tools for social media trends and others that purport to forecast if a post will go viral, but he said they’re not focused on competitors, brands’ own posts and engagement.
Opinions expressed in this article are those of the guest author and not necessarily MarTech Today. Staff authors are listed here.
About The Author
