CONCURED Uses Deep Learning Artificial Intelligence to Score and Improve Your Content Marketing
Enterprise content marketers share a few common problems. They don’t know how their content performance compares to competitors or their industry. They often don’t have a complete picture of the content they have available. And they need help figuring out what content their audience truly desires.
All three problems have solutions thanks to artificial intelligence tool CONCURED. The solution uses natural language processing and deep learning to score and improve content marketing for enterprises. It can identify content gaps, conduct automated content audits, and recommend high-performance content topics. We spoke with CONCURED CEO Tom Salvat (LinkedIn) to learn how the solution works.
In a single sentence or statement, describe CONCURED?
CONCURED is a fast growing AI company with a SaaS platform that enables clients to carry out automated content audits, benchmark against their competitors and define success, discover content gaps that are trending upwards in popularity, and validate their content ideas with quantitative data so that they can engage their audiences and maximize ROI.
How does CONCURED use artificial intelligence?
CONCURED's SaaS platform uses natural language processing (NLP) throughout, and deep learning for some innovative tasks. The NLP is critical for understanding each content item, whether an article, blog post, video, podcast, etc. For example, a video mentioning eggs, bacon, smoothies, California, and Venice is really about brunch in LA (Venice Beach), not about Californian food in Italy (another Venice).
Accurate and relevant semantic understanding is fundamental to all subsequent steps in our pipeline, whether it's showing you at a glance what topics your competitors wrote about in the last week, highlighting current content gaps, or detecting emerging trends across industry publishers at a scale that would be too laborious to do manually. Clients have the power to drill-down from the high-level overviews to inspect individual content items, either a short auto-generated summary (more NLP) or click on a link if they want to go to the original source (as we don't store the content itself).
We match up these insights from content items with engagement data, such as social, website views and conversions, and email clicks, to create a kind of semantic topic ranking engine. “Topic” here means a cluster of terms that represent a contextual concept (like brunch in Venice Beach). This means that when our client is comparing content ideas, she can, for example, validate that if she produces content on Bitcoin and Inflation, it's likely to perform better than content on Bitcoin and gold.
We use deep learning for categorizing content from publishers into one or more sub-industries, and for company sites to automatically determine whether an item constitutes content marketing or not. The first is important since it means we consider only content for the industry relevant to our client. For example, a retail bank looking at the Wall Street Journal probably only needs to track content on personal finance and not the sports section.
From the other perspective, we find and include all relevant content, regardless of any categorization from the publisher. For example, if there was an article on the New York Times about a new medical device startup that had received funding from investors, should this only be classified as healthcare? We use fuzzy classifications based on training over multi-industry labelled datasets to handle content that spans multiple industries.
The second use of deep learning, what we call our content marketing detector, is important because every client and competitor site is different, and, unlike a one-off content audit, our computer systems are running continuously, at a scale where it's not feasible to manually locate content marketing on the websites we track.
So we developed a web crawler that identifies candidate content marketing pages. This finds a lot of content on websites, not just the content marketing, so it then hands-over candidate pages to the content marketing detector to check whether it really is something we should be including when generating insights for the client, or whether it's something else, such as a product description.
In the future, our system will learn by itself which regions of the site contain the relevant content, which will reduce the number of candidate pages it parses and allow it to find new content even faster.
Internally, our development team is super-excited about the experimental results we're seeing from a technique called reinforcement learning: it compares candidate topics generated by our recommendation system with their subsequent performance in the real world: e.g. did they initially accelerate then fizzle out, or did they show sustained growth? The idea is that the recommendation system learns from its past successes and failures, so that it improves over time. This is like the technique Google's DeepMind employed to train computers to play computer games and beat the world champion of Alpha Go.
What do you see as the limitations of artificial intelligence as it exists today?
AI is currently good at answering well-posed questions or performing well-specified tasks, but it will be a long time before AI can ask good questions.
What do you see as the future potential of artificial intelligence in marketing and sales?
In a future ideal world, we as consumers will be exposed to exactly the information we want, exactly when we want it, even if it was a latent desire that we wouldn't have been able to articulate until we saw it, like a perfect butler working in the background who anticipates our desires.
This includes product information that is so well-targeted that we actually want to receive it, and exactly fits our buyer persona and the stage of the buyer journey we are in. In fact, it’s so good that it will know our preferences and be able to order the product for us automatically, at exactly the price we are willing to pay, with a money-back guarantee if we, the humans, have a different opinion.
Such a future would save ineffective marketing dollars on the part of companies, as well as hours of product research and decision-making on the part of consumers. Of course, the big issue here is trust. How do we know that our opinions have not been manipulated by the information bubble we live in, and whether we are being offered a competitive price, not just the maximum we are predicted to accept based on our demographic profile?
What makes CONCURED different than competing or traditional solutions?
Most tools that help content marketers with ideation are keyword-based, and designed to help clients with SEO to game the ranking algorithms used by the major search engines (which are getting more sophisticated at penalizing this, by the way). CONCURED's solution is intended more for optimizing engagement, whether measured as social shares, video or page views, newsletter signups, etc., or a combination. It's also contextual (as in the Bitcoin example) and semantic (Venice Beach vs Venice, Italy). And by the way, CONCURED works on content in any language! The first thing our clients appreciate is time saved. Our technology helps content creators get their message closer to the middle of the bull’s-eye better and quicker than ever before.
Who are your prototype customers in terms of company size and industries?
Our clients currently include enterprise companies (thousands of employees, billions in revenue) primarily in the B2B technology, finance, manufacturing, and healthcare industries. We also serve digital marketing agencies, and are about to release a version of the product for agencies' and their clients.
What are the primary use cases of CONCURED for marketers and sales professionals?
Clients use CONCURED’s SaaS platform to audit their own content, benchmark their content against their industry peers to help define what success looks like, help companies better understand their content assets, what they are about and how they are performing, and to understand quickly which topics are resonating, trending and why, so they know what to create content on next and can validate if their ideas are likely to succeed.
In marketing we define these use cases as:
I don’t know how to validate if my content is performing well or not, or how it compares to competitors or my industry.
I don’t know what content assets I have, how they are performing, or if I can re-use or adapt what I already have.
I need help identifying what my audience wants content on, and validating if my ideas will resonate.
Any other thoughts on AI in marketing, or advice for marketers who are just starting to explore the possibilities of AI?
Treat any “black box” results or recommendations with skepticism and don't believe these until you have seen the rationale and evidence behind them. Finally, as marketers, let’s remember that the motivation for gaining a better understanding of our current and potential customers is so that we can ultimately serve them with more relevant and timely information. We also fully agree with the advice that marketers focus on one or two AI use cases for things that are time-consuming and that It can be your competitive advantage as a marketer.
About Paul Roetzer
Paul Roetzer (@paulroetzer) is founder and CEO of PR 20/20, author of The Marketing Performance Blueprint and The Marketing Agency Blueprint, and creator of The Marketing Artificial Intelligence Institute and Marketing Score. Full bio.