What is Big Data?

First, some definitions. Big data involves computationally analyzing extremely large data sets to reveal patterns, trends, and associations; especially those relating to human behavior and interactions. It is used in everything from predicting stock performance to seasonal buying behavior to helping the NSA know whether your post about “blowing up the joint” refers to your bomb-making or DJing skills.

Every human who uses any form of digital communication generates data constantly, both about themselves and about humans in aggregate. Big data refers to the ability to find, sort, and make sense of this ocean of ones and zeroes. It encompasses structured, semi-structured, and unstructured information, both human-generated and from sensors, machines, and public records.

Structured data generally means information residing in a fixed field within a record or file, such as that found in spreadsheets and relational databases. Information that’s tagged to show some elements within the data, such as metadata in email or photos, is semi-structured data. Unstructured data meanwhile, includes content such as untagged text, images, audio, video, and so on.

Big data can also include demographic or psychographic information about consumers. Think product reviews and commentary, blogs, content on social media sites, and the digital exhaust streamed 24/7 from mobile devices, sensors, and technical devices.

Defining Artificial Intelligence

The definition of AI is more nebulous because what is considered AI is constantly changing. One way of thinking of AI is as intelligence exhibited by any device that perceives its environment and takes actions to maximize its chances of achieving a goal. Another instance is when a machine mimics “cognitive” functions such as “learning” and “problem solving”—also known as machine learning.

Capabilities currently classified as AI include understanding human speech, self-driving cars, and interpreting complex data. As the technology improves however, capabilities once defined as AI are removed from the definition. For instance, optical character recognition is no longer perceived as artificial intelligence, but as a routine technology. The same with GPS navigation systems.

Another way of thinking of AI is that it merely refers to algorithms we don’t fully understand yet.

The implications of this technology could feel as if we’re living in an anime cyber thriller, hurtling towards some utopian (or dystopian) future—the finale isn’t clear yet.

Applying Artificial Intelligence to Content Marketing

We’re barely at the beginning of applying the technologies of AI—such as natural language processing and machine learning, to content marketing. Artificial intelligence today has very narrow applicability. It is typically built to do one complex thing—usually a complex data-driven thing—more efficiently than a human can do it. Over the last two to three years, several technology startups launched that are purpose-built for individual content tasks. These companies are trying to apply AI capabilities to a task that’s very time-intensive for humans, to make it more efficient and effective.

AI technology is still raw, and arguably pretty stupid. “It’s basically no smarter than a preschooler in a lot of cases,” says Paul Roetzer, the Cleveland based founder of the Marketing AI Institute, a resource for those interested in applying AI to content marketing. “But it can be trained to do one thing exceptionally well, almost super-human, and that’s where a lot of the companies are focused.”

AI as it applies to content could involve any content-related task we do as marketers, predicts Roetzer. Whether finding keywords, picking blog post topics, determining what to share on social media, writing hard copy, creating landing pages, or writing headlines. Everything we do that requires us to manually create a strategy or plan, develop content, and promote it can be automated or enhanced with artificial intelligence.

It pays to be skeptical about anything touted as ‘AI’ however. Thirty years after the 80’s, AI is once again a buzzword. Many software tools that call themselves AI aren’t really. They’re more like 80’s expert systems that merely rely on hard coded rules.

Who’s Making it Happen?

The main software leaders in the AI field are the obvious players, such as IBM’s Watson, Google, Apple, Amazon, Microsoft—all the big tech companies.

However, this field is young. Think Internet in the early 90’s young. There are few content AI or marketing AI platforms as yet. There are mainly tools that perform specific tasks in equally specific contexts. You currently need multiple different tools to build AI into a content strategy. One related challenge marketers will face, argues Roetzer, is that the majority of those tools won’t be independent companies within 18 months. The big companies are buying up promising AI upstarts because there is a lack of talent that can actually build AI solutions.

What About Purely Marketing Companies Developing AI?

San Francisco’s Salesforce has made massive investments, buying multiple AI companies for a cool couple of billion dollars. That technology was then built into Salesforce Einstein, which launched in September 2016. San Mateo-headquartered Marketo now has predictive content recommendations, although this technology seems to have been mainly from acquisitions.

The Marketo and Salesforce marketing automation platforms use lead scoring technology. It examines which content someone’s consumed, how much they’ve consumed, and their demographic and firmographic to give them a score. Based on this data it automatically sends them to sales, and starts sending them relevant content. It will then send them different content based on what they continue to consume.

“I don’t know if it’s quite intelligent though,” muses Curata CEO Pawan Deshpande (Curata is my employer). “Because it’s basically a hardcoded set of rules determined by the marketer. It’s not learning from past performance and evolving and adapting. Marketing automation doesn’t really have much in the way of machine learning in it at present—but it certainly could.”

Boston-based Curata has been around since 2007. It uses natural language processing and machine learning to power two software platforms. Curata CMP offers full funnel predictive content analytics and editorial calendaring. Curata CCS is a curation platform that discovers content, filters out noise, sanitizes text, extracts metadata, automatically summarizes, and makes it easy to review, curate, publish and promote content. It uses machine learning to self-optimize and learn user preferences to find better content.

Other marketing companies using AI in content tools include Manhattan-based opentopic, which has a personalization project called Sia built on Watson. Austin-based OneSpot uses image recognition and natural language processing to automatically tag and categorize content and images. It then uses machine learning to automatically surface relevant information to the right visitor on the right channel at the right time. Conversica is headquartered in California with offices in Missouri and Washington. It uses AI to automate the lead contact and qualification process, and identify which leads intend to purchase and are ready to buy.

Storytelling Machines

Natural language generation refers to a computer using data to produce natural language as a human would write it. Machines write 100 percent of Associated Press earnings reports, along with some of their basic sports news stories. Two major players in this area are Narrative Science, headquartered in Chicago, and North Carolina-based Automated Insights.

Persado has $66 million in funding and offices in New York, San Francisco, Chicago, Rome, Athens, London, and Germany. Persado uses natural language processing and natural language generation to automatically create Facebook ads, landing page content, and email subject lines. They are unique in that they are in the creative realm of using machines to create content that isn’t data-driven.

There are dozens of other players in this field.

So What is Content Intelligence?

Content intelligence may draw on artificial intelligence and big data, but it is neither of those two things. It’s the systems and software that transforms data into actionable insights for content strategy and tactics. Content intelligence means having the full context of an individual piece of content. Not only that—but the whole corpus of content, to make better decisions about anything pertaining to the content in question.

Forrester analyst Ryan Skinner defines content intelligence as “technology that helps content understand itself—what it’s about, how it speaks, how effective it is at accomplishing certain goals, what emotions it calls to mind, etc.”

So what does having the full context of a piece of content mean? It’s understanding what the content is, what it’s related to, how it’s performed in the past, and how related content has performed in the past. This includes understanding how competitor content may have performed as part of the broader context it sits in. As well as other content it’s competing with in search engine optimization and search engine results pages.

Content intelligence means understanding everything there is to know about a piece of content. And to the extent that the past can help predict the future, using that comprehensive understanding to guide decision making for that piece of content. It doesn’t necessarily have to include the automation and execution of those decisions.

The Evolution of Car Navigations Systems

Once you have a content item’s full context, the next step is automation and execution of certain tasks. Think of the evolution of car navigation systems. For decades when people wanted to get somewhere, they would look at a paper map, then approximate a route.

But the information a map gives you is woefully incomplete. It just doesn’t give you much context for your trip. For example, you don’t know any of the speed limits, or where there’s stop lights, rest areas, or gas stations. Forget about real-time changing conditions such as roadworks or traffic. So people would plan out trips just based on a map and guesstimate most things.

Then GPS comes along and lets you pinpoint exactly where you are on your journey. (Although it didn’t necessarily tell you which path you needed to take.) Next came the navigation system. It could suggest an optimal route, absent traffic information. It could say where you are, where you want to go, and some good ways of getting there.

The next evolution was Google Maps and Waze. These platforms have real-time traffic data—not to mention satellite navigation such as Sirius, offering a real-time dynamic understanding of traffic flows and patterns.

The next step beyond is automation. It doesn’t just give you intelligence on what to do and guide your decisions. It actually does it for you: i.e. the self-driving car. We haven’t quite conquered this as yet—but we’re close.

Content Marketers Are Still Using Maps

Many content marketers are now at the very first stage of using maps when it comes to content strategy. They don’t really understand where they are because they mainly look at ‘vanity metrics’ such as pageviews, social shares, and so on. These are top of the funnel metrics that offer a rough approximation of how content is doing. But they don’t measure content lower in the funnel, or show content’s impact on your business.

Some marketers today however, go beyond vanity metrics. They connect all the dots by pulling in significant quantities of data that’s hard for a human to compile, let alone compute. These companies pull together data from many disparate sources and apply that to content to get a fuller understanding, manifested as analytics and reporting. They can look at a report and get comprehensive data at a glance. They can then make an intelligent and informed decision about a course of action, which was formerly not possible.

These marketers understand the business results they’re driving exactly. This includes the leads, revenue, and sales pipeline a particular piece of content is generating. This is where the technology is at today.

Content intelligence technology is exciting relative to where we’ve come from, but it’s still nascent. At this stage it’s mainly collating full context data to provide intelligence. “I think we’re really at the data aggregation stage,” says Deshpande. “Simply collecting all the data about content from disparate systems is a challenge. After that, machine learning systems have the full context to make intelligent recommendations.”

The next step for this technology will see it do the majority of computation and inference to determine the best course of action, based on a given set of data. Currently this is where marketers apply their intuition. This could include choosing whether or not to refresh an evergreen article. Or whether to spend money on paid promotion for an article that’s popular and could go viral. The best course usually isn’t obvious however, requiring a lot of data exploration digging, which machines are much better at.

Which Problems Does Content Intelligence Address?

Every marketer alive today proclaims how “data driven” they are. But if you look at how content is utilized, it’s still mostly based on intuition and guesswork. At marketing conferences, if you want to know what content to create you’ll always be advised to go ask your sales team. Or to go search online and see what questions people are asking. Or to look at search volume. These are all good inputs, but they’re mainly based on what’s worked in the past.

What to Share

Sales teams typically share content based on anecdotal feedback. You’ll often hear one sales person yell out on the floor, “Do we have a piece of content about analytics that I can share with this prospect?” And someone else will say, “Yeah, here’s a tasty piece.” These anecdotes reinforce behaviors. So if something worked well and someone hears how a content item helped them close a deal, the next sales rep likely repeats the pattern. Demand generation teams will do AB-testing on emails when looking at what data to share. But beyond that a lot of it is just very anecdotal, or looking at very rudimentary metrics.

Jeff Brewer, Lead Software Engineer at Lux Research and a former quant in futurist Ray Kurzweil’s hedge fund, believes content intelligence can really pay dividends when paired with consumer behavior models. “This means moving beyond content suggestion by anecdote to using consumer models and characterizing content from previous interactions to suggest what to share next,” he says. “Besides improving sales outcomes, a data-driven suggestion platform can evaluate new, untested content to determine efficacy with minimal risk to the sales pipeline. This helps both sales and marketing departments hone their craft. These models can be as unique as every company’s set of products, content, and markets. AI based content intelligence customizes these models for each company.”

There are many other channels where most decisions about how to utilize content come down to intuition. These include social channels, websites, advertisements, and so on. This is where content intelligence can add the most value. It can make those decisions informed, and even automate some of those decisions in a scalable manner.

What to Read Next

On a web page you’re usually presented with the most popular or most recent stories. With content intelligence, expect instead to see stories based on your previous browsing history and position in an organization’s hierarchy. Along with what your title is, what you’ve consumed in the past, what other people in your organization have consumed in the past. Even which content—or certain pieces of content, shared in succession—has the highest conversion rate at the stage you’re currently at in the consumption cycle.

Those are just a few factors. But there are literally hundreds of possible factors that can be fed into an algorithm to find the right weight for all those inputs. So you know exactly what context something has, and what the right piece of content is. Content intelligence helps utilize your existing content inventory most effectively.

Which Content to Update

Another way content marketers can use their inventory more effectively is by knowing when to refresh evergreen articles. Say a particular article or blog post performed really well, but it hasn’t been refreshed in a year. You are automatically sent a prompt for this article to be refreshed and shared.

There’s a constant stream of things a content marketer can do to be a lot more effective, that don’t take up much time. But it’s hard to figure out what they are. Content intelligence can surface that information in an easy-to-understand, and easy-to-act-on way. So you can login every day, or every week, get some really useful feedback or advice, make those changes, and see specific growth in certain areas.

You can do this at every different touch point with content intelligence. Your website, a sales person on the phone or emailing with a prospect. Any time you have a touch point with a customer, you can show targeted information that’s highly valuable to them.

Which Content to Create

The content creation process is another area to be optimized. Content intelligence can make recommendations about what you should be creating based on what performs well, or what your competitors are doing. And the recommendations can vary based on what your goals are. So you could say, “we have a goal of a certain number of leads or pageviews generated for this quarter.” And you could receive recommendations to help you achieve that. The recommendations will differ based on the goal.

While machines can mainly only create data-driven content for now, that’s evolving quickly. IBM partnered with a movie studio and created a video trailer using AI, which took over the creative process. Coca-Cola has used AI to generate TV adverts, selecting the music and creating scenes. If Pepsi had content intelligence, they may have avoided uniting the entire Internet in opposition to their Kendall Jenner ad.

How to Promote

Think about how hard it is to know what to promote, when to promote, and where to promote it. Content intelligence will enable highly personalized, cross-channel promotion that humans are just not wired to do.

“I am a big believer that the first major AI platform to be built, will probably be the first billion dollar AI company, because it will completely redefine marketing strategy,” says Roetzer. “Think about a company trying to spend a hundred thousand dollars, or a million, or hundred million dollars on marketing. Humans are incapable of figuring out the best way to spend that money. Given all the channels, all the need for personalization, all the different possibilities of what you can do with that money, there is just no way… the greatest strategists in the world together can’t figure out the optimum way to spend a million dollars. But AI can. It’s the hardest problem to solve, but that’s the one that is going to change everything.”

The Solution to Content Shock?

Tennessee-based marketing strategist, speaker, and author Mark Schaefer coined the term content shock in 2014. It describes the phenomenon of an ever increasing arms race to produce more, and more compelling content. This content is seldom personalized, and consumers’ attention span is resolutely finite. But marketers keep producing more and higher quality content for a slice of a pie that’s not growing.

Content intelligence is a significant competitive advantage for organizations wanting to overcome audience content shock. They have the insights to produce better, more engaging content. And more intelligent means of distribution to get content to the right sections of audience where it’s most compelling. Content intelligence enables you to present the right content to the right person, every time.

For a B2B marketer, downloading an eBook on most sites means you’ll likely be presented with that same eBook as a call-to-action if you revisit two days later.

If you’ve created enough great content however, with content intelligence you can truly tailor the experience someone has with your brand and site to an individual level. It’s the Netflix or Amazon approach. You don’t feel there are too many products on Amazon, because you only see the stuff that matters to you, based on your behavior. If content is the same way—if you only see what’s relevant to you all the time—it doesn’t matter to you how many millions of eBooks are out there.

Who Are Some of the Players in Content Intelligence?

The term content intelligence has been around for over a decade. But the companies building the technology are generally less than ten years old.

Curata CMP pulls in data from many different sources. This including the content itself, social metrics, traffic metrics, lead metrics, marketing pipeline metrics, sales pipeline metrics, and revenue metrics. It shows the precise business impact of every piece of content you produce.

Headquartered in Brighton, England, BuzzSumo pulls together social sharing data for all content across the web. Idio has offices in London and New York, and uses machine learning to analyze content, marry it to a consumer profile, and serve personalized content. Washington D.C. based TrackMaven focuses more on competitive analysis. It helps you track how your marketing performs against competitors, peers, and industry influencers. Conductor Searchlight has offices in New York City and San Francisco. It shows you how your content is doing in SEO and how your competitor’s content is doing. Florida-based Ceralytics is a platform for creating, analyzing, and promoting your most effective content.

Where is the Technology Moving Next?

Content intelligence is increasingly looking at what’s driven business results in the past, and advising you what to do going forward based on your goals. Or telling you where you will end up in terms of those metrics. It will automate parts of this process. This year expect a focus on helping to optimize your current content inventory for immediate benefit. Longer term, expect more of a focus on external data sources.

The past doesn’t always predict the future. Content marketing is inherently a very creative process, and thinking outside the box should not be underestimated. So expect the more mundane, repetitive, unskilled parts of content marketing to be automated by content intelligence. But not the need for experimentation.

How to Prepare for Content Intelligence

The insights content intelligence provides are only as good as the amount and quality of the data you have. To leverage this technology for competitive advantage requires a sufficiently long history of well-structured, well-maintained, trustworthy data. Deshpande offers an example of the importance of this from when he worked at Google in 2005.

“Back then Google Translate worked best for Chinese and Arabic,” he says. “And it wasn’t because Google had the best algorithms for those particular languages—it was because they had the most data. The Department of Defense translated so many documents in those two languages that Google had the most training data to input into the system. It really demonstrates how data is more important than the algorithms behind it.”

Know What You’ve Got

One of the most important steps an organization can take is to conduct a content audit of inventory. Best practice is to do this at least once a year, and it’s also vital to effectively utilize content intelligence. This audit should capture the content text, images, metadata, and other associated attributes invisible to the content consumer. For example: the persona, the buying stage it’s designed for, the vertical it’s for, and who wrote the article—especially for organizations using ghostwriters.

Given sufficient examples/inputs and their resulting outputs, a machine learning algorithm learns which inputs correlate with which outputs. Then it can predict which inputs correlate to positive outputs, and optimize for a desired output. Content marketing inputs are those in your content inventory. Performance related data makes up the outputs. This means things such as leads generated, revenue generated, social shares, traffic data, variants such as time on page or bounce rate; whatever you’re trying to optimize for.

You can then examine marketing pipeline metrics to see how these inputs turn into opportunities for sales. For example, which leads consume which content, how often they consume it, when they consume it, and so on. Did someone consume something when they were an opportunity, at purchase decision, or much earlier? Which of these pieces of content then led to revenue, and how much revenue? Then there’s outside, broader context data such as competitive data, and related topics and trends at the time.

Promotion and Distribution

The last phase is understanding how content is distributed and promoted. It can be hard to tell why a piece of content did well. Is it because it’s fundamentally strong, compelling content, and therefore got traction because people are sharing it? Or was it just heavily promoted mediocre content?

For example, on the homepage of Google News they used to have a ”recommended stories” section. These stories got the highest click-throughs, so the team felt they were making good recommendations. But it turned out they were getting the highest click-throughs because it was the first thing on the page. It was the problem of display bias.

Collecting, Storing, and Cleansing Data

Again, clean data—and lots of it, is imperative. You can’t just go out and buy a machine learning algorithm, flip it on, and start seeing immense value. Any time you use an AI application, with machine learning in particular, you have to teach the machine by giving it data.

Cleansing your data means de-duplicating it. I.e. making sure there aren’t two contacts with same email address, or the same contact with two different email addresses. Simple things like that ensure your data is correct. There’s a tool called IBM Watson Analytics which allows you to upload a dataset to find out how good your data is.

Marketing automation systems warehouse content consumption data that pertains to leads further down the funnel. But vendors don’t currently retain that data well for storage reasons. For example, Marketo only stores web activity data for 90 days, after which they start purging the data. Oracle stores your web activity data for 25 months.

The longer your data goes back, the better. So it will pay to pull data outside of your marketing automation system and build a data warehouse. Other systems pull social data. It’s easy to go and see how a piece of content is doing today in terms of social shares. You can look at a snapshot and say, “Ok, I got this many shares on LinkedIn.” It’s harder to go back and see how the numbers changed over time. So the sooner you start cataloging and storing that information the better, so you get more historical context.

Google Analytics is pretty good for storing historical data, but even that has issues. If a company has over 500,000 pageviews in a given time period, Analytics starts sampling data. It’s too much for their channel, so you’re don’t actually get truthful data.

Every system has its downsides, and it’s important to know what the downsides are. To avoid being hindered by those limitations, it will pay to store data outside many systems.

Clippy’s Revenge

There are some applications around now that are about as useful as Microsoft Clippy. Clippy was the famously intrusive Microsoft Word paperclip that used to give unhelpful, obvious suggestions. (Microsoft euthanized Clippy in 2007.) Many content intelligence tools are arguably still at the Clippy stage.

While everyone likes to make fun of Microsoft, they’ve morphed Clippy into other technologies. Now if you use the same language or phrasing over and over in a Word document, it offers subtle, non-obvious stylistic suggestions beyond just spelling and grammar. Rather than being an annoying paperclip in the corner, it’s built into the workflow. That’s where content intelligence is working towards to provide value.

The adoption of content intelligence will likely come down to a question of trust in the technology. Firstly trusting that the data is right. Then that the insights are right. Then trusting that the system’s suggestions will actually help rather than hurt. And finally trusting the system to automatically perform the suggested action on its own.

Think about this analogy. In the 1920’s when elevator technology first started displacing elevator operators—that was a big deal for elevator passengers. Many people would take the stairs because they just didn’t trust an elevator sans operator. This dynamic exists with people’s attitudes towards self driving cars. Content intelligence needs to overcome the same level of trust. The technology needs to develop and provide insights and automation. But even with that, it may take even longer for folks to really trust it to make the right decisions.

The Content Intelligence Disruption

We’ve reached the point with certain systems where we trust machines more than we do ourselves. It doesn’t make sense for us to tell an elevator what the best routing is to get to a floor. Many people now trust a GPS navigation system over their own intuition in most cases when driving cars. This is the point content intelligence needs to reach before we see wider adoption.

Content intelligence right now is reducing or removing the need for freelance writers who produce low level copy. It may do the same for certain marketing operations and demand gen positions. (There will be plenty of software engineers kept in gainful employment however.)

It will likely take another two years for the intelligence to offer more consistently helpful insights. And perhaps another three years to gain wider adoption and trust. That said, content intelligence, like all new technology, will offer first movers a significant competitive advantage—whether vendors or users. It will also eventually devolve into a more utility-like function as economies of scale allow full industry penetration. From now until then however, marketers ignore this technology at their peril.