Is the AI Revolution Real?

Hardly a week goes by recently without some Marketing Technology vendor promoting their new AI product. This is not an especially new trend. The AI buzz has been building since late 2016. Every major Data Platform is branding themselves as an AI vendor. Even Marketo are getting in the act with their upcoming Audience^AI capabilities.

Nor is this a trend confined to MarTech. Even Microsoft Office is getting an AI makeover. The likes of Microsoft, Google and Intel have devoted large proportions of their considerable R&D budgets to Artificial Intelligence for many years. The results of this investment are now being seen across the entire technology sector. All the major technology companies are promoting the AI revolution as the next big thing, and have been using their customer conferences to talk about AI, and its impact on their products.

There's one thing they're not telling you though - this trend is not a result of high profile breakthroughs in AI research over recent years. Instead, the driving force behind the AI revolution is cloud computing. This has allowed unlimited amounts of resources to be put behind the algorithms and data models that tech firms are branding as AI.

Few of these underlying data models are actually new. They are powered by the same algorithms that the same companies were calling predictive four years ago and machine learning eight years ago. As these models have evolved, they have become more accurate to the point that many firms can remove the probabilities and confidence scores from their output, and just give you the actual results.

Machine Learning

The technology used to enable all this is called machine learning. At its simplest, machine learning is a form of pattern recognition. It takes in large quantities of data and uses it to spot lookalike matches at a scale no human analyst could hope to match. The initial promise was that models built using this technique would become more accurate over time. This has proved to be true.

Take predictive scoring. The vendors in this space have been around since the late 2000s. Their business model is based on them reviewing your sales history and your marketing database, performing a data analysis with expert data scientists, and then using the resulting data model to give every contact a lead score. The key difference was that their model was vastly more sophisticated than a traditional lead score model, taking in many more data points. Over time their model used machine learning techniques to become more accurate, evolving to identify who your customers are based on the data you provide it. Now their models have reached the point where predictive data vendors simply tell you who your targets should be. The score is optional.

The main advantage such data platforms have is that they work by combining your data with publically available demographic and firmographic data, and then add web tracking and intent data from firms such as Bombora on top. All this data is then aggregated across customers to see if there are any general industry trends which can be applied to all scoring models.

This aggregation has allowed them to overcome the critical weakness in their approach, namely that to be useful machine learning requires a large, clean database and extensive training.

Applying machine learning to small databases gives inaccurate results because it relies on high data volumes to spot the patterns in the underlying data. Secondly, it is allowed vendors to compensate for the Garbage In Garbage Out principle. Much like a human data analyst AI is only as good as the data that it has been given. In some cases it can be worse as a human analyst will compensate for poor data quality using experience and gut instinct. A robust data model using the results of multiple customers can do the same thing. It can apply the data trends of successful organisations with good data quality to a low quality database.

Over time vendors have been able to broaden the market for predictive scoring beyond just the Fortune 500 to businesses of any size. More recently, they have also extended the scope of their services beyond scoring to full data management. In the process, they have rebranded themselves as Customer Data Platforms (CDP)

Account Based Marketing

The rise of Account Based Marketing (ABM) was the cause of this trend. A successful ABM strategy relies on a strong alignment between sales and marketing, and good data about your target accounts. AI vendors can't help with alignment but are ideally placed to help with data quality. Most vendors have offered account scoring since their earliest days, so are already integral to the process of target account selection.

ABM is about more than just accounts. In an ABM model, the funnel is structured around demand units - the individual teams in the enterprise that have the need and authority to purchase products and services. Most enterprises are comprised of multiple demand units, due to different divisions or different regional or national subsidiaries. Modelling this relationship in a typical CRM or Marketing Automation platform is challenging. It requires a structure more complex that of the traditional contact and account separation, and needs information about individual responsibilities and corporate hierarchies that few databases have.

Customer Data Platforms are the solution to this problem. CDPs have seen growing adoption in B2C for several years as a consumer-centric alternative to CRM systems. Aggregating all customer activity and profile data in a central database allows the detailed segmentation and one to one personalisation required for successful consumer marketing at scale. The leading CDPs such as Tealium then integrate directly with ESPs and Marketing Automation to launch campaigns on the back of this data.

The more in-depth personalisation requirements of ABM mean that the predictive data platforms are now expanding their capabilities, just as existing CDPs are adapting their platforms for B2B. It's a natural progression. They have all your data already because they use it for scoring and target account selection. They have all the available information about individuals and businesses too, as they use it as the foundation of their scoring models.

What they lacked until now is the ability to organise all this data into a comprehensive single customer view organised by account, contact and demand unit listing the full profile and activity history of each across all channels. To do this manually is a massive undertaking requiring teams of data analysts pulling in information from multiple sources, deduplicating it and aggregating it into a data warehouse. Data vendors have been able to tune their existing scoring algorithms to do this entire operation automatically in real time at scale. In doing so, they are offering the holy grail of marketing operations - a genuine single customer view that tells Marketers who their customers are and who their next customers will be.

The Real Revolution

Marketers can collect activity data across web and social channels, and then use it to drive highly personalised outbound campaigns. Abandoned basket emails are a classic example of the capability this provides. Website activity such as an abandoned basket in a web store is tracked by the CMS powering the store as part of its normal operations. This data is then passed back to the CDP in real time, and then onto Eloqua or Marketo so that opted-in customers can be sent an email listing the contents of their basket.

This is a revolutionary capability that gives enterprise marketers the ability to achieve true personalisation at scale. B2C marketers already use it to dynamically calculate the detailed demographic profiles of their customers based on activity data. Now it is being used to determine the specific interests and pain points of individuals within businesses to drive highly targeted messaging as part of an ABM strategy.

Modern marketing has always relied on a strong data foundation, yet in practice, few companies have the database to match this aspiration. Building and maintaining a clean B2B marketing database is hard. AI is going to make it a lot easier.