Machine Learning Applications in Marketing: a beginners guide.

Machine learning tools have scaled up in capability rapidly, building a huge media hype, and massive waves of investment. One of the largest low hanging fruit for these tools is digital marketing. This is the era of the brick and mortar retail apocalypse after all. As companies re-structure to focus on retail in the digital space, or spring up as solely digital distributors the days of old school marketing analytics are rapidly flying by the wayside. Digital marketing in a competitive field creates something with astonishing speed and volume, Data. Savvy companies like Amazon who led the digital first retail transition have realized the power of this data. Now we are coming into an age of marketing powered by machine learning that can take advantage of this large data for untold insights, and increased ROI for companies of any size. With a projected CAGR of 20.79% in AI marketing between 2018–2025 (source), it’s definitely a curve that executives should plan on getting ahead of. Let’s look at some of the use cases for machine learning solutions in marketing, and some of the companies who are building them.

The top of the sales funnel is exposure for lead generation and the key to that is Content. Ideally, it is high quality content that won’t get suppressed by one of the few algorithmic distribution vehicles that dominate the exposure market. Your team of copywriters, graphic designers, and increasingly videographers, is about to get augmented by the age of AI. Tools like Wordsmith used by the Associated Press can utilize natural language processing to write articles. Computer generated models (no not RNN’s or random forests, actual models) have already been implemented, and are even beginning to foray into influencer marketing. Companies like Wibbitz are using the vast amount of video content available on the internet to curate captivating videos based on text analysis. Their process is likely automated editing of stock videos to marry well with text, not true generation, it still makes some amazing videos. These machine learning programs certainly still require a human in the loop, but they are allowing content curators the ability to produce better quality pieces much more rapidly.

Now we have all this beautiful content curated with machine learning, how do we deploy it in the most effective manner? This is another great opportunity for the power of machine learning to increase the value of all the data generated, and maximize the efficiency of the now augmented marketing team’s efforts. In the Social Media realm players like SproutSocial and Movyl Technologies provide advanced analytics that not only gives insights, but they include deployment methods and create a data based feedback loop. Understanding which content is resonating with leads and how things like time of distribution or which channel it is deployed on affect engagement are key to marketing success, learning these insights and automatically deploying their insights for future distribution is timed saved for the marketing team and increased ROI for the company.

More frequently we are seeing companies moving to take advantage of analytics on this deployed content and the results when they are plugged into a CRM. Not only do you get marketing contribution as a KPI other insights can appear from the data as well. Using machine learning and statistical methods such as game theory and Shapley values CMO’s can derive more accurate marketing attribution metrics across an incredibly large number of touchpoints, allowing them to strategically invest their team’s efforts to ensure the best growth. This also makes accountability with the board a data>opinon driven discussion. Tools like this, that used to only be available to advanced marketers and teams using google analytics premium, are becoming increasingly easier to build and deploy.

So now your enterprise is running on a top of the funnel ML optimized marketing team and probably seeing results. Beautiful AI designed content that’s getting engagement, analytics based distribution, and content attribution analytics. Well, our job is done here let’s hand it off to the UX team and sales department. Not quite. You’re still marketing, that beautiful video distributed on the best platform at the analytically verified correct time isn’t getting its full value without conversion rate optimization. Your site needs to be designed from the ground up with the intent of pushing users towards their end goal. Of course, you’ve thought of this and your dev team has planned for it but is your UX+CRO fully optimized?

CRO is a major space where businesses are seeing improved results with AI based solutions. Often using customer’s conversion data during UX adjustments over time to train a neural network model will allow your enterprise to test for future UX adjustments likelihood of success using the model you’ve generated. This is neural network based multivariate testing. One of the drawbacks of this approach is the ‘cohort concept’, this is the idea that your model is trained using batched historical data. The insights derived will be generally useful but they are designed given the large herd of customer’s past choices, not truly optimizing conversion for everyone that lands on your product page. One way around this is to use what’s known as an evolutionary genetic model. Mimicking nature to weight the different variant’s effect on conversions over time, using many nodes that can adjust for a truly user optimized experience. If you have a large in house data science team you should consider this approach, if you’d like to investigate it as a service we’ve only found one platform that offers it, Sentient.AI

Looking at all the ML implementations on the marketing/sales funnel in this fashion can be kind of daunting. Each element has a unique use case that generally requires a specific set of data recorded and unique ML models to functionally optimize the data. Each element has a limited number of service providers, with some overlapping on their use cases and analytics quality. This puts CMO’s interested in applying more machine learning in the position of making the age-old question of buy vs build. How do you quantify investments in these elements and their ROI? Every enterprise is unique but when looking at the broad nature of implementing machine learning we can look at two incredibly strong solutions that encompass the vast-majority of the marketing ML implementations we’ve discussed above.

The buy:

Sentient Technologies. While they don’t offer content curation currently, they can optimize every many aspects of your marketing and sales funnel. Using a distributed evolutionary algorithm, they can provide true multivariate testing across your website. Effective conversion rate optimization ensures maximum marketing value.

The build:

DataRobot. If you have a small data science team, or you’re marketing team has a limited machine learning skillset, DataRobot can dramatically increase their efficacy and productivity. They provide an AutoML platform which offers assistance in all aspects of the data science pipeline. DataRobot offers data preparation assistance, automated model selection, and an easy deployment methodology. This allows your marketing team to research different approaches, test and tune different models, and deploy much more rapidly.

Hopefully this synopsis of machine learning implementation throughout the sales and marketing funnel and a look-into some of the service providers active in the space will help your enterprise or team take the next step into growing their capabilities with machine learning. Stay tuned for our next blog post, “from the funnel to the hourglass, how machine learning in marketing adds value after the close.” where we discuss machine learning applications designed to retain customers, decrease churn, and provide better products and services to ensure legacy branding.

Benjamin Cox —