Is Marketing Ready for Artificial Intelligence?

The Rise of Data Science as a Service

After a few AI winters, we may have finally entered the golden age for Machine Learning and Artificial Intelligence. The environment for the growth of narrow artificial intelligence has never been better – exponential growth in data volumes meeting algorithms and AI techniques solving all manner of business challenges. Coupled with the incredible efficiency of cloud storage and compute services, the world is well positioned to see some practical advancement in AI for marketing.

 “Deep learning, when married with big amounts of data really gets very accurate predictions."

- Frank Chen, Andreessen Horowitz

A question for marketers is, are we ready to embrace AI and Machine Learning in our brands, and how and where do we need to adapt our organizations to take full advantage of the available technology? Importantly, how do we sift through “AI” vendors and separate what’s real and effective, from the latest buzzword.

The environment for AI in marketing is promising:

• There’s plenty of room to grow in marketing efficiency and effectiveness
• Digital marketing spend (and corresponding datasets) represent a large and increasing share of budgets
• Culturally, marketing organizations are building data driven marketing capability, extending into the fields of Marketing and Decision Science

Below is some practical advice and insight for marketing teams thinking about the applicability and readiness of AI in their processes.  

The Magic is in the Channel 

We’ve observed that the promise of AI and Machine Learning has tremendous value within channel teams and their budgets i.e. Display, Search, Social, Email, etc.

Yes, AI and Machine Learning is being applied to multi-channel data sets, and informing budget allocation decisions aka Attribution, but the primary gains are in optimizing channel performance metrics, then letting investment follow performance for channel budget allocation. The promise of Multi-Touch Attribution is relevant but will improve as we close or can credibly infer data gaps in the customer journey.

Organizationally, this may come as a relief for marketing teams. The convergence of marketing technology, for example DMP+CRM blurring traditional “Advertising” and “Marketing” lines meant that marketing leaders had to grapple with how best to organize across their traditional channel teams. For example, anonymous audience data typically used for acquisition and awareness can be used to personalize email; and email data can be used to inform acquisition and awareness campaigns. This drove new thinking organizationally, and the emergence of cross-channel Centers of Excellence, or Innovation Pods.

It’s less disruptive to marketing organizations to think about AI optimizing within the channel, leveraging the deep expertise of existing resources, and accessing a unified data layer for blended audience, campaign and messaging data.

Evolving the Insights Model

A recurring theme coming from brands and agencies is that today’s marketing analytics and reporting offer only a present or static view of marketing performance, i.e. they are “stuck at a point in time” and offer little or no actionable recommendations.

Fortunately, the capability is there to revolutionize the way we think about generating and customizing insights. Forecasting procedures like Facebook’s Prophet help to accurately forecast and predict revenue and engagement, dealing effectively with outliers and customizing to the seasonality in your business, what we like to refer to as your digital marketing DNA

Data Scientists are tuning this model to bring personalized forecasting capability to most marketing data. For extra points, the forecasting procedure can be used as part of a data science model that generates recommendations such as budget allocation, while measuring the proposed budget move against a saturation curve, to maximize the potential opportunity and impact from the suggested move. In other words, budget allocation decisions are optimized for the greatest marginal utility, or to borrow from Modern Portfolio Theory – to operate at the Efficient Frontier of Marketing.

Transparency in Decisioning and Automation

We’re all a little wary of black-box automation. The availability of granular data sets and advanced visualization is ushering in a “no-excuses” environment where recommendations are supported by the corresponding analysis for human interpretation and approval.

For example, at conDati we maintain a long tail of dashboards that stand at the ready to support recommendations. For example, when our recommendations dashboard suggests a time shifting of advertising budget to Monday and Friday mornings, in a click a user can see a heatmap that visualizes the “why” behind the recommendation.  

With this data science approach, humans can resolve for broader context and help to narrow any data gaps.

For an interesting read on the importance of transparent modeling see Caruana et al. and their study on intelligible machine learning predicting survival rates in pneumonia patients (Friends Don’t Let Friends Deploy Black-Box Models).

Data Science as a Service

Data Scientists are a scare and extraordinarily valuable resource for most brands. Limited in supply, and at most companies, the attention of data science teams regularly extends beyond marketing tasks. We’ve observed marketing related data science tasks scheduled 18 months down the queue.

Data Science as a Service starts with providing a platform for the collection, blending and analysis of marketing data as a necessary step. Value is unlocked through the application of algorithms at scale and decisioning power to generate recommendations that optimize marketing performance. Perhaps platforms like these will facilitate the rise of what Gartner calls the “Citizen Data Scientist”, with estimates that the democratization of data science and augmented analytics will generate twice the growth for entities that embrace it.

I joined conDati because I believe we are the forefront of the Data Science as a Service platform. The building blocks guiding our path, that could be helpful to marketers thinking about AI in their business are:

• Data models optimized for granularity, access and elasticity
• Algorithms matched to marketing challenges and tested for scale
• Transparent recommendations and visualization
• Speed – in analysis and provisioning
• Relentless customer-led innovation

In summary, it’s the perfect time to build Machine Learning and Artificial Intelligence into marketing processes, and starting with a channel optimization approach means marketing organizations can introduce this capability step-wise into their organizations.