The Million Dollar DevOps Question
PHOTO: Martin Pettitt

Back in 2016 I wrote an article exploring the four questions marketers must ask when considering the replacement of an analytics solution.  

2016 might as well be 1816 when it comes to tech, because new sources of data have given rise to new customer experience metrics whose demands far exceed the analytics solutions I had in mind at the time. Machine learning tactics now underlie data usage, as well as any associated analysis. Marketers have to step up.

Marketers worried about gaining budgetary approval for new solutions should feel relieved that organizations are recognizing the evolution of reporting and analysis choices. So too are vendors, judging by the over 6000 martech solutions they've launched to date, according to Scott Brinker's annual Marketing Technology Landscape supergraphic

The challenge marketers face then is how to cut through the quagmire of jargon and over-selling of features that even the most savvy marketers could fall prey to. Believe me, the "shiny bauble" effect is real.

It's possible to select a few areas to focus on that have a far-ranging impact on how a data-driven project is managed. Selecting around these ideas can help triage which features will best blend data, analytics and machine learning development into a technology stack.

What API Connectors for Data Sources Does the Platform Offer? 

We've all heard of APIs (application programming interfaces), but let’s reflect on what an API's main function would be in this context.

An API is essentially a doorway through which data can pass in a controlled manner. The passage is a controlled digital access to an application — the machine that solves a problem for you. The interface is how that machine takes input and issues an output.

When viewed in a machine learning context, you have to think about the application as the data source that we want to access conveniently. The interface, consequently, becomes the way we query data from that source, as well as where the output is delivered. 

So when looking at APIs, a marketing team should think about what kinds of interfaces the analysts can use intuitively. An API eases access to the data, but in a secure, controlled way, and in a way that makes it convenient to create advanced models, or make adjustments to existing ones. For example, there are libraries in R programming that are designed to access data from various kinds of sources, such as an SQL database, a cloud solutions like Azure or AWS, or a platform like Twitter or YouTube.    

Related Article: Here's Why Every Modern Marketer Needs R Programming

How Does an Integrated Development Environment Help Communicate Updates?

You'll see integrated development environments (IDEs) topic come up a lot among the developer community: it’s the software that presents a single program from which all code editing can be done, be it websites, apps or chatbots. When it comes to machine learning, IDEs are essential, whether you're talking about R programming in R studio or using Python within Microsoft’s Visual Studio Code. 

While marketers may not be programmers, they should take interest in how well the tool managing collaboration on code revisions and on data blending. This is where version control can be handy.  Understanding how updates are grouped together can foster better communication among teams that are increasingly reliant on code to identify development and production issues.

Related Article: Developer Collaboration Tools Beyond Slack

What Customization Is Available for Data Visualization?

A customized data visualization can sound like an oxymoron: aren't all data visualizations customized for different data sets to begin with?

Honestly, yes. But some visualizations are handled through a solution with its own proprietary user interface. While that may make it convenient as a quick way for analysts to simplify large amounts of data into more basic, understandable forms, it can also limit the insights provided. Distilling a large number of metrics into a handful of pictorial representations can sometimes lead analysts to reach unfounded conclusions or completely neglect certain signifiers which otherwise would change their assumptions about the data and its relationships.

Proprietary user interfaces work fine in some instances. The reports in web analytics solutions are proprietary interfaces, acceptable for general distinctions in a digital marketing analysis. But in many instances statistical analysis for machine learning requires more imaginative options to display results. Therefore, marketers should seek agnostic visualization options that provide a wide assortment of data input and graphing options.  

Keeping these few tips in mind will help marketers cut through a lot of the vendor noise so they can make good decisions on what platforms suit their organization's needs. The choice will have a long-term impact on how your solutions enhance the digital experience for your customers.