Find the right predictive solutions for your email marketing!

carlos-muza-hpjSkU2UYSU-unsplash-200

In my last blog post, I provided a quick and easy roadmap to analytics for email marketers. In this post, I’ll continue the discussion on analytics by taking a deeper look at predictive analytics, artificial intelligence (AI), and machine learning solutions in email marketing. [Moving forward, I’ll refer to the group of predictive analytics, artificial intelligence (AI), and machine learning email solutions as predictive features.]

Email Service Providers (ESPs) and vendors are offering exciting new predictive features. These features have the potential to send each subscriber the right email, at the right time, with the right content... but how can you know if these predictive features are a good fit for your email program? I’m here to help!

I’ll share four questions at the bottom of this post that will help you identify how predictive features could help you, but before we get there, let’s first a look at the analytics techniques these features use.

[If you are new to the world of analytics, I've compared the three most common types of analytics: Descriptive, Predictive, Prescriptive in my last post.]

When dynamically assigning content or selecting email send times, predictive analytics is to forecast an individual subscriber's preferences and actions, and prescriptive analytics maximizes an objective (e.g., clicks, revenue) by assigning the content or the send time to each subscriber. Within every ESPs predictive features, there is an underlying math model that makes decisions on your behalf. And like all things, these predictive features have both strengths and weaknesses.

Strengths:

  • process a vast amount of information in a short amount of time,
  • discover new insights from that information,
  • use the insights to prescribe a solution and
  • implement the solution on your behalf auto-magic-ally.

It is pretty neat when you think about it -- you can use the data within your ESP account to send more relevant content at the best time. Yay! However, this benefit hides major potential issues.

Weaknesses:

  • the quality of the solution is only as good as the quality of the data and the model(s)
    (this is extremely important)

To ensure ESP predictive features are more helpful than harmful, you must be able to ensure that you are using:

  1. the right data and
  2. the right model(s)

If the data that you use within these models is junk (inaccurate, not applicable) the results are also junk. Perfect data about the wrong audience is not helpful to you. The data used must be accurate and applicable.

If the model that you are using doesn’t reflect your business or its goals, then the results are junk. A model that optimizes open rate is useless to you if your focus is conversions. The model(s) used must reflect your business and its goals.

When using predictive features, you are outsourcing some decisions currently made by a person to a computer. And if you are going to outsource crucial decisions about your email campaigns, you must ensure that the computer “thinks” in the same or similar way as that person does (or at the very least that you understand how the computer “thinks” and are OK with the differences). Failing to do so, or not being able to do so, could result in the “magic” of predictive features to turn into a nightmare.

To find the right predictive features for your business, answer these four questions:

  1. What are your decisions? What can you change? What are your options?
    • Example: the image at the top of a campaign; the featured articles in an email, time to send the campaign
  2. What are your goals? Do you have one or multiple?
    • Example: sales; opens; clicks; conversions
  3. What are your rules? What are the relationships between decisions?
    • Example: there can only be one image at the top of an email; there are at most three featured articles; emails can only be sent during business hours US/Eastern
  4. What data do you have? What initial information can you use to initialize the model(s)?
    • Example: historical actions of subscribers (e.g., opens, clicks, page visits, purchases); append data

If you can’t answer these questions, then most likely predictive features aren’t a fit for you right now.

After you know the answers to all four questions above, ask your ESP how your answers align with their predictive features so that you can make an informed decision.

If you are not familiar with analytics, it is essential to only work with analytics companies or experts that can clearly explain the decisions, goals, rules, and data their models use in a way that you can understand.

Use these four questions whenever you are considering using predictive solutions in your email marketing -- and do not hesitate to speak with ESP and vendors about how their features work. When you find the right match, you’ll be ready to use analytics to improve your subscribers experience and increase your revenues!

Cover Photo by Carlos Muza on Unsplash

Related Posts