1 upvote 0 discussions

MarTechExec takeaway

“Marketers need to look at predictive analytics as more than just a single step.  If you want it to work, it can’t be disconnected from everything else. Your input data, your model and your output data are one and the same.” – Lana K. Moore, Executive Editor at MarTechExec (@martechexec)

Definition

Predictive analytics is the process of using software to predict likely business outcomes. Once the software user asks a specific question, predictive analytics sorts through past data to come up with an answer. This is done using techniques like data modeling, machine learning and AI.

Guiding principles

Predictive analytics deserves your full attention — if you hope to do it well, anyway. But we get that your plate is full and you may need to come back to consume our advice later on.

In the meantime, here’s our short and sweet version of how to succeed with predictive analytics.

Use a quality foundation

Predictive analytics is only as smart as the data it analyzes. Give it a solid foundation to work from by cleaning and organizing your stored data. For best results, feed in online and offline data by integrating your predictive analytics solution with your existing databases.

Don’t bite off more than you can chew

Great data means nothing if you can’t process it right. Research different predictive models and what they’re best suited for. Pick one that can not only help you reach your objectives but that can also work using the data you have. The most complex model isn’t always the best model.

Tweak where needed

History doesn’t always repeat itself. Customers change their behavior all the time, and looking at predictive results alone can be misleading. Consistently monitor for outliers and be wary of correlations that aren’t what they seem. Then, adapt your model the next time around.

Predictive analytics, demystified

Data is something I research and write about almost daily. And while I pride myself on the strength of my left-brain, it’s really my right-brain that runs the show.

That’s why I tend to use creative analogies to help wrap my brain around complex data topics — like predictive analytics.

Is Martha Stewart in the house?

So here it is: Predictive analytics is like baking a cake.

No, it’s not a piece of cake, as in it’s super easy to do. But the pre-, mid- and post-predictive analytics processes are scarily similar to the cake-baking rituals of Martha Stewart and Rachael Ray (Trust me. I’ll explain).

Here’s the thing, though: We can’t just talk about predictive analytics, because understanding why and how to use it means understanding how it fits into the martech ecosystem.

In my opinion, the simplest way to understand this is by visualizing it. And I think this infographic from McKinsey does that well.

Let’s apply this five-step process to my epic baking analogy to see how predictive analytics plays a central role in data science, the process of gathering insights from data.

Step 1: Gathering ingredients

Before anything else, data science needs a foundation made of — you guessed it — data.

Step 1 includes collecting the key ingredients for your cake — or, the data you need in order to glean great insights.

This includes:

  • Website and social media data
  • Purchase and inventory data
  • Customer contact information
  • Customer acquisition and retention

...to name just a few.

This data can be both structured and unstructured, meaning it can be nice and tidy or a total mess. Step 1 is just about gathering your ingredients, so you need not worry about its presentation just yet.

Step 2: Stirring ingredients

This is when you mix all your data into one scrumptious concoction, like in a database.

Like a good cake batter, a good database has a smooth consistency. In marketing language, this translates to using things like data aggregation and data mining to scrub-a-dub-dub your data until it’s pristine enough to use.

It’s this data that can then be used for descriptive analytics — the “preliminary stage of data processing that creates a summary of historical data.”

Descriptive analytics that specifically looks at business performance, such as sales, costs and customer churn, is known as business intelligence.

Step 3: Baking the cake

This is is when your data goes in the “oven” that is predictive analytics.

If your foundational data is no good, your “cake” will fall flat. And if your model “bakes” the data in the wrong way, all your data efforts are for naught.

In other words, you need to have great data to use with your model… and a great model to use with your data.

If you’ve done step 2 right, the “great data” aspect should be taken care of already. Now, you just need to choose a data model to use with it.

Some of the most common data models include:

  • Cluster models, which create audience segments based on variables you set.
  • Propensity models, which make predictions about customer behavior, like predictive lifetime value, propensity to buy and propensity to churn.
  • Collaborative filtering, which is used to deduce what products or services a customer will be interested in so you can make accurate recommendations.

All of these models use a tactic known as regression analysis, which is how we learn what variables are related to one another.

A regression analysis could find, for example, that customers who refer a friend to your business are more likely to convert on upsell opportunities than non-referrers.

If your business goal was lead generation, you could then find your most frequently “upsold” customers and target them with a referral campaign. Or, if your business goal was to increase purchase amount, you could find your top referrers and offer coupons to encourage upsells.

Step 4: Refining the process

During step 4, you’ll use the insights you gathered in step 3 to refine your process for next time.

This is also when prescriptive analytics comes into play, since your insights guide you to “prescribe” a solution to past problems.

Why marketers love predictive analytics

Predictive analytics is a pivotal part of your data-baking process. Do it well, and you reap the rewards of the tedious steps that came before it. But do it poorly, and you could risk the quality of all your efforts thereafter.

We reached out to some predictive analytics advocates to learn why they tout this tactic — and why they say others should, too.

Money-making mixes

It goes without saying that utilizing big data is no easy feat. According to eMarketer, 54 percent of U.S. executives are currently trying to make money from their big data.

Predictive analytics can hurry this process along by analyzing your big data in a completely automated way.

“Very simply put, the more data we have the deeper and more precise the insights and predictive and prescriptive analytics can become. Those that understand the importance of data in the next few years will look to monetize it and will create a competitive advantage.  Those that don’t, will fall behind.” — Korey Thurber, Head of Analytics at Harte Hanks

Much butter churn rate

Studies show that lowering your churn rate by just 5 percent can make your business 25 to 125 percent more profitable.

When we have an idea of what factors cause our customers to churn, we can add them to our predictive analytics model to find out which of them we’re most likely to lose — like this infographic from Presidion shows.

This is a saving grace that gives us more time to reach out to high-risk churners — before they leave us for good.

“Soon, companies will be able to take all the data they are collecting to accurately predict churn, understand if opportunities are on track and to figure out who else would be a great customer.” — Chris Rothstein, Chief Executive Officer at Groove.co

Smart as a cookie cake

Predictive analytics helps us analyze advanced buying signals, which we can factor into our predictive models to make smarter predictions.

This also opens the door to upsell and cross-sell opportunities, since predictive analytics gives us insight into what product skews a customer might like.

According to eMarketer, 35 percent of U.S. retailers want to utilize predictive analytics to make personalized product recommendations.

"For the first time, the possibility exists to create 1-to-1 targeted marketing at scale. This is good for businesses, but it’s good for customers as well." — Bradley DeLoatche, Co-Founder and Chief Technology Officer at Snappy Kraken

Why marketers hate predictive analytics

As sweet as predictive analytics may seem, it’s not always so good for us. Or our wallets. Or our customers.

Predictive analytics is far from perfect, and marketers aren’t hesitating to express their frustrations.

These are the most common issues expressed by skeptics of predictive analytics.

The flavor of false

According to Teradata, the main issue with false positives and false negatives is that they waste time. To illustrate why, here are examples of each.

  • False positive: Your predictive model tells you that John Doe — or a segment of similar John Does — would be an ideal target customer for your new product. But after launching a series of targeted ads, you find this to be totally untrue. No one from that audience segment even clicked through.
  • False negative: Your predictive model tells you that Jane Doe is at a high risk of churning, but that her sister, Jenny Doe, is at a low risk of churning. But joke’s on you: They both churn. In this case, Jenny was a false negative. And while you may have accurately predicted one, you basically lost a customer that you could’ve easily retained had you identified her as part of your churn reduction campaign.

Avoiding false positives and false negatives has a lot to do with how your predictive models work. And if predictive analytics vendors can’t improve on their flawed models, we as marketers are basically stuck on a sinking ship.

“Predictive analytics is heavily reliant on the data it can ingest to produce insights. Often analytics companies are sitting on a massive amount of data that is either unclassified or cannot be used for modeling.” — Varun Gudiseva, Vice President of Market Development, Analytics at Tapad

Everything’s not as it creams

By the same note, marketers can fall victim to “spurious correlations,” meaning the predictive model shows that two variables are related when really… they’re not.

This chart comparing ice cream sales to shark attacks is a great example. If we took the data at face value, we would conclude that ice cream causes shark attacks. Realistically, the relationship can probably be explained by another factor: Time of year.

More people are in the water during the summertime, which equals a higher chance of shark attacks. And with the summer heat comes refreshing treats, which mean ice cream sales will skyrocket.

There are entire websites dedicated to pointing fun at correlations like these. And ridiculous as they may seem, these faulty correlations happen. All. The. Time. And it’s up to us to exercise common sense and read between — or beyond — the lines.

"Marketers know the data is laden with insight to customer traits and behaviors that can help target ideal prospects, make recommendations and predict likely outcomes, but its size, diversity and and velocity make it unruly to work with." — Mark Gamble, Senior Director Product Marketing, Analytics at OpenText

Please eggscuse the dirty data

We know we’re beating a bad egg here, but it’s worth repeating: Quality data is the heart of predictive analytics. But the sad truth is marketers don’t always have the best data to pull from. And they know it.

According to eMarketer, only 32 percent of U.S. marketers believe their own first-party data is “very accurate.” Another 51 percent said it’s “somewhat accurate,” and 16 percent it’s neither accurate or inaccurate… whatever that means.

“The most important element is that the data that goes into anything predictive or AI-driven is accurate, in-depth and up to date. That’s hard to do, and companies that get it right will win.” — J.J. Kardwell, Chief Executive Officer at Everstring

“While the digital landscape has launched marketers into the “Big Data” age, the availability and granularity of offline data is still in its infancy.” —Donald Gallant, Vice President of Analytics and Innovation at Marketsmith, Inc.

Can you past the butter?

By nature, predictive analytics only looks at data from the past.

But trying to determine what will happen based on what already happened is like predicting the success of your cake-bake based on how much flour you used.

While “wrong amount of flour = unsuccessful cake” isn’t a false assumption, this thinking limits you to only considering a strict set of variables. In the end, fewer variables in the equation means a less precise prediction.

“In spite of all the technology that tracks our lives, predictive analytics are often only as good as the last month. In other words, the 30-day cookie is holding the market back.” — Mike Gilloon, Director of Strategic Communication at Bozell

“The history of how that data changes over time is actually what’s most interesting for predictive models. When we can see the complete history of how and when that data has changed we can start to analyze the actions and milestones that correlate to data changes and that will take us to the holy grail - causality.” — Manny Medina, Chief Executive Officer at Outreach

Tools of the trade

As the Periodic Table of Predictive Analytics shows, solution providers aren’t few or far between.

How do you choose the right tool to help your predictive analytics succeed? Here are a few factors to keep in mind.

Always integrATE

The more data your models can access, the more accurate their predictions will be. Look for a solution that integrates with your existing data platforms, like your CRM and your DMP.

Not only does integration make it easier to “feed” your predictive models, but it also ensures your plethora of stored data doesn’t go to waste.

Boxed isn’t better

Some predictive analytics solutions use “black box” models in their predictive processes. In short, these models don’t clearly show how they calculate their results. And this leads to a lot less transparency from the marketer’s perspective.

Before choosing a predictive analytics solution, take a long, hard look at the models they use and how much control you would have to tweak them.

Recipe for success

Your best predictive analytics solution is the one that helps you reach your objectives. And like we’ve gone over already, predictive analytics is only one slice of the process.

Prescriptive functionality is huge when you’re evaluating different solution providers, so dig deep and look for those that offer it.

Future of predictive analytics

In the spirit of predictions, we asked some martech experts what they expect for the future of predictive analytics.

Spreading the sweets

Predictions-as-a-Service is on the horizon, folks. And marketers should be excited. Because, you know, we could use all the help we can get.

Insights Success explains how Predictions-as-a-Service will lead to “greatly expanded and enhanced predictive analytics, as a company’s internal data is compared against an aggregated mass of data.”

This will also make predictive analytics way more accessible to businesses, since they’ll be able to outsource all of it to a dedicated service.

“Where things get really exciting is how the space has been moving away from services engagements to fully productized and packaged solutions that empower teams to apply predictive analytics to their respective work.” — Leah Pope, Chief Marketing Officer at Datorama

“We’ll see more software solutions that provide context to data and better predictive analytics that can actually help organizations grow their businesses – in addition to better determining tactical output or measuring how a campaign has performed as this technology has done to date.” — Kate Richling, Chief Marketing Officer and Co-Founder at Birdsnest

“Predictive analytics will be used in more packaged applications and be more-user friendly. As Gartner has noted, predictive models will become more embedded in business applications which we see manifesting in how our clients are thinking.” — Kerrie Wuenschel, Director of Analytics at R2integrated

Catering to customers

Predictive analytics casts light on customer traits we may not have seen otherwise. In particular, predictive insights can help us add another level of personalization to the customer experience.

“By studying the impact of prior site openings in similar geographic areas, marketers are better equipped to understand and anticipate how consumers will react and can adjust their marketing tactics accordingly in order to preserve market share and loyalty.” — David Bairstow, Vice President of Product at Skyhook

Have your cake and eat it too

One of the main complaints with predictive analytics is that it’s only a means to an end, meaning that it only tells you what could happen — not how to make it happen (or not happen, if it’s something negative).

But in the next few years, the line between fact-based predictive analytics and action-based prescriptive analytics will start to blur.

“Predictive analytics and applied artificial intelligence promises to make marketing and sales better, smarter, faster and, most importantly, more personalized. The technology is moving from easing execution to now supporting go-to-market decision making.” — J.J. Kardwell, Chief Executive Officer at Everstring

“For the most part, predictive analytics has failed to deliver more than small insights into current behavior, as predictions often don’t have a way of translating to real action in the field. The emergence of machine learning and AI technologies allows these predictions to create actions that will change the way marketers think about the content they are about to create.” — David Keane, Chief Executive Officer at Bigtincan

Trends

Mind you, we’re not just pulling these ideas out of thin air. The predictive analytics landscape is looking more lush every day, and current trends indicate we have a lot to look forward to.

Here are a few big ones to note.

Variety is the price of life

Dynamic pricing is one predictive analytics tactic that has both marketers and customers on edge. On the customer front, Wiser Retail Strategies explains that “Consumers can be scared off by frequent price changes.”

And on the marketer front, the fear of a “margin depleting price war” makes many wary of using dynamic pricing at all.

If and when dynamic pricing makes cost a less important factor, focusing on customer experience will be key to differentiating yourself from competitors.

Real-time rewards

IoT connectivity will include way more variables which predictive models can use for real-time solutions. Look around and you’ll realize this is already happening — like with Amazon Dash and Alexa-powered Echo.

IBM estimates that there are 1 trillion connected objects generating data today. It should go without saying that the more data we have on hand, the faster we can make decisions and the more accurate they’ll be.

“It now becomes possible to mine every photo a consumer uploads to a public social media profile. It will be possible to classify these photos using deep learning, and include that information in the predictive analytics pipeline.” — Michel Ballings, Assistant Professor at Department of Business Analytics and Statistics, The University of Tennessee, Ballings.co

Smarter taste tests

Adaptive content is the new kid on the block. Granted, it’s a very cool kid. Instead of performing stagnant A/B tests that yield limited results, adaptive content opens the door to an abundance of content options that fit an array of customer tastes.

This is the latest iteration in a series of related concepts, such as responsive content and mobile-friendly content. But adaptive content is different in that it caters to each channel, thus making the C.O.P.E. method (Create Once Publish Everywhere) easy to execute.

Marketing Insider Group talked about how NPR used adaptive content and saw an 81 percent ROI on the efforts they made to put it in place.

The insights created by predictive analytics will key you in on what each user needs to see in order to convert — plus when and where they should see it.

"For the B2B predictive marketing technology, the ability to target precisely (no false positives) and completely (no false negative) the potential buyers of your product will become more and more available and more and more used. This means a more efficient marketing and sales process." — François Bancilhon, Vice President of Public Affairs and Innovative Projects at Sidetrade

“In the next 3-5 years, you will be hard-pressed to find a company that has not already embarked on a predictive directive. The differentiator will be the current focus and data we are collecting now.” — Scott Teger, Platform Operations and Analytics at Mundo Media, Ltd. 

Conclusion

Predictive analytics is one big piece of an even bigger data science dessert.

For first-timers, the thought of using statistics and scatter plots to solve real-world problems can be a little scary. It’s part “go with your gut” and part “trust the model,” which leaves many a marketer unsure what results they can trust.

But as we can pretty much guarantee that legends like Martha Stewart weren't born baking, you can rest assured knowing you won’t be a predictive pro right off the bat.

Follow our advice and remember the experts’ tips to make sure your predictive analytics insights are as sweet as they can be.