Predictive analytics in marketing: everything you need to know

Before you start sweating: nope, your competitors aren’t very likely to be advanced users of predictive analytics in marketing. Yet.

But since the predictive analytics market is growing at a 23.2% rate year over year and we know that predictive analytics can help you make better marketing decisions (which equals more revenue), we’re tempted to say that it’s only a matter of time until your rivals catch on.

So to kick off your education in predictive analytics (and to ultimately show your competitors who’s who in advanced marketing data wizardry), grab a cup of coffee and make yourself comfortable, because you’re about to hit the ground running with predictive marketing analytics. ??‍♂️

And if you’re only here for one thing or otherwise impatient (I feel you), help yourself to a specific section of this post:

• What is predictive analytics?
• What is predictive analytics in marketing?
• 5 examples of predictive analytics in marketing
• The 7-step predictive marketing analytics process
• A 4-point checklist for getting started with predictive marketing analytics

Ready? Let’s go!

What is predictive analytics?

Predictive analytics is the process of using current and/or historical data with a combination of statistical techniques — including (but not limited to) data mining, predictive modeling, and machine learning — to assess the likelihood of a certain event happening in the future.

To drive home what predictive analytics really means, let’s compare it to a few other branches of business analytics:

• Descriptive analytics answers definitive questions like “what has happened?”
• Predictive analytics answers hypothetical questions like “what is likely to happen?”
• Prescriptive analytics takes predictive analytics one step further by answering questions like “what should we do based on what has already happened and what is likely to happen?”

In business, one of the earliest and most intuitive applications of predictive analytics is credit scoring. By using historical information about a person’s loan applications, past payments, and credit history, banks and other financial institutions use predictive analytics to calculate a score that reflects the likelihood of that person making their payments on time in the future.

In case you’re still awake, a slightly sexier and more recent example comes from Netflix. They pulled historical data about the success of previous tv shows to create House of Cards, the award winning series that follows a carefully constructed recipe: 

• David Fincher as the director
• House of Cards as a concept (tested in the UK)
• Kevin Spacey as the leading actor

But what about marketing then? How can marketers benefit from predictive analytics?

What is predictive analytics in marketing?

In the marketing context, predictive analytics refers to the use of current and/or historical data with statistical techniques (like data mining, predictive modeling, and machine learning) to assess the likelihood of a certain future event. Duh.

But to understand what this actually means, let’s look at a couple of practical examples.

5 examples of predictive analytics in marketing

1. Customer and audience segmentation (using cluster modeling)

If you don’t know whether you should segment your audience based on their behavior, demographics, firmographics, interests, or any other variable, predictive analytics can help.

By experimenting with different cluster models, you’ll be able to find patterns that you may not have expected, and that way arrive at audience segments that make the most sense for your business.

2. New customer acquisition (using identification modeling)

Taking your segmentation one step further, you can use your customer data to create identification models. In practice, this comes down to identifying and targeting prospects that resemble your existing customers in some meaningful way.

A common example of this is Facebook’s lookalike audiences. You can use this feature to upload a list of the emails of your best customers, based on which Facebook starts targeting your ads to people similar to these customers. 

3. Lead scoring (using propensity modeling and predictive scoring)

Already in 2015, a Forrester study identified predictive lead scoring as one of the top three  use cases of predictive marketing analytics. In practice, the process comes down to using past customer data to rank identified prospects according to their likelihood to convert. 

Depending on your business model, you can use this data to trigger relevant marketing messages and/or prioritize your sales team’s outreach efforts when a prospect reaches a certain threshold in your lead scoring model .

4. Content & ad recommendations (using collaborative filtering)

While pretty much all successful ecommerce businesses (think Amazon and Zalando) and streaming services (think Netflix and Spotify) are experts in using collaborative filtering to come up with relevant product/series/song recommendations, most marketers have yet to embrace similar tactics.

In practice, collaborative filtering comes down to using past behavior (e.g. aggregate-level content consumption patterns within a particular segment) to make recommendations for content consumption, cross-sell, or upsell. 

For example, let’s say that you found out that most of your new customers in the retail industry started a trial immediately after reading a particular case study of a Fortune 500 retail business. Based on this behavioral data from a specific segment, you might well want to introduce this particular case study to your retail prospects at an earlier stage to see if you can shorten the sales cycle.

5. Personalizing customer experiences (using automated segmentation)

For the longest time, personalization was synonymous with “Hey {firstName}” emails. But here’s the good news: predictive analytics can help you go far beyond that.

Going back to your meaningful audience segments, lead scoring (aka recognizing an individual prospect’s propensity to buy), and triggered content recommendations, you’re able to increase not only the relevance of your marketing activities but also their return on investment.

The 7-step predictive marketing analytics process

Now that you hopefully have an idea of what you can achieve with predictive analytics, it’s time to look at what the process of getting that done might look like in practice. 

And worry not, I’ll save you from machine learning algorithms and the like. Instead, you’ll get a simplified example just to give you a fairly straightforward glimpse into the necessary steps.

1. Define the question you want to answer

Before you jump head first into data, you’ll want to have a clear idea of what you’re doing. As mentioned earlier, the question you’ll define here should be of the “what is likely to happen based on what’s happened before?” variety.

Good examples include:

• “Which MQLs are likely to buy within the next 30 days?” (based on what’s happened before)
• “Which pieces of content should I serve to people whose trials have expired if I want them to convert?” (based on what’s happened before)
• “Which audience segment should I target in my next Facebook campaign?” (based on what’s happened before)

2. Collect the data you need to answer your question

Let’s say you settled on the first example question: “Which MQLs are likely to buy within the next 30 days?”

For context, let’s say that your company is trying to close a big funding round, and that’s why you’re in a rush to get some new business in so that you’ll get a better deal from the investors. So here you are, trying to find the lowest hanging fruit that you can close within 30 days. 

And to answer the question of which of these MQLs are most likely to convert within the next 30 days, you’ll need at least:

• Historical data on your MQLs
  • The average and median of the number of days it took from MQLs to become customers
    • Split by channel (e.g. Facebook, Twitter, organic traffic to blog etc.) and individual touchpoints (e.g. ads, blog posts, and website pages etc.)
  • Firmographic information per MQL (e.g. industry, company size)
  • Demographic information per MQL (e.g. title)
• A list of your current MQLs that haven’t bought yet

3. Analyze the data you’ve collected (aka do some good old-fashioned descriptive analytics)

Now that you have all the data you need for analysis, it’s time to start crunching. In my example above, I would try to list and find answers to questions like: 

• Does the average number of days to convert vary between different channels?
• Does the average number of days vary from ad creative to ad creative?
• Do firmographic variables like company size or industry correlate with the number of days to convert?
• …

The list here is practically endless but you get the point. 

4. Build and test your hypotheses with statistical techniques

Once you’re happy with your list of questions and you’ve gotten into number crunching mode, it’s time to test your hypotheses.

Just because you might logically reason that it takes larger companies a longer time to make a purchase than it takes smaller companies, that’s not necessarily true. Test out all your hypotheses and go with your data, not your gut.

5. Create a predictive model

Once your hypotheses have been tested and either validated or thrown out the window based on your data, it’s time to create a predictive model. Simply put, this comes down to using statistics (and often machine learning) to predict outcomes.

By now, you’ll probably need an engineer or a data analyst who knows Python or R. 

6. Deploy the model

Now that you have an existing predictive model (yay!) it’s time to start putting the results of your model into practice. Remember that predictive analytics won’t make any decisions for you. It’s up to you to look at the data and turn it into actionable insights.

In this case, our predictive model would spit up the MQLs that are the most likely to convert into customers within the next 30 days. Based on this information, we could then direct our marketing and/or sales efforts to those prospects in an attempt to convert them by the deadline.

7. Monitor, iterate, and create new models

Finally, remember that external variables (COVID-19, anyone?) can throw your model off completely.

But because some of the external variables (think seasonal fluctuations and trends in customer behavior) aren’t associated with something as obvious as a global pandemic, it’s a good idea to adjust and/or replace your models with new ones every now and again.

Now, take a short break and congratulate yourself, because you’ve breezed through all the seven steps in the predictive marketing analytics process. ?

 All you have to do now is to get started for real.

How to get started with predictive marketing analytics?

Now that you know what predictive marketing analytics is, what you can do with it, and how the process should work, I hope you’re excited about getting your first few models up and running.

Here’s what you should do next.

1. Get buy-in and secure resources

Here at Supermetrics, we’ve never met a cowboy who’s managed to successfully use predictive analytics in marketing on their own. The simple reason is that marketers are not often engineers, and engineers are not often marketers.

That’s why before you do anything else, you need to pitch your idea to the C-suite to get budget, a dedicated team, and some technology. However, if you’re lucky (and working at a large enough company), chances are that some departments other than marketing are already working on predictive analytics.

And if that’s the case, as soon as you’ve gotten the green light from management, you can pretty much go with the knowledge and the technology that you already have in-house. 

But if you’re not so lucky, this might be the time for some serious discussions about your marketing data stack.

2. Choose technologies based on your needs

What kind of technology do you need then? 

Centralizing all your marketing data in a data warehouse is a good start. Bonus points if you can also store sales and other business data in the same place. This way, your data team can quickly pull out cleaned and mapped out data for your predictive models since they don’t have to waste their valuable time on data wrangling.

Long story short: you’ll want to work with your data team to figure out what the business and technical requirements are, and work out the best solution based on those requirements.

Psst! If you decide to go with a data warehouse, check out this post where we compare BigQuery, Snowflake, and Redshift.

3. Encourage and educate the whole staff

Since predictive analytics is (still) no job for cowboys and since other departments can also learn from your predictive marketing analytics forays, you’ll want to kick off this massive undertaking with transparent communications.

Share your successes, share your failures, and share your learnings. This way, other people in the organization won’t have to repeat the same mistakes you’ve made, and the company will save a ton of money in the process.

4. Learn and adjust

Think of predictive marketing analytics as a marathon, not a sprint. 

Sure, your first couple of attempts at predictive modeling may be a bit on the confusing side, and you might not get a lot of valuable information out of them. But as you learn what you can and can’t do with predictive marketing analytics, your models will improve, and so should your results.

So keep going and never stop learning.

Over to you, you (future) predictive marketing analytics genius! ?

To recap, commercially viable applications of predictive marketing analytics are increasing fast, and you can freely decide whether you want to be among the first marketers to jump on board (or whether you’d rather try to catch up with the fast moving ship later on).

So what’s it going to be?

Psst! If you’ve done something cool with predictive marketing analytics already, I’d love to feature you and your marketing team in a future post. Any leads will be warmly welcome at pinja@supermetrics.com.

And if the idea of moving your marketing data into a data warehouse excites you, why not check out this webinar, where Sebastian Mehldau explains how VanMoof (a really cool electric bike company from the Netherlands) has been able to streamline their marketing analytics with BigQuery and Supermetrics. ?