It’s been estimated that the average American sees about 4,000 ads per day.
As a marketer, let that sink in for a second. Amazing but daunting time to be alive, isn’t it?
Your work is cut out for you: ensure your marketing efforts are reaching the right customers at the right time. And as you likely well know, there are at least two fundamental questions to answer:
- Who are my best, most valuable customers?
- How do I reach and acquire more of them?
The practice of predictive analytics can yield sights on these questions, but like any advanced, powerful tool, it needs to be properly understood and applied to be effective.
You’ve probably read a few blog posts, watched a few webinars and maybe even suffered through a lengthy whitepaper or two, but may still be thinking:
“Wow, predictive analytics seems so useful but how can I use it to solve my business problems?”
“Predictive analytics is too complex but my team doesn’t have the resources or time to learn and use it”
If you’re having thoughts like this, you’re in the right place. The goal of this post is three-fold:
- Explain the basics of predictive analytics
- Provide an overview of applications for improving marketing ROI
- Share a recipe for success
But first, let’s quickly cover some definitions to clear up any confusion
Defining Predictive Analytics
SAS states it quite nicely as:, “Predictive analytics is the use of data, statistical algorithms and machine learning techniques to identify the likelihood of future outcomes based on historical data. The goal is to go beyond knowing what has happened to providing a best assessment of what will happen in the future.”
In marketing-layman terms, a predictive analytics – or a predictive analysis – can provide insights by predicting unknown values related to customer behavior that lead to reducing cost and/or increasing profit margins. Thus, a predictive model is a formula for estimating a specific unknown value of interest, called the target.The target could be a prediction of things like:
- Expected KPI for a future week or month based on historical performance
- Incremental revenue generated from untested website UX changes or SEO efforts
Let’s get even more specific with a couple examples, shall we?
Applications in Marketing
Here are two applications of predictive analytics for marketing:
- Forecasting KPI targets and attributing revenue impact of untested website UX changes
- Identifying anomalies and potential issues in website checkout funnel performance and paid search campaigns
- Problem: Marketing management/customer acquisition team was having difficulty setting and meeting their goals for new revenue generation. They were consistently missing weekly and monthly KPI targets, management was unable to identify root causes, create accountability or most importantly, answer questions like “Did our website redesign 2 months ago drive the increase in number of micro and macro conversions we’re seeing?”
- Solution: Predict future KPI’s based on segmented, historical data and implement predicted targets into existing reporting workflows. Predict impact on KPI’s (e.g. – sales) by comparing a) prediction or estimate of sales if no website changes were made and b) actual KPI performance after website changes are made
- Results: KPI target hit rate rose to 90% by using forecasted KPI targets. ~12% increase in quarterly organic and referral revenue by properly attributing short-term improvements in KPIs to website redesign efforts and reallocating resources towards website Conversion Rate Optimization in the long-term.
- Problems: It’s unclear when our website checkout conversion funnel is declining or increasing due to a) technical issues or b) random noise. It can also be difficult to identify when campaign performance is declining from a) our messaging and targeting or b) our old friend, random noise
- Solution: Create anomaly detection model, dashboard and alert system to identify issues with website performance and paid search campaigns
- Results: ~8% increase in marketing ROI from reducing ad spend during identified problem periods with website checkout funnel and reallocating spend from poorly performing campaigns to higher performing ones
Recipe for Success
A successful predictive analytics project is like baking a cake: you get out of it what you bake into it. Delivering value is most common when you have the following ingredients:
- A crystal-clear business objective and question to answer
- Clean, quality data
- Experienced data analyst or data scientist
- And…YOU the experienced marketer and project manager
While predictive analytics isn’t for the faint of heart with arithmetic, if you properly understand your business objectives and questions then you can provide value by managing a predictive analytics initiative since a model is only as strong as the assumptions it makes. In other words, a racecar is only as fast as the driver behind the wheel – who doesn’t also need to be the mechanic!
Wait..what is a clean and quality dataset?
A dataset is clean when we have consistent labelling and values throughout our dimensions. For example, we should have:
- Consistent acquisition data (e.g. – source/medium labels and/or channel groupings)
- Consistently populated and human readable custom dimension and custom event labels
- Consolidated url’s
- Documentation of time periods of incorrect data (e.g. – double or under counting hits due to tracking errors)
A quality dataset means we have:
- The proper level of granularity (hit and customer level dimensions)
- Meaningful metadata (e.g. – customer subscription status or length of time as a subscriber)
- Required volume for modelling (e.g. – 2 years of historical website sales to use in a forecasting model)
For example, if we were using the Google Analytics API to extract data for a customer-level predictive model (e.g. – predicting a customer’s probability of conversion based on view history), we would require at least the following custom dimensions implementedl:
- The client ID set by Google Analytics (session-scope)
- A session ID unique, randomized value (session scope)
- A user ID set when someone logs into your website (user and/or hit scoped)
- A hit timestamp or the actual timestamp of each hit in local time, with the timezone offset included (hit-scoped)
An alternative to a clean, quality dataset extracted from the Google Analytics API, are the Google Analytics 360 and Firebase data exports to BigQuery, an enterprise ready data warehouse solution. Google also offers access to several other valuable, hit-level datasets via BigQuery Data Transfer Services:
- Google AdWords (beta)
- DoubleClick Campaign Manager
- DoubleClick for Publishers
- YouTube – Channel Reports (beta)
- YouTube – Content Owner Reports (beta)
Just like your favorite Italian restaurant and their traditional, marinara sauce, we at E-Nor have our own recipe for success or methodology for predictive modeling:
- Define – the business objectives and questions to answer and explore the source data
- Prepare – the data by extracting, transforming and loading it to the location and format for modeling
- Understand – Create, test and validate a predictive model to answer the business question
- Communicate – the analysis results in the form of data visualizations and written analysis
- Integrate – or activate the results into Google Analytics for use in remarketing campaigns (e.g. – predict a customer’s likelihood to churn then load that value into a custom dimension slot in Google Analytics), Email Service Provider (ESP), BigQuery or your own data warehouse for further analysis
Hopefully you now have a better grasp on what predictive analytics is and some concrete use cases for improving your marketing ROI and bottom line.
Predictive analytics can yield valuable insights, but like any advanced, powerful tool, it needs to be properly understood and applied to be effective.
At E-nor, we have the expertise and can help with the entire process. From defining the business objectives, the nuances of collecting and storing clean, quality data and of course, selecting and using the right predictive models that deliver value to your business.
Contact us to learn more about how we can help with your data collection, Predictive Modeling and more!
About the Author
Machine Learning Engineer
Justin is passionate about numbers and solving problems. He loves learning new tools and ways to tackle old and new challenges alike. Before joining E-Nor, Justin spent 6 years as a consultant, including running his own firms, helping companies improve their websites, mobile apps and digital marketing through data engineering, visualization and analysis.