Marketing automation is a software platform that helps companies to highly personalized customer relationship and experience at scale, by automating marketing campaigns workflows to generate more leads, close more deals, and better measure marketing success through different communication channels:
Marketing automation tools are used to tackle the whole customer lifecycle by accompanying prospect and customers to enhance their journey and higher their ARPU at each stage. The most common marketing automation use cases are as follows:
Marketing automation leverage the high potential of the amount of data that companies possess, by using machine learning for segmentation (refining the campaigns targets), scoring (assessing customer attitude) and opportunities detection (revealing associations and hidden correlations), allowing to insure marketing campaign effectiveness, reach operational efficiency and grow revenue faster. Predictive and descriptive machine learning models help to identify customers and their needs, so as to increase their likelyhood to respond to a given campaign through specific channels:
Customer’s response to marketing communications can be different depending on many criteria like the sales channel, customer gender, location, activity, transactions and other relevant information. Segmentation is an effective tool that help to group customers with similar characteristics using historical data (their activity, purchasing habits and behavioral traits) and algorithms like Principal component analysis (PCA), K-means or Two-Step methods to find the clusters.below a step-by-step explanation of the PCA algorithm for segmentation:
The resulting clusters must be build with marketers, so as to map them to understandable profiles that represent distinct “buyer personas”. This will help to better personalize communication with company’s audience, depending on their propensity to respond to particular offers or promotions, by building a refined strategy and messages for each of those personas (or segment) to fits each stage of the customer’s journey.
Step 2: Scoring
Enrich customer information by augmenting them with new highly valuable information generated by machine learning algorithms that help marketers to maximize prospect conversion and customer’s ARPU, these scores are used in marketing campaign as condition to take the adapted action:
Lead scoring: allows categorizing leads, by differentiating between those who are really interested in the product from those who just starting to search some information. the higher is the chance that the specific customer is ready to convert. It can be calculated using 2 ways :
- Rules engine: by increasing and decreasing leading score based summing interaction’s weights, example: [+1 point] for website visit, [+5 points] click on Email contact, [+10 points] click on product catalog, [+20 points] download buyer guide, [+30 points] access payment form, [-10 points] after 1 month of inactivity, [-30 points] unsubscribe from the newsletter. Drawbacks: interaction’s weights are defined manually and need constant adjustments
- Predictive analytics especially regression, such as logistic regression that can be seen as the probability of conversion, it allows to :
- Get rid of choosing predictors manually, by using feature selection algorithms like stepwise backward to pick up the most relevant information about the leads from demographic information, online behavior and Email/social engagement.
- Get rid of defining weights, since it’s automatically defined by the regression algorithm during the model’s training.
RFM scoring [Recency, Frequency, Monetary]: it provide accurate definitions of the best customers, most loyal, biggest spenders, almost lost, lost customers and lost cheap customers.
- Recency score: Identify [purchase most recent date, purchase furthest date] interval and bin it to 3, 4 or 5 ranks. Customers who purchased more recently are more likely to purchase again than are customers who purchased further in the past.
- Frequency score: Identify [highest frequency of purchases, lowest frequency of purchases] interval and bin it to 3, 4 or 5 ranks. Customers who have made more purchases in the past are more likely to respond than are those who have made fewer purchases.
- Monetary score: Identify [highest monetary value, lowest monetary value] and bin it to 3, 4 or 5 ranks Customers who have spent more (in total for all purchases) in the past are more likely to respond than those who have spent less.
- RFM score = [Recency score] x 102 + [Frequency score] x 101 + [Monetary score] x 100
Next best offer (NBO): use association algorithms like Apriori and CARMA that are trained on historical data (customer spending habits) to propose to each customer new products and upgrades that best fits his needs, which help companies to adopt a customer-centric approach, increase conversions and encourage sales.
Churn score (attrition rate): predict customers who have a high likelihood to cancel a subscription to a service.
Step 3: A/B testing
A/B testing is used for Marketing campaign to assess its different variants, so as to determine the most effective one (send the right message at the right time) that generate the highest number of customer response (best click view, click-through and open rates). A/B testing provide good insights around what wording, visuals and sending time will work best for each specific segment. The tested points are as following:
• Email subject line which is seen first by the audience first and can affect the open rate.
• Subject Design & content like using different text, various layout, visuals, video vs 2D images, which affect your clicks & conversions.
• Sending Time & frequency: find a happy balance for sending frequency and time take into account customer’s behavior/feedback, the type of your business and the strategy adapted by industry peers.
To execute A/B testing, we must split the audience (the whole target or preferably only new prospects and customers) to as much segment as campaign variants that must be tested, and then it uses prospect/customer’s feedback history and 2 kind of tools to assess the significant difference between the tested cases:
- Statistical (frequentist) methods: 2 types of statistical tools are used:
- Machine learning methods: like Bayesian A/B testing, where each case is treated as random variable to which we assign a prior probability inferred from past knowledge of similar experiments, we combine this prior probability with current experiment data to find the right answer to the test.
Posterior = Data + Prior
Usually, we test one thing at a time, to get accurate results, and tests are differentiated depending on the location, gender and segment. Also, continual testing and optimizing is essential after campaigns go live.
Step 4: building Marketing campaign Workflow
To automate business scenarios, Marketers build campaign workflows, a sequence of steps and tasks with a predefined check points of events (rules and conditions) that trigger touch points (email, SMS, …), that go along with the customer to guide him and light up his choices throughout a predefined path. 4 main elements are used in the workflow:
You can see below, 2 use cases of campaign workflow:
- Cross-sell email campaign workflow:
- Recover abandoned shopping carts workflow:
Step 5: Measure Campaign Performance
At this stage, the goal is to monitor campaign results in real time and track its effectiveness, by detecting opened Email (Views, reads, location), click made from emails, website traffic generation (number of visits of landing pages) and sales through website/outlet (Cost Per Lead, Lead to close ratio, Conversion Rate, Return On Investment ROI).
For campaign In Social media, the following metrics can be assessed: new followers, comments, likes, retweets, channel views, bounces and subscribers.
Performance of a campaign is also assessed by comparing those KPI for the target vs the control group, to see if there was a real impact that leads to substantially exceed the organic sales growth.
At this stage, Machine learning can be used (Regression or Neural network) to predict those KPI before campaign execution, based on historical data of the former campaigns, including customers’ feedback and behavior, so as to avoid bad campaign parameters that lead to weak performance.
Step6: refine your model
Applying machine learning on marketing system allow to keep campaign performance at the Top, by proceeding periodically and automatically to a tuning of the used models, based on the new data collected from the ongoing campaigns, which help to improve their accuracy and power to provide the right prediction.
This is where machine learning gets really interesting, as you an end up with a system which changes and improves itself over time.
Marketing automation is a must have tool for marketers that get enhanced by leveraging machine learning capabilities, which provide the capacity to process data at scale, to get the right target and the right action at the right time, helping to optimize User eXperience, increase customer loyalty and commitment to the brand, and raise the ROI.