Have you ever wondered how you can predict how likely a customer is to buy something before they even make a purchase?
Today, with the help of machine learning, this is possible. Predictive modeling allows businesses to use past behavior data to understand how likely current or future customers are to buy a product or service. Predictive analytics—already a $10 billion industry—can help CIOs and their teams to better understand what customers are most likely to buy and how best to target them. From purchase history to web browsing data, there are a number of different factors that can be used to build a propensity model, and generate insights that can boost revenue.
Given these diverse use-cases, it's no wonder that predictive analytics is forecasted to become a $28 billion industry by 2026.
In this article, we will explore how predictive analytics can be used to improve everything from customer acquisition to churn reduction, focusing on the benefits of using propensity modeling to improve business outcomes.
Perhaps no word is more commonly bandied about in business than “AI.” But what does AI actually mean, and how can businesses use it to improve their operations?
In short, AI is a process of programming computers to do things that ordinarily require human intelligence, such as understanding natural language and recognizing objects in images. More broadly, AI can be used to make better decisions by automating processes and optimizing outcomes.
There are many different ways businesses can use AI. For example, retailers might use it to personalize product recommendations or target ads. Banks might use it to prevent fraud or assess credit risk. Manufacturers might use it for the optimization of production lines or to predict machine failures.
In addition, AI can be used for a wide range of other tasks such as human resource management, marketing, and even cybersecurity. As the capabilities of AI continue to expand, so too will the ways in which businesses can make use of it.
Most businesses today are only starting to scratch the surface of what AI can do for them. The potential uses of AI are vast and evolving, and the best practices for implementing it are still being developed. However, CIOs who start exploring and experimenting with AI today can gain a competitive advantage over their peers.
There are many different subsets of AI, one of which is propensity modeling.
In 1983, statisticians Rosenbaum and Rubin introduced the idea of propensity modeling as a way to infer the causes and implications of behaviors. This is useful when methods like randomized trials and A/B testing can’t be feasibly carried out, such as when there’s a risk to revenue, when it could cause smaller bonuses for selected sales personnel, or when there could be ethical or health issues.
Over the past few decades, propensity scores have become a mainstay of medical and social research. They are also becoming an important tool for business leaders, including CIOs. In this section, we will introduce propensity models and explain how they can be used to improve decision-making.
Let’s start with a simple example. Assume you want to predict the likelihood that leads will convert to customers. In other words, you want to know what factors are associated with conversion. You might start by looking at variables such as age, sex, income, and location. However, there are many other potential factors that could impact conversion, such as the customer’s attitude or the quality of the product.
In order to account for all of these potential factors, you could build a propensity model. A propensity model is a mathematical formula that takes into account all of the known factors that are associated with conversion. The model then uses this information to estimate the likelihood that a given lead will convert to a customer.
There are many different types of propensity models. The most common type is a logistic regression model, which is used to predict the probability of a binary event (e.g., conversion). Other types of models include decision trees, neural networks, and Support Vector Machines.
There are many ways in which propensity models can be used by businesses. Here are a few examples.
When it comes to marketing, generating leads isn't the whole battle: converting those leads into paying customers is where the real money is made. In fact, statistics show that the global average website conversion rate based on multiple marketplace conversion rates is just 4.31%.
That means that over 95% of visitors to your site aren't converting into paying customers. If you're not constantly testing and optimizing your website to improve your conversion rate, you're leaving a lot of money on the table. By using propensity models to forecast the likelihood of a lead converting to a customer, businesses can allocate their resources in a more informed manner.
Of course, "conversion" can relate to any number of positive target actions, such as subscribing to an email list, engaging with a piece of content, or even just making a purchase. In many cases, propensity models can even be used to distinguish between different types of conversions, such as customers who are likely to become long-term vs. short-term customers.
These insights can then inform all kinds of business decisions—from the design of marketing campaigns to the allocation of discounts or free shipping.
CLV is a Northstar metric for many businesses and CIOs. CLV is the total dollar value of a customer’s transactions over their lifetime with a company. It is important to know which customers are most valuable so that you can focus your marketing efforts on them.
As McKinsey reports, customer spending is on a post-pandemic rise, making CLV more important than ever for businesses. In order to maximize CLV, businesses must focus on customer retention and expansion - that is, keeping existing customers happy and encouraging them to spend more.
A propensity model can be used to predict CLV. The model can identify which customers are most likely to stay with the company, buy more products, and generate more revenue over their lifetime. This information can be used to create targeted marketing campaigns and improve customer retention rates.
From a retailer's stock perspective, understanding what percentage of purchasers are those that the business would like to see more of (higher than the recent average purchase order amount) is valuable information. The application of a propensity model to post-transaction analysis can help identify which shoppers showed increased interest post-purchase and should be sent subsequent offers or coupons, for example.
This same idea could be used by other types of businesses as well. Airlines could use propensity models to identify customers who are likely to book another flight in the near future. Hotels could use them to identify guests who are likely to stay again, and so on.
Churn rate is the percentage of customers who discontinue their service with a company. It is important for companies to understand what causes customers to churn so they can take steps to prevent it.
With the prospect of post-pandemic churn looming large, now is the time for companies to really focus on churn prevention. With 80% of OTT services worried about post-pandemic churn, it's clear that this is a top concern for businesses.
A propensity model can be used to predict churn rate. The classifier model can identify which customer data factors are associated with churn and then estimate the likelihood that a customer will churn. This information can help companies focus their efforts on those customers with a high propensity to churn, improving the customer experience in the process and building a more data-driven organization.
From providing incentives to higher-touch support, there are many ways to keep that customer base on-board.
Propensity modeling isn't always a walk in the park, particularly with legacy systems that need to be revamped in order to support advanced artificial intelligence features. However, with careful planning and execution, CIOs can overcome these challenges and reap the benefits of propensity models in their organizations.
One of the key considerations for implementing propensity models is ensuring that sufficient resources are allocated to the effort. This includes not just compute power and data but also the necessary manpower to design, build, test, and operate the models. Often, this requires a significant change in organizational culture and a shift in thinking from simply using data to make decisions to using datasets to automate those decisions.
Another challenge is striking the right balance between accurately targeting customers and annoying them with too many automated messages. This depends on the particulars of each business and its customers, but it's important to have a good understanding of who is likely to respond to a particular offer or message. Too much automation can lead to decreased response rates and reduced profits.
Finally, it's still necessary for humans to intervene in some cases even when propensity models are deployed. For example, if new data becomes available that suggests a customer should be targeted differently than was originally thought, the model needs to be adjusted accordingly. Or if there's a problem with how the model is functioning, human analysts may need to step in to diagnose and fix it.
Despite these challenges, CIOs who are able to successfully implement propensity models can see impressive results. By automating decision-making processes and increasing customer engagement, businesses can realize significant cost savings and growth.
Businesses are increasingly looking to artificial intelligence to help them achieve their goals. However, many companies struggle to implement AI solutions due to a lack of expertise in coding and data science.
Akkio offers a solution to this problem with its no-code AI platform. Akkio allows businesses to build predictive models without any coding or data science experience. From conversion rates to customer churn, Akkio’s platform can be applied to a variety of business use cases.
Akkio has a simple 3-step process to integrate AI into your organization: Click to connect your data, select a column to automatically build a model, and deploy in any setting.
Let's discuss a case study of a business that has seen success with AI propensity modeling using the Akkio platform.
AngioDynamics is a medical device company that provides products and services for the treatment of cancer and vascular diseases. The company runs a large number of clinical trials, and clinical trial abandonment is a huge challenge for the company. Clinical trial abandonment is the process of discontinuing a clinical trial before it is completed.
Considering that clinical trials for novel therapeutic discovery cost between $12 million and $33 million per trial, abandonment has serious financial repercussions, and even worse, potential human costs.
AngioDynamics decided to use Akkio to predict which clinical trials were more likely to be abandoned. Algorithms were fueled with “over two hundred thousand clinical trials conducted over the last few years.” These studies included data on 80 trial attributes, such as title, location, anonymized demographics, the number of participants, all the way to detailed elements of the trial design.
Akkio’s natural language processing weighed those free form variables alongside quantitative data to predict trial status. The company found that Akkio was able to identify factors that are associated with trial abandonment with a high degree of accuracy.
There are three steps to using Akkio for propensity modeling:
The first step is to upload your data into Akkio. This can include historical data from your business, as well as data sources such as surveys or social media. Akkio includes native integrations with tools like Hubspot, Salesforce, Snowflake, and Google Sheets, but you can also simply upload a CSV or Excel file.
The next step is to train the model. You’ll simply select the column you’re interested in, and Akkio will automatically identify the best machine learning models for your data. It will also identify the most important features (predictors) in your data, and how to best use them to predict the target outcome you’re interested in. Akkio will also automatically conduct validation to ensure that your model will perform well in the real-world.
The final step is to use the model to make predictions about customer behavior. As with our data integrations, you can deploy your model in a wide variety of ways, from directly in CRMs to deploying the Akkio API or leveraging our no-code Zapier integration. With our visual interface, marketing teams can deploy these models without the need for data scientists.
This can be done in real-time, or retrospectively using historical data. You can also use the model to segment your customers into different groups, and then create targeted marketing campaigns based on those segments.
See our applications page for an exploration of a variety of AI use-cases.
Propensity modeling is a powerful tool for predicting future behavior. By applying machine learning algorithms to data, businesses can better understand the likelihood of specific actions or events occurring. In this report, we have shown how Akkio can be used to create propensity models for tasks from conversion prediction to churn prediction.
With Akkio's free trial, businesses of all sizes can start using propensity modeling to improve their predictions and make better decisions for their company.