Using Data Science to Improve Customer Acquisition
Data science allows companies to work smarter not harder when it comes to customer acquisition strategy
As data analytics increasingly becomes used by companies and non-profit organizations of all sizes, businesses that decide not to invest in understanding their own data are falling behind the eight-ball. The term data analytics describes the process of using data or statistics to discover, interpret, and communicate meaningful patterns or trends.
Within the larger field of data analytics, data science is a sub-field that focuses on predicting the future rather than describing the past. As defined by IBM, data science combines math and statistics, specialized programming, advanced analytics, artificial intelligence (AI), and machine learning with specific subject matter expertise to uncover actionable insights hidden in an organization’s data. These insights can be used to guide decision making and strategic planning.
Don’t let these verbose definitions intimidate you. Data science is all about turning data into useful information. How can we make better decisions as a business based on what we understand about the past and can predict about the future?
Data science is leveraged by some of the most successful and recognizable companies. Amazon uses data science to proactively predict purchases and move products to local delivery centers. Netflix uses data science to suggest which shows its viewers should watch next. Target uses data science to provide custom promotional deals that are more likely to get customers in their stores.
But data science is no longer a luxury of established giants, especially when it comes to customer acquisition-related use cases. Even SMBs and nonprofits can now take advantage of data science. In this article, we will look at a few different data science techniques that can help even resource-limited organizations improve their customer acquisition efficiency and return on investment.
Building High-Value Lookalike Audiences
Within a business’s full population of customers, there are often different types of customers with unique traits, needs, and behaviors. For example, a barbershop might have one group of customers who are regulars that come monthly and another group of customers who have low return rates. In this example, let’s say those regular customers tend to be over 40 years old, and the non-regular customers tend to be under 35 years old. Understanding this average age difference between these two customer categories is valuable information, because this barber now knows that acquiring a customer over the age of 40 is more likely to lead to return business than a younger customer.
A lookalike audience is a group of people who are not active clients of a business, but who share unique characteristics, behaviors, or traits with a group of existing customers. Understanding the characteristics of a business’s best customers can lead to understanding the right type of potential customers to target for future sales and marketing. How can a business go about doing this?
The first step of building lookalike audiences for your business is to explore commonalities amongst the highest value or most profitable existing customers you have. This is done by identifying behaviors and characteristics that these highest-value customers have in common that lower-value or less reliable customers do not. This could include demographic information such as age or gender, the products or services being purchased, and the source from which customers were originally acquired. This could even include more complex data, such as customer emails or other forms of unstructured text.
The key here is to find meaningful differentiators between the customers who bring in the most profit and all other customers for your business. It is important to note that while this exercise can be done through manual or non-technical methods, machine learning methodologies like classification modeling can save time and offer more precise (and often overlooked) findings.
Once you have created a persona for your highest-value customers, the next step is to determine the place or places where these customers are frequently acquired. How did this type of customer find your business? What are the best places to reach this audience? To which incentives or messaging has this audience historically responded positively?
Using this information, specifically target leads or audiences (i.e. lookalike audiences) with similar traits or behaviors. If you target potential customers that are similar to your highest-value customers, those who you convert will be more likely to fall into the “high-value” pool of customers. If you use lookalike audiences correctly, your business can expect to see an improved average customer lifetime value (LTV) as well as improved customer retention rates.
A good place to start in terms of the data to analyze would be customer sales data and marketing data, including data from third-party tools that might be collecting this information on your behalf (SalesForce, Google Analytics, etc). Give this article a read to learn more about how to get started with lookalike models.
Enhancing Marketing & Sales Mix with Predictive Modeling
Most businesses do not rely on a single channel for 100% of their sales. Depending on the nature of the business, they likely rely on a combination of marketing efforts, sales teams, and referrals. Even within marketing and sales, a well-organized customer acquisition effort includes multiple sales and marketing channels. Diversifying your customer acquisition efforts both online and offline is critical to a healthy flow of new business. Learn some of the ways businesses should be diversifying their digital acquisition efforts in this article.
Some of the most common channels for customer acquisition include Google Ads, search engine optimization (SEO), social media, content marketing, sales reps, cold calling, print advertising, and direct mailing. There are many other channels as well, and the truth is there are so many ways to reach customers in the modern world that it would be impossible to list them all. So if a healthy business is investing time and money into a combination of acquisition efforts, how do they know the optimal way to distribute their resources?
That is where predictive modeling comes into play. Predictive modeling is a mathematical process used to predict future events or outcomes by analyzing patterns in a given set of input data. In layman’s terms, predictive modeling is about predicting the future based on understanding the past. There are different types of predictive models- regression, classification, and clustering to name a few- but for the sake of this article, we won’t go into that level of detail.
To enhance your marketing & sales mixes for your business, the first step of the predictive modeling process is to collect and store all relevant marketing, sales, or lead generation data (i.e. SalesForce, HubSpot, Google Analytics). Identify all of the data available related to these sales and marketing initiatives, including but not limited to data points like sales rep, lead source, demographic information, communication medium, lead open date, and lead industry.
There are some simplified tools to run predictive models. For example Excel has relatively unsophisticated yet user-friendly plug-ins you can use. For the sake of leveraging predictive modeling in a way that is going to be precise, accurate, and financially useful, it is recommended that you use machine learning. Using data science techniques, run a few different model types to determine the most appropriate and accurate predictive model based on the ability to uniquely predict sales outcomes for your business. For different datasets and different businesses, the best models to use will vary.
Once your business has found the most accurate model based on its internal data, break apart this model to determine the variables that have the strongest ability to predict (and perhaps even impact) sales outcomes. Evaluate these variables to determine which the business can control (i.e. lead source, sales rep) and which are random or uncontrollable. From there, you can adjust your marketing or sales strategies based on the variables that can be controlled.
These levers have the strongest impact on your marketing or sales return on investment. When leveraged appropriately, predictive models should improve a business’s return on marketing investment, return on ad spend, and cost to acquire each conversion/sale.
Delivering Customized Promotions and Discount Offers
Businesses use promotional deals and discount offers to incentivize customers to make transactions that they might not have otherwise made. There is something psychological about getting an exclusive or temporary deal that increases people’s likelihood of purchasing something. But not all deals are appropriate for all potential customers. Different customers are motivated by different things, and being able to service customers at the individual level ultimately is more effective than offering one-size fits all solutions.
Promotion customization involves providing unique promotions, offerings, or discounts to different segments of your customers or leads with unique characteristics so that each individual is likely to spend the largest collective amount based on personalized considerations. This practice can take many forms. One example of promotion customization is targeting different income levels with price-appropriate products or services. Another example would be offering different promotions or discounts based on what each segment is inclined to take interest in (i.e. cross-selling product A vs B for a discount). One more example would be offering unique promotions or discounts to different traffic channels based on the characteristics of the users in each marketing or sales channel (i.e. unique offerings for leads who found your business through Facebook vs. Google).
Let’s look at an example of a business that successfully pulled off promotion customization. Loanry is a marketing lead generator designed to provide quick and easy access to third-party lenders. Loanry split their customers up into age and location to segment them in terms of affluence and perceived wealth. This allowed them to target higher-earning customers with higher rates of investment and real estate opportunities. Loanry then used different low-level incentives for lower-income customers as not to offer opportunities that would be a financial stretch. The result of a segmented incentive program was an overall increase in response rate from 7% to 11% year over year- a massive increase given the scale.
If a business is interested in creating similarly-customized promotions, the first step is to collect and take account of the data available to you. From there, perform a clustering or segment analysis using machine learning (or comparable data science no-code-required tools) to better understand the different “personas'' of users who interact with your business. What are the unique interests or behaviors of each persona? Through the use of historical data as well as leveraging A/B testing, you can run a series of experiments targeting each unique “persona” to determine how each segment of your customer base responds to a series of potential promotions or discounts. Based on the performance of these tests, you will be able to identify which marketing communications, promotions, or discounts generated the best response for each unique segment of your customer pool.
Companies who successfully undertake this endeavor will ultimately be able to provide their customers with customized deals that are built to the individual. The deals themselves are not just customized in terms of pricing, but also in terms of timing and messaging. Customized promotions are ultimately a tool to increase conversion rate and overall sales.
Leveraging Recommendation Systems
Have you ever been shopping on Amazon or another e-commerce platform, navigated to your cart, then received recommendations of other items you might want to purchase before checking out? The recommendations you received were not random. Every item shown was selected based on past purchasing patterns and behavior. A recommendation system is an artificial intelligence or AI algorithm, usually associated with machine learning, that uses Big Data to suggest or recommend additional products to consumers.
Recommendation systems are designed to generate the maximum amount of revenue possible by understanding what a customer is most likely to get upsold or cross-sold. If a customer is recommended something that they are not likely to purchase, then the recommendation did not provide any additional value to the seller.
A good example of a successful recommendation system in action is DoorDash, the third party food delivery company. When purchasing food with one of DoorDash’s partner businesses, DoorDash will recommend additional things to buy that complement the existing cart. If you are ordering a pizza, DoorDash might recommend you also buy marinara sauce or garlic dip, for example. This algorithm is based on two primary factors: 1) which products are frequently purchased together and 2) which products have the individual customer (or audiences similar to the individual customer) purchased in the past.
In order for a business to implement their own recommendation algorithms, they should perform a machine learning-based analysis of all of their products or services to better understand the relationship between unique offerings. They can use this analysis to identify which products and/or services are most frequently purchased together by their customers and explore reasons why their pairings happen to go together. From there, when users are going through the purchase funnel (whether that be online, in person, or over the phone), a business can make personalized suggestions to the customers about potential cross-selling or upselling opportunities.
There is a more manual way to go about this as well, and to a certain extent, a lot of this could be replicated in a simple Excel workbook. Machine learning allows for this exercise to be dynamic and auto-adjust over time. Machine learning also allows you to identify patterns that are less obvious, such as pairings that infrequently purchased together but lead to higher customer retention rates.
Recommendation algorithms are about more than just increased customer satisfaction. They lead to increased cross-selling & upselling, customer retention, and overall conversion rates. A business who is successfully using a recommendation engine is giving customers purchasing options that they are more likely to respond positively to based on the individual needs of that customer.
All of the examples covered above serve one larger purpose- building smarter customer acquisition processes. Data science can help identify trends, opportunities, or patterns that would not otherwise be obvious to business decision makers. This increased insight ultimately leads to better decision making, which in turn leads to increased sales, profit, and business growth.
Give this article a read if you are interested in learning more about affordable and effective data analytics & data science use cases for small businesses. If you are interested in partnering with an agency to help you fulfill your data science needs, book a free consultation with Render Analytics.