The traditional ingenuity and competitiveness of the Retail industry is being put the severest test. Customers have become impatient, competition has grown, and lifestyles/ needs are changing fast and in ways not easily fathomable. The “new normal” is just around the corner and the industry needs to take a major evolutionary step forward. The next advance in industry processes has to be around customer understanding. What products and services do customers want? Where and when do they want to make the purchase? What are the brand and product attributes that will win their loyalty? How do they want to pay? Smart investments in near-real-time customer analytics can unlock precise answers to these questions—and at a pace that provides a competitive edge. But that’s only part of the answer.
Having a finger on the pulse of the customer was always central to success. Marketing spends were traditionally based on needs, propensity to buy, wallet size and perceived value. These haven’t changed. But the tools to understand the customer, at a highly granular level, have improved dramatically. Today, retail leaders are harnessing the power of data, predictive analytics and Machine Learning to gain a deeper understanding of the customer. The outcome of using these technologies is a comprehensive and dependable view of Customer Lifetime Value (CLTV). CLTV allows retailers to target profitable customers, strategically position themselves in the customer journey and deliver high lifetime value. What happens is this: The retailer makes every marketing dollar go the extra mile. CLTV is an estimate of how valuable a customer will be to business across a specific period of time. It is the total worth of the customer to business over the tenure of engagement.
Retailers can make better strategic decisions by calculating and tracking CLTV with their Customer Acquisition Cost (CAC) to measure overall profitability. Ideally, a retailer’s CLTV should be higher than its CAC. Thus, the CLTV metric plays a crucial role to evaluate the customer base health and acquisition strategies. If a retailer cannot reduce the churn rate, then they are leaking value. To keep the business profitable, measures should be taken to either reduce the CAC or increase the CLTV.
CLTV metric helps a retailer to optimize sales and marketing strategies
The CLTV acts as a benchmark for the growth and expansion of a business. It bridges the gap between the marketing and financial metrics implying that the marketing activities depend on the finances. The CLTV metric helps a retailer to optimize sales and marketing strategies and generate opportunities to achieve better value through superior customer targeting and personalization of offers and promotions. The golden rule of 80:20 shows that 80% of an organization’s revenue comes from 20% of existing customers. The CLTV metric helps the retailer to identify those 20% profitable customers and define strategies to optimize the allocation of resources to maximize ROI. CLTV is thus a customer-centric metric which provides a holistic and evidence-driven approach for brand building along with long term revenue objective.
CLTV is based on customer association, transaction frequency and customer behavior
Before calculating the CLTV, the nature of the customer-organization engagement should be thoroughly understood. Based on the engagement the method for measuring CLTV is selected. A customer’s engagement can be based on 3 drivers for CLTV calculation:
❖ Customer lifetime or duration of customer association with the organization’s products or services
- Contractual – An organization knows when a customer will churn out. The buyer-seller relationship is governed by a subscription or contract. Example – insurance services and subscription-based services like Hotstar, Amazon Prime and Netflix
- Non-contractual – An organization is unsure when a customer will churn out. The buyer-seller relationship doesn’t have a contract and is governed by demand and supply. Example – FMCG products and customer-facing services
❖ Inter-purchase time of a customer
- Discrete – The transaction occurs at a distinct point. Example – subscriptions of Netflix, Spotify, magazines and newspapers. These are mostly deterministic, based on a contract
- Continuous –The transaction occurs at any point. Example – grocery purchases and hotel stays. These are highly stochastic
❖ Customer spend behavior
- Fixed – The amount spent by a customer almost same in every transaction. Example –Magazine subscriptions
- Variable –The amount spent by a customer varies scenario basis. Example – grocery purchases, retail and airline bookings
Modeling approaches to calculate CLTV
Various modeling approaches can calculate the CLTV. These approaches can be classified into two categories – past customer behavior models and future-past customer behavior models. The major differences between these models include the assumption of activeness of targeted customers in the future and the customer spend in the future. The first category includes – PCV Models (Past customer value), RFM (recency, frequency, monetary), SOW model (share of wallet) which calculate CLTV by using past data of the customers whereas the second category includes probabilistic and econometric approaches which consider the future behavior of the customer.
Based on a customer-organization engagement, different methodologies are used to derive CLTV:
❖ Historical CLTV – Sum of all the gross profits from the historical purchases of a customer. The Simple CLTV Calculation – (Average sales per customer * Number of transactions) – Customer acquisition cost. The drawback of this calculation is that the annual revenue of a customer should remain consistent.
❖ Predictive CLTV – More efficient way of calculating CLTV. It considers the history of all the transactions made along with the other purchase behavior characteristics to predict the CLTV. This traditional method can handle revenue fluctuations over time and is adjusted by the discount rate to account for inflation.
The Traditional CLTV Calculation – Average Profit Margin per customer lifespan * (Retention rate / 1 + Discount Rate – Retention rate) – Acquisition cost
Use probabilistic and econometric models for CLTV analysis in retail
The retail business is largely confined to non-contractual, continuous, and variable purchase settings. The standard method of CLTV has limitations here. The most suitable approaches are the probabilistic and econometric models.
❖ Probabilistic model-based approach – This is the buy till you die approach that uses (BG/NBD) models which measure CLTV by evaluating the frequency of future transactions and the probability of churn of a customer and (Gamma-Gamma) model to assess the expected spend of a customer per visit. All the values are then substituted in the traditional CLTV formula to obtain a value for each customer. There are other prototype models as well that are used – Pareto/NBD and Extended Pareto/NBD model
Limitation – The BG/NBD approach assumes that the number of transactions made by each customer follows a Poisson distribution, but real-time data sometimes violates this assumption
❖ Regression and classification model-based approach – A regression model estimates the spending of a customer in each of the upcoming periods of engagement whereas a multi-label-classification model predicts the probability of a customer to buy in that period. The overall expected spending is calculated by multiplying the two. After estimating the parameters, the CLTV calculation is made using the traditional CLTV formula
Limitation – The regression models consider various historical transaction attributes like RFM to forecast the future buying behavior of customers. This has major implications while making future predictions depending on the past customer behavior as the RFM values can differ from one period to another thus resulting in different model parameters
Our approach to calibrate CLTV based on frequency of customer visit and per visit spend predictions
The two key metrics required to calculate CLTV are the frequency of visits of a customer and the predicted spend in each visit. The broad steps taken by us are:
❖ Need and Behavior Segmentation: Before building the CLTV model we must understand the customer base and to do so we segment customers using advanced segmentation techniques like Hierarchical Clustering, K-proto Clustering and PAM Clustering. The segmentation is done based on the needs or behavior of customers (Need-based Segmentation and Behavioral Segmentation). Need-based Segmentation provides customer profiles based on the quantity/type of items they have purchased whereas Behavioral Segmentation gives us customer profiles based on their demographic (age, gender, etc.), purchase (ATV, UPT, recency, frequency, number of brands purchased) and socio-economic attributes (income, credit score, employment status, education indicator). By merging the customer profiles from Need and Behavior Segmentation we prepare our final customer focus groups.
❖ Amalgamation of Time Series and Probabilistic Modeling techniques: ITC Infotech has taken an approach for calibrating the lifetime value of each customer group in the next 2 to 3 years, depending on the loyalty index of customers. For predicting future visits and the spend of a customer we take an amalgamation of Time Series and Probabilistic Modeling techniques. The forecasts have been fine- tuned by incorporating essential factors like discount rate, retention rate, gross profit, etc. Depending on requirements we can further dissect the lifetime value at a granular customer level.
Increased revenue by ~ 3% using CLTV
To do a comprehensive Customer Lifecycle Analysis we have deployed the lifetime value model along with propensity to buy models, advance recommendation engine and basket analysis, churn prediction model and demand forecasting models for our retail clients. For one of the retail giants based out of South Africa, these exercises have resulted in a promising uplift of 3% for the financial year 2019-20. The lifetime value models have helped retail clients identify valuable customer segments, strategize their marketing decisions, and understand the contribution of each customer to the bottom-line.
Our customer intelligence platform for retail accelerates the customer lifetime value
ITC Infotech’s end-to-end integrated customer intelligence platform with pre-built AI/ML models for predictive and prescriptive analysis use the above-mentioned analytical models and the approach to improve the customer lifetime value. Our deep retail domain expertise with data science skills and experiences with leading retailers and CPG companies has helped us collaboratively develop this platform for accelerated value realization. Do connect with us to know more about our Customer Intelligence Platform with unique commercial model and plug and play components.
Author – Astha Jaggi
Associate Data Scientist
Author- Nivedita Roy
Principal Data Scientist
Author- Sumela Banerjee
Senior Data Scientist