Discounting is a timeless marketing strategy. If done well, it can give just the nudge needed to turn hesitant decision makers into buyers, reward the right customers, and boost sales. If done poorly, it devalues the product or service, incentivizes the wrong customers and throws money away.
So, how does AI help brands make better discounting decisions? There are essentially 4 types of decisions:
1. Who to give discounts to?
2. When to give discounts?
3. Which products to give discounts on?
4. How much discount to give?
Let’s unpack how AI helps with each of these decisions, using the example of a fictional fast fashion brand called Sizzle.
Who to give discounts to?
AI models can help determine which customers are most likely to buy when offered a discount. This is typically done through propensity modeling, in which based on customer’s past behaviour, a propensity score is calculated. The higher the score, the higher the chances of a customer responding positively to a discount. For example, AI model may determine Sizzle’s fashion forward customer segment to be responsive to discount on new arrivals. Customers in this segment may have features like – they regularly buy latest arrivals, follow trends and engage with the brand on social media.
Grouping customers on discount sensitivity can also be done through classification algorithms. The model can classify the customers as “likely to respond to a discount” and “unlikely to respond to a discount” and focus on the customer segment that is likely to respond.
Another approach is to predict the LTV (Lifetime value) of customers based on their transaction data and offer selective discounts to high LTV customers to drive up loyalty and engagement. For example, the AI model may identify offer discount to a customer Alisha who has high predicted LTV based on her behavior – She shops twice a month on Sizzle on an average, buys across product categories, has low return rate and leaves positive ratings on products.
Customers who are likely to buy a product based on their similarity to other users or other products they bought in the past represent another category of customers who can be targeted for discounts. By predicting which product a customer is likely to buy using techniques like collaborative filtering or content based filtering, discounts can be targeted towards these products at the optimal time. For example, customers who buy Boho chic floral maxi dress also tend to buy strappy block heels. By offering discount on block heels when a customer buys the maxi dress will help drive sales of both products.
If a brand wants to attract new customers or enter a new market, it can leverage AI to determine which new segments to target or which geographical region to enter. Similarly, while launching a new product, AI can determine which customer segments are likely to be interested in that product, and then offer discounts to those customers on the new product. For example, while launching an eco-friendly line of clothing, the AI model may determine that female customers in the age group of 25 to 36 who typically buy casual wear and accessories, are active on social media, and follow sustainable fashion influencers may be most interested in this product line.
Inactive customers is another segment where AI can help offer the right discount prevent customer churn. Let’s say the model determines Betsy to be a customer likely to lapse, based on her last purchase being older than 3 months, low engagement with brand and low open rate on marketing emails. The model also determines that Betsy being a value shopper will buy products she likes if available on a discount. So, the brand can offer a discount on her next purchase while highlighting products she may like.
When to give discounts
AI models can help predict the time of a customer’s next purchase, and brands may offer discounts around that time to nudge the sale. For instance, a survival analysis may determine that Sizzle’s customer Dina typically makes a purchase every 30 days, and it's been 25 days since her last purchase. So, there is a high probability that she will make a purchase in the next 5 days. So, Sizzle may offer a discount around that time to increase the likelihood or value of the sale.
Time series forecasting is another method to determine the timing of discounts by predicting trends and patterns over time, such as seasonal fluctuations or peak shopping periods. For example, Sizzle’s AI model may predict that summer dresses that have consistently sold well during the spring and summer seasons may see an unexpected dip in the coming spring season. This may be due to factors such as shift in fashion trends, a downturn or customer’s spending behavior where they may want to spend less in spring after heavy spending in winter. This insight will allow Sizzle to discount this product to push sales.
Brands can predict periods of high or low spending using AI using linear regression models using past purchase history and time the discounts. They can time the discounts with high spending periods to push sales even further or time them during low spending periods to stimulate sales.
Brands can also deploy reinforcement learning techniques to determine best times for discounts through trial and error. In this type of machine learning method, an agent learns to make decisions by taking actions and receiving rewards (increase in sales) for correct actions and penalties for incorrect actions.
Which product to give discount on
Association Rule Learning can identify associations between products and by offering a discounted price on that bundle, it may stimulate higher sales of one or both products. For example, Sizzle’s satin marble dress offered as a bundle with sequined silver clutch may drive higher sales.
Using collaborative filtering, if customers like a certain product, then by predicting which products they may be interested in and offering discounts on those products may drive higher sales. For instance, the model may determine that customers who like strappy crepe dress also tend to like flared jumpsuit, and therefore, by offering discount on the flared jumpsuit, Sizzle can drive additional sales.
If a predictive analytics model predicts that a customer is likely to purchase a certain product, offering a discount on that product may increase conversion rates.
By using Natural Language Processing in Sentiment Analysis through customer reviews and social media posts, a brand can identify products that are currently popular or trending. Offering discounts on these products can capitalize on their current popularity.
Using inventory Management Models, AI can predict demand for different products and optimize inventory accordingly. If an excess of a particular product is predicted, discounts can help balance supply and demand.
How much discount to give
There are different AI models such as regression models, decision trees, random forest, gradient boosting models, reinforcement learning models which predict optimal discount to maximize revenue. The models balance the increase in sales with the decrease in profit per unit due to the discount. They consume data such as product type, customer demographics, time of the year, and past sales data. Then they identify the most relevant variables that impact sales and revenue such as original price, previous discount levels, time of the year, and customer purchasing behavior. Once the model is trained, then the model predicts the discount.
No customer will ever say – “I dislike getting discounts on products I like when I need them”. As AI models become more sophisticated and powerful, brands that can effectively leverage AI powered discounting will gain competitive advantage. A good place for brands to start with their AI efforts is to get their data in order.
(Article originally published at: https://www.linkedin.com/pulse/how-ai-transforming-discount-dilemmas-sales-success-neha-verma-bacgc/ )
Posts
No comments:
Post a Comment