How Machine Learning Is Helping In Providing Dynamic Pricing
This article dives into Machine Learning and Dynamic Pricing optimisation and how various companies are leverages it.
Overview Of Dynamic Pricing
Dynamic pricing is the practice of setting a price for a product or service based on current market conditions. Businesses reap the benefits from a huge amount of data amid the rapidly evolving digital economy by adjusting prices in real-time through dynamic pricing. “Dynamic pricing uses data to understand and act upon any number of changing market conditions, maximizing the opportunity for revenue,” says Alex Shartsis, founder and CEO of Perfect Price.
Dynamic pricing can be used in various price setting methods. According to Yigit Kocak of Prisync, the three of the most common methods are cost-based, competitor-based, and demand-based.
- Cost-based pricing “adjusts prices dynamically according to business costs and keeps profit margins on a certain level.”
- Competitor-based pricing takes into account competitor pricing decisions.
- Demand-based pricing speaks for itself: Prices increase with growing consumer demand and dwindling supply, and vice versa.
The challenge of setting the right price
Setting the right price for a good or service is an old problem in economic theory. There are a vast amount of pricing strategies that depend on the objective sought. One company may seek to maximize profitability on each unit sold or on the overall market share, while another company needs to access a new market or to protect an existing one. Moreover, different scenarios can coexist in the same company for different goods or customer segments.
These are some of the crucial questions that retailers recurrently face:
- What price should we set if we want to make the sale in less than a week?
- What is the fair price of this product, given the current state of the market, the period of the year, the competition, or the fact that it is a rare product?
What are traditional approaches to pricing?
Sellers used to set the price for a product or service based on a manual analysis of the cost, demand, supply or competition. Without sophisticated algorithms, two pricing strategies were common:
- Premium Pricing: Premium pricing is where companies set the price higher than average competitive price. The key factor for the success of this strategy is differentiation. Premium pricing effectively works when the product has a unique feature that differentiates it from similar products in the market and has a competitive advantage.
- Penetration Pricing: Penetration pricing is basically setting the price relatively lower than the market competition. Companies use this pricing strategy to raise brand awareness and increase customer loyalty. Initially, penetration pricing may cause revenue loss but the main goal of this strategy is market penetration.
Algorithms & machine learning for dynamic pricing
Algorithmic pricing is a process of setting optimal prices using the power of machine learning and artificial intelligence to maximize revenue, increase profit or gain other business goals set by retailers. Algorithmic pricing is one of the most powerful means of getting a competitive advantage.
How does dynamic pricing algorithm work?
In most cases, advanced pricing algorithms are powered by the combination of AI and ML technology. In contrast to traditional pricing, the dynamic approach ensures scalability of the pricing decisions. Subsequently, algorithmic pricing enables retailers to shift from SKU-centric to portfolio-based pricing in which all sorts of both explicit and implicit dependencies are considered.
Dynamic pricing algorithms also bring flexibility as retailers can set prices targeting different groups of shoppers. The latter is achieved by crafting an optimal value offering based on market trends, demand fluctuations, customer behavior, purchasing power, and plenty of other factors.
Dependency between price and demand is a core estimation when it comes to dynamic pricing algorithms. With relevant data on this dependency, the revenue-optimal price could be calculated using the formula below.
In the equation, p marks the price while d(p) stands for a demand function. The dynamic software engine extends this formula adding a range of other pricing and non-pricing factors to be considered. Typically, these may include procurement expenses, inventory costs, demand cannibalization between particular products, competitor prices, promo activities, and other factors. The more diverse and relevant data points are processed by pricing algorithm, the more accurate results it generates.
The vast majority of pricing algorithms use historical sales data based on which the demand function is estimated. The workflow of a typical pricing algorithm goes through the four main stages:
- Historical data on price points and demand on particular products is consumed by the engine to be processed using the dynamic pricing algorithm
- The demand function is build based on identified dependencies
- The state-of-the-art math processes dozens of pricing and non-pricing factors to generate optimal prices
- After the recommended prices are applied, the algorithm goes through the cycle again taking into account the latest repricing results
The workflow outlined above represents a unified basic pattern while, in each specific case, the model of dynamic pricing algorithm is aligned with the targets and constraints of a retailer.
Dynamic pricing providers offer a wide range of tech approaches to make their pricing engines more effective. Many optimization algorithms use the latest-generation neural networks capable of processing billions of repricing scenarios to ensure the integrity of results with the price effect prediction accuracy of 90–98%. But building a forecasting model is a complex process that might be different in every case depending on the particular goals and needs of a retailer.
Most dynamic pricing engine is based on a two-stage machine learning. The first stage implies calculating the precise effect of price changes on sales. And the second stage is state-of-the-art math price optimization which uses the results of the first stage to recommend prices for the whole portfolio.
USE CASE:
Implementation of Thompson sampling for dynamic pricing
An example of a dynamic pricing implementation with Thompson sampling is shown below. By running this implementation and recording how the parameters of the distributions are changing over time, we can observe how the algorithm explores and learns the demand function:
In the beginning, the demand parameters are the same for all price levels. The algorithm actively explores different prices (the red line in the bottom chart), becomes certain that the price of $3.99 provides the best revenue (the yellow curve in the middle chart), and starts to choose it most of the time, exploring other options only occasionally.
Different Modules of Dynamic Pricing
Because of the complexity of dynamic pricing, different modules are sometimes used for different product categories and market responses to manage complexity.
mage credit: McKinsey&Company
Long tail module
This module is for new products or long-tail products with little or no historical data. Main challenge for this module is to use product attributes to match products with little purchase data with products that have rich purchase data so prices can be informed by rich data.
A US retailer with more than two million range of products customized its long tail module algorithm. To build the long-tail module, company gathered a rich set of data for its 100,000 top-selling SKUs including competitor prices, data on customer behavior, product attributes and descriptions, and online metrics. Developers then worked with category managers to create attribute similarity scores and leveraged rich data of popular products to price products in the long-tail. Pilot resulted in 3% increase in both revenue and margin .
Elasticity module
Elasticity module calculates the impact of price on demand considering seasonality and cannibalization.
A leading Asian e-commerce player built an elasticity module based on a multi-factor algorithm that drew on ten terabytes of company’s transaction records. Data included product price, substitute price, promotions, inventory levels, seasonality, and competitors’ estimated sales volumes. Though price recommendations were generated real-time, category managers made the final pricing decisions. Pilot led to an increase of 10% in gross margin and 3% in GMV.
Key Value Items (KVI )module
Key value items are popular items whose prices consumers tend to remember more than other items. KVI module aims to manage consumer price perception by ensuring that items that strongly impact customer’s price perception are appropriately priced.
This is important for resellers like grocery companies. Because they are not selling their own products, they need to make sure that customers see them as the lowest cost option. A leading European nonfood retailer built a sophisticated KVI module statistically scoring each item’s importance to consumer price perception on a scale of 0 to 100. This scale guided pricing decisions and company was willing to lose more on KVIs to retain and improve the customer price perception about their company.
Competitive-response module
This module leverages granular pricing data from competitors and impact of those prices on company’s customers to react to competitors’ prices in real-time.
Though this is a relatively simple mechanism, two competitive-response modules competing with one another can create quite unexpected results like asking $23.6M for a book! Two 3rd party Amazon merchants had dynamic pricing models. While first merchant’s system aimed to sell its book at a price 27% more than the second merchant, that merchant dynamically sets its price to 1% less than the first merchant. Predictably, the price of the book skyrocketed at every iteration of the algorithms. This is why including price data in your dashboards make sense.
Omnichannel module
Companies manage prices between channels both for price discrimination and also to encourage customers to visit less costly channels. Omnichannel modules ensure that prices in different channels are coordinated.
Time-based pricing module
Online retailers may charge customers more or less at the specific time of the day due to the following reasons:
- seasonality of products
- retailers charge more between 9 AM-5 PM since most online retail customers shop more during weekly office hours
- if customers want a same-day delivery or shopped right before end of the working hour, retailers are eager to charge more
- if the product has an expiration date, as time goes by, the price of the product decreases.
Conversion rate pricing module
If most leads (which are the views of the website in retail terminology) aren’t turning into sales for specific products, then retailers may lower the price to increase conversion rate.
More opportunities of using Machine Learning for price optimization
Machine Learning can be used for other tasks related to pricing in retail. For example, given a new product, a clustering algorithm can quickly associate it with similar products to obtain a probable price segment. Another compelling possibility is to jointly predict prices and demands for items that were never sold.
More generally, Machine Learning can be a tremendous tool for insights:
- In what way is the sale of pants impacted when shirts’ prices are drastically cut?
- When efforts are made to sell more pens, are the related products, such as ink, notebooks or work agendas, impacted?
- Are customers who buy a certain computer more or less likely to buy monitors the following month?
- Are inactive clients in the last year sensitive to a promotion campaign?
These are just some examples of the questions that Machine Learning models can help answer.
Advantages of price optimization with Machine Learning
In addition to automation and speed, there are several advantages to using Machine Learning to optimize prices.
First, Machine Learning models can consider a huge number of products and optimize prices globally. The number and nature of parameters and their multiple sources and channels allow them to make decisions using fine criteria. This is a daunting task if retailers try to do it manually, or even using basic software.
For example, it is known that changing the price of a product often impacts the sales of other products in ways that are very hard to predict for a human. In most cases, the accuracy of a Machine Learning solution will be significantly higher than that of a human. In addition, retailers can modify the KPI and immediately see how the models recalculate prices for the new goals.
Second, by analyzing a large amount of past and current data, a Machine Learning can anticipate trends early enough. This is a key issue that allows retailers to make appropriate decisions to adjust prices.
Finally, in the case of a competitive pricing strategy, Machine Learning solutions can continuously crawl the web and social media to gather valuable information about prices of competitors for the same or similar products, what customers say about products and competitors, considering hot deals, as well as the price history over the last number of days or weeks.
A system that can learn most of what is happening in the market allows retailers to have more information than their competitors in order to make better decisions.
Which industries use dynamic pricing?
Airlines
Airlines are the earliest adopters of dynamic pricing. A ticket for the exact same flight with the same destination and at the same date can have a number of different prices for different customers. Because airline sales moved online earlier than other categories and because airlines are expected to charge different prices for the same ticket bought on different days, it was easy and acceptable for airlines to move to dynamic pricing.
e-Commerce
Retailers, especially e-commerce companies like Amazon, eBay, Myntra, etc. use dynamic pricing for personalized pricing. If you consistently buy from Amazon or another e-commerce website, prices will be higher. Algorithms calculate the loyalty level of each customer and set the price lower if a person is a newcomer.
Dynamic pricing is now used for almost every product and service. From the price of a concert ticket to the price of a hotel booking is calculated by algorithms. Even Uber is using surge pricing.
Hospitality
For hotel management and tour companies, seasonality is an important factor. Using time-based pricing that means increasing prices during peak season and lowering when the season ends increases profitability. During the peak season, hotels’ supply needs also increases that’s why charging guests higher is not an immoral idea but the goal of the management should be finding the highest price that consumers are willing to pay.
Car Rental
Prices of rental cars fluctuate depending on season and day-of-the-week effect. According to a study made by Thinknum, weekend prices are more expensive than weekdays and summer prices are higher than winter prices. Here are the datasets about seasonal average car prices and discount percentages from the same study.
Final Thoughts
Dynamic Pricing is a very good strategy especially for industries that seeks to optimize profit and minimize cost. It is hard time different companies and product managers leveraged the power of dynamic pricing in their products to achieve the best of their product goals.
If you like this article, kindly give it a thumps up. Thanks in advance.
If you wish to write in this Newsletter, kindly reach out to us via Whatsapp: +919467891831
You may want to check: Full Stack Data Scientist BootCamp
— — — — — — — — — — — — — — — — — — — — — — — — — — — —
SUBSCRIBE TO Newsletter For FREE: https://lnkd.in/ewB9KR4j
OTHER ARTICLES:
- Robots Are Hiring Data Scientist. What You Should Know Before Applying For Your Next Job
- Data Science & Analytics are Changing University Course Design BUT Software Engineering Has A Word Of Caution
- Should You Work In A Startup or MNC? As A Beginner Data Scientist
- Which Programming Language Is Best For Data Science In 2022: Python or R?
- How to Properly Analyze Your Personal LinkedIn Data With Python