Dynamic pricing: In-depth Guide to Improved Margins 
In this paper we present an end-to-end framework for addressing the problem of dynamic pricing on E-commerce platform using methods based on deep reinforcement learning DRL. By using four groups of different business data to represent the states of each time period, we model the dynamic pricing problem as a Markov Decision Process MDP.
Compared with the state-of-the-art DRL-based dynamic pricing algorithms, our approaches make the following three contributions.
First, we extend the discrete set problem to the continuous price set. Second, instead of using revenue as the reward function directly, we define a new function named difference of revenue conversion rates DRCR. Third, the cold-start problem of MDP is tackled by pre-training and evaluation using some carefully chosen historical sales data. Our approaches are evaluated by both offline evaluation method using real dataset of Alibaba Inc. In particular, experiment results suggest that DRCR is a more appropriate reward function than revenue, which is widely used by current literature.
In the end, field experiments, which last for months on stock keeping units SKUs of products demonstrate that continuous price sets have better performance than discrete sets and show that our approaches significantly outperformed the manual pricing by operation experts.
Reinforcement learning Dynamic pricing E-commerce Revenue management Field experiment. Dynamic pricing, to adjust prices according to inventories left and demand response observed, has drawn great attentions during the past decades since the deregulation of the airline industry in the s.
During the recent development of business, many industries have become more active in revenue management. Retailers like Zara have implemented systematic dynamic markdown pricing strategy [ 4 ]. Kroger is now testing electronic price tag at one store in Kentucky [ 5 ]. Online retailers have a stronger desire for dynamic pricing strategies due to the requirement of more complex operations. For example, Amazon. Operation specialists have to set prices for these items periodically to remain competitive while maximize revenue, which will be mission impossible when the number of items goes this high.
As a result, Amazon has implemented automatic pricing systems and it is reported that Amazon. In this paper, we proposed a reinforcement learning approach to address the dynamic pricing problem for online retailers. The scenario we consider is how to dynamically price for different products on Tmall. There are many difficulties for pricing on such an E-commerce platform.
First, the market environment is impossible to be quantified. Second, it would lead to non-convergence policies if the reward function is not set properly under such complicated environment. Third, it is not appropriate to apply a learning model online directly, since a slightly inappropriate price online could quickly cause large capital loss. That is an obstacle to evaluate the performances of different pricing policies during the field experiment.In recent years, artificial intelligence has enabled pricing solutions to track buying trends and determine more competitive product prices.
While static pricing keeps prices absolute, dynamic pricing adjusts prices to offer customers different prices based on external factors and their individual buying habits. While there are some dynamic pricing options on the market, automatic dynamic pricing software is in its infancy.
These patterns are unveiled by analyzing a variety of sources, such as loyalty cards and postal codes, in order to predict what the customer is willing to pay and how responsive they might be to special offers.
Machine learning algorithms can also reveal pricing gaps of which businesses can take advantage. In this article, we aim to provide business leaders with a glimpse of:.
Pace claims to have developed a software that uses machine learning to enable hotel management to explore pricing that matches supply and demand. This could allow hotels to maximize their profits by offering the price that customers are willing to pay based on their demographics and the time of year among other factors. Hotels could forecast increases or decreases in bookings based on certain customer demographics and adjust prices accordingly.
For example, students and elderly customers might pay less than people staying at the hotel for business. Holidays may also affect the price for each of these segments.
Pace has not released any videos demonstrating their software. Pace does not mention any larger clients on their website. Artur Matos is the CTO at Pace, holding a Doctorate degree in Computer Science from the Nagoya University, where he conducted research in genetic programming, genetic algorithms, and other evolutionary computation techniques. Perfect price is an AI-powered dynamic pricing solution that claims to enable companies, such as car rental companies, to do dynamic pricing.
According to Shartsis, Perfect Price calculates pricing based on microsegments to better fit price with demand. Traditionally, car rental companies segment based on time of day, boosting morning prices to match business travelers who are assumed to be more willing to pay.
What is Machine Learning’s role in Dynamic Pricing?
With Perfect Price, every car class and location is set as its own microsegment, and it can separate days into 24 microsegments. Shartsis claims the application is able to determine if a certain car shows higher demand in a specific area and at a specific time of day, resulting in surge pricing while not affecting other car classes.
With automated pricing capabilities, the system could need minimal human oversight, he added. How exactly the software does this does not seem to be publically available. It could also test promotions and price changes on whether or not they affect revenue. Perfect Price does not provide any video demonstration of their application, but the company makes available some case studies.
In one study, the Betabrand crowdfunded clothing community reports that it needed to reflect new prices on its products, but did not know how to test it.
They turned to Perfect Price, which used the e-commerce edition of its application to experiment with new pricing and promotion. Perfect Price does not mention any larger clients on its website.
How Dynamic Pricing Uses Machine Learning to Increase Revenue
CTO and Co-founder Youngin Shin is responsible for data mining and machine learning within the company. The company claims that the software uses machine learning to track more than 1 billion unique products across more thanbrands over 1, categories. To use the application, the user inputs their business website into the search engine, chooses the industry to which the business belongs, and lists their top 5 competitors. The data is presented by price range and product performance, allowing the user to see trends and competitor activity.
The user can also drill down to view more detailed data at a subcategory and product attribute level. E-commerce businesses with this information could spot what is currently popular in their market and know exactly which products shoppers want.
They could access rich market data down to individual SKUs, product attributes, categories, and brands and gain visibility into product catalogs. They could even pinpoint hot categories and products for timely promotions to increase sales. Additionally, knowing what shoppers are buying could help businesses adapt their merchandise and prevent out-of-stocks or excess inventory.Dynamic pricing allows large and small companies improve their margins quickly.
Any corporate leader needs to know about dynamic pricing and we answer all dynamic pricing questions here:. 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:. Profit maximization is not always possible with both strategies. At premium price level, demand would be low.
Even if you have a high demand for penetration pricing, the price will remain low. What if you can cover all the price segments and respond faster to demand fluctuations in the market? This is possible with price discrimination. In the pre-internet days, companies capabilities for setting different prices for different customers were limited. Student fares are a common mechanism to set cheaper prices to customers who have less willingness to pay.
Time-specific products such as transportation or hospitality products also have the advantage of changing prices based on time of purchase. However, e-commerce enabled companies like Amazon to develop digital personalized stores for each customer.
Each customer gets personalized product suggestions and personalized prices. This started a golden age for price discrimination where companies can offer customers prices based on their exact willingness to pay. Price discrimination in the digital world is commonly called dynamic pricing. Price may even change from customer to customer based on their purchase habits.
Dynamic pricing is the strongest profitability lever. It is difficult to achieve significant financial improvement in a large company. Most large organizations. However, dynamic pricing is one of the few approaches that can lead to quick results in large companies and make the responsible team heroes. We explained why dynamic pricing is so important for large companies in detail.
Dynamic Pricing on E-commerce Platform with Deep Reinforcement Learning
These are some of the variables used to make pricing decisions:. After knowing all of this, extensive machine training is required to build a successful dynamic pricing model. Because of the complexity of dynamic pricing, different modules are sometimes used for different product categories and market response to manage complexity.
This module is for new products or long-tail products with little or no historical data.Price setting is one of the most important problems in retail because any price setting error directly results in lost profit.
However, traditional price management methods almost never achieve optimal pricing because they are designed for traditional environments, where the frequency of price changes is inherently limited e.
Dynamic pricing algorithms help to increase the quality of pricing decisions in e-commerce environments by leveraging the ability to change prices frequently and collect the feedback data in real time. These capabilities enable a company to respond to demand changes more efficiently, reduce forecasting errors, and automate price management for catalogs with hundreds of millions of items.
This article is a deep dive into dynamic pricing algorithms that use reinforcement learning and Bayesian inference ideas, and were tested at scale by companies like Walmart and Groupon. We focus on the engineering aspects through code snippets and numerical examples; the theoretical details can be found in the referenced articles. Traditional price optimization requires knowing or estimating the dependency between the price and demand.
This basic model can be further extended to incorporate item costs, cross-item demand cannibalization, competitor prices, promotions, inventory constraints and many other factors. Since the price-demand relationship changes over time, the traditional process typically re-estimates the demand function on a regular basis.
This leads to some sort of dynamic pricing algorithm that can be summarized as follows:. The fundamental limitation of this approach is that it passively learns the demand function without actively exploring the dependency between the price and demand. This may or may not be a problem depending on how dynamic the environment is:. The second case represents a classical exploration-exploitation problem: in a dynamic environment, it is important to minimize the time spent on testing different price levels and collecting the corresponding demand points to accurately estimate the demand curve, and maximize the time used to sell at the optimal price calculated based on the estimate.
Consequently, we want to design a solution that optimizes this trade-off, and also supports constraints that are common in real-life environments. More specifically, let's focus on the following design goals:. In the remainder of this article, we discuss several techniques that help to achieve the above design goals, starting with the simplest ones and gradually increasing the complexity of the scenarios.
This scenario is often a valid approximation of flash sales or time-limited deals. For instance, a variant of the algorithm described below was tested at Groupon with very positive results.
In an extreme case, only one price change is allowed — a seller starts with an initial price guess, collects the demand data during the first period of time explorationcomputes the optimized price, and sells at this new price during the second time period that ends with the end of the product life cycle exploitation.
It can be shown that in these settings, the optimal durations of the price intervals have to be exponentially increasing, so that a seller starts with short intervals to explore and learn, and gradually increases the intervals until the final price is set for the last and the longest interval, which is focused purely on exploitation:. This layout is illustrated in the figure below:.
Next, we need to specify how the prices are generated for each time interval. One simple but flexible approach is to generate a set of parametric demand functions hypotheses in advance, pick the hypothesis that most closely corresponds to the observed demand at the end of each time interval, and optimize the price for the next interval based on this hypothesis.
Next, let's implement the above algorithm and run a simulation. We use a linear demand model to generate the hypotheses and it is a reasonable choice for many practical applications as wellbut any other parametric demand model, such as the constant-elasticity model, can also be used.
We use this code to generate a sample set of demand functions and the corresponding optimal prices:. For the runtime portion of the algorithm, we generate the price interval schedule in advance, and use it to determine whether or not we need to generate a new price at every time step as we mentioned earlier, the schedule depends on the properties of the demand distribution, which is unknown to the seller, so the fixed schedule is a heuristic approximation :.
The execution of this algorithm is illustrated in the animation below. The top chart shows the true demand function as the dotted line, the realized demands at each time step as red crosses sampled from the true demand function with additive noiseand the black lines as the selected hypotheses.
The bottom plot shows the price and demand for every time step, with the price intervals highlighted with different bar colors. The algorithm described in the previous section is a simple yet efficient solution for settings where the demand function can be assumed to be stationary. In more dynamic settings, we need to use more generic tools that can continuously explore the environment, while also balancing the exploration-exploitation trade-off. Fortunately, reinforcement learning theory offers a wide range of methods designed specifically for this problem.
In this section, we will discuss a very flexible framework for dynamic pricing that uses reinforcement learning ideas and can be customized to support an extensive range of use cases and constraints. Let's start with an observation that the approach used in the previous section can be improved in the following two areas:.Have you ordered an Uber during this rainy season and wondered why the prices fluctuate from time to time?
The reason for this is a pricing strategy known as dynamic pricingan approach to pricing goods based on real-time changes in the market such as demand for certain goods and services, time of purchase, change in certain conditions etc.
The keyword here being real-time. But where does Machine Learning come in when talking about dynamic pricing? Throughout this rainy season, a lot of businesses are adjusting their prices from time to time, from the popular Uber to the Matatus we use on the daily and I could bet that gumboots prices have shot up. These changes in prices especially how Uber does it in real-time got me curious as to how data and Machine Learning is being used by businesses to set flexible prices for products and services.
There are a tone of other areas where dynamic pricing can be utilized. Like in Gaming applications, online retail shops, banking, it could also be used to reduce customer churn whereby when your application starts identifying a customer as being likely to churn, it reduces the price of certain products to keep the customer interested.
For many years, retailers have been using what is known as traditional, static, or fixed pricing where a fixed price point is determined and maintained for an extended amount of time. This method becomes difficult when a new product is being introduced into the market and are unable to accurately predict the effect of the changing demand into the price.
Retailers may also set the price in accordance to their desired target market. Recently, Ebay a giant e-commerce corporation, announced in Decemberthat it will acquire a Canadian data analysis firm Terapeak which is good at predicting supply, demand and pricing products so as to give their sellers price guidance and comparisons. You can read more about this here. Data Mining, Machine Learning and Statistical Methods can be useful in predicting the purchase behaviour of an online customer by selecting an appropriate price range for them based on dynamic pricing.
By obtaining a solution for price prediction for products and services, it will be easier for sellers to sell and enlarge the number of goods or services being sold as well as increase the shopping community. With the flexible prices, retailers can bring in higher profits for each sale made. There are lots of other advantages of dynamic pricing especially in e-commerce businesses. The purpose of this challenge is to build a system that offers pricing suggestions to their sellers.
The basic Idea here is to find the best price by analyzing product characteristics. I will briefly take you through my thought process for building such a model using the various submissions made on the Mercari Kaggle competition. Disclaimer: This is just an overview of how Machine Learning can be used to suggest product prices using the Mercari competiton as an example.
You can find the various codes to build such a system on the Kaggle kernel and discussion page of the competition. The following diagram shows the framework for building such a model or rather any machine learning model. The following table shows the features provided by Mercari. This will involve handling missing values in the data as well as some text-processing on the item-description.
A lot of knowledge on Natural Language Processing will be needed here to extract meaningful insights from the item description, as well as cleaning the text data.
Data Exploration is done to get meaningful information from the data. In this case we could explore the data to find out things like:. This is the process of using knowledge of the data to create features that will make your machine learning algorithm work.
This is a fundamental step in any machine learning problem. For this specific competition it was seen that XGBoost also known as extreme gradient boosting was a relatively accurate model to use for this competition as well as RMSLE Root Mean Squared Logarithmic Error as the statistical performance indicator.
Suggesting product prices to online retailers is just one of the many ways machine learning can be used in Dynamic Pricing. Companies such as Airbnb are using data to build algorithms that take into account things like location, type of property, duration etc to help people set their prices for their location.
Machine Learning can also be used to predict the purchase behavior of online customers by selecting an appropriate price range based on dynamic pricing. Dynamic pricing is one of the many applications of Machine Learning that is rapidly growing. Can you think of other areas that can utilize dynamic pricing? Let me know in the comments. Walters, Troy.
Joseph Pisani HOw much for that tequlla shot?In continuation to my previous blogwhich discussed on the different use-cases of machine learning algorithms in retail industry, this blog highlights some of the recent advanced technological concepts like role of IoT, Federated learning and Reinforcement learning in the context of retail industry.
This blog is structured as follows:. In retail sector, use of IoT-enabled devices have played considerable role in controlling and streamlining supply chains, capturing real-time metrics to track product availability, sales and deciding the best placement locations for different products. However a few common problems encountered while collecting and analyzing IoT data from disparate sources and ways to handle them are :. Firstly, the sensor readings from retail stores and warehouses may be intermittently absent or randomly missing at consecutive timestamps, or lost at a certain time-stamp for an entire store.
Missing data makes the traditional regression-based methods or non-negative matrix decomposition method useless due to one or many columns and rows are missing at the same time. Secondly, those sensor data generated by sensors deployed in different locations e. Even matrix completion methods to interpolate the missing is limited to capture the one-dimensional spatial similarity. Spatio-temporal multiview-based learning ST-MVL method is used to collectively fill missing readings in a collection of geo-sensory time series data, considering :.
ST-MVL integrates the advantages of global viewsi. It can handle the block missing problem, combining the four views in a multi-view learning framework.
Though ST-MVL achieves satisfactory performance in terms of filling missing geo-tagged sensor readingsit ensembles five different models and each model requires to fine-tune several parameters, which is labor intensive. Moreover, ST-MVL is still limited to capture one dimensional spatio and temporal information and fail to model high dimension spatial features e.
NN-based heuristic searching methods have evolved to map the sensors with irregular geo-locations into a matrix. The process occurs with iteratively searching the spatially nearest neighbor for each sensor. It only requires to tune one key parameter, without requiring non-missing training data. It can accurately model spatial and temporal dependencies, periodic patterns among sensors to enable a high performance model on missing sensor data recovery.
This method is primarily designed for recovering noisy images or videos naturally can be seen as a tensor. At first the NN-based heuristic searching method is used to transform the irregularly deployed sensors into an array, with adjacent sensor data being placed closed to each other.
Many sensors are irregularly deployed throughout the city generate huge amount of time series data with two dimensions — time and spatial dimensions.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. Given that in these days it is very easy for a customer to compare prices thanks to online catalogs, specialized search tools or collaborative platforms, retailers must pay close attention to several parameters when setting prices.
Factors such as competition, market positioning, production costs, and distribution costs, play a key role for retailers in order to make the right move. ML can be of great help in this case and have an enormous impact on KPIs.
Its power lies in the fact that the developed algorithms can learn patterns from datainstead of being explicitly programmed. ML models can continuously integrate new information and detect emerging trends or a new demands. The use of ML is a very attractive approach for retailers.
Instead of using, for example, aggressive general markdowns which is often a bad strategythey can benefit from predictive models that allow them to determine the best price for each product or service. Briefly, price optimization uses data analysis techniques to pursue two main objectives :.
Understanding how customers will react to different pricing strategies for products and services. Pricing systems have evolved since the early s until now, from applying very simple strategies, such as a standard markup to base cost, to being capable of predicting the demand of products or services and finding the best price to achieve the set KPI.
Price optimization techniques can help retailers evaluate the potential impact of sales promotions or estimate the right price for each product if they want to sell it in a certain period of time.
Current state-of-the-art techniques in price optimization allow retailers to consider factors such as :.
It is important to differentiate price optimization from dynamic pricinggiven that these terms are sometimes used as synonyms.
The main difference is that dynamic pricing is a particular pricing strategywhile price optimization can use any kind of pricing strategy to reach its goals. Despite having many advantages and being quite used, dynamic pricing has some disadvantages when used in an extreme way. Simply put, using a dynamic pricing strategyretailers can dynamically alter the prices of their products based on current market demand.
In contrast, price optimization techniques consider many more factors to suggest a price or a price range for different scenarios e. We all know and somehow accept because it seems reasonable, that the price of a hotel room or a plane ticket varies according to the season, the day of the week or the anticipation with which we booked. However, when prices change too fast — sometimes in the course of a few hours — some customers might have the feeling that prices are unfair or that the company is practicing price gouging.
Dynamic pricing is, therefore, a strategy to be used with caution. The pricing strategies used in the retail world have some peculiarities. For example, retailers can determine the prices of their items by accepting the price suggested by the manufacturer commonly known as MSRP. This is particularly true in the case of mainstream products.