Danillo Pereira and Betsy Romeri
January 20, 2020 Download PDF
In the time of COVID-19, businesses have to be more dynamic than ever in their pricing, responding to greater short-term fluctuations in demand and supply, while keeping an eye on a longer term recovery.
The topic of dynamic pricing has been discussed often in technology and business circles in recent years, mainly due to the availability of massive amounts of data, increased computing power, rapid growth of AI-driven business applications, and technologies like electronic price tags that make it much easier for companies to adjust prices quickly and effectively. With AI, revenue improvements are possible across many verticals and business functions, including retail, eCommerce, CPG, supply chain management, and others. These improvements stem from the ability of AI-driven solutions to ingest large amounts of data and different types of data simultaneously, find hidden relationships, and predict and prescribe optimal solutions.
Businesses have always tried to optimize their prices for maximum revenue. Before AI and machine learning (ML) became more mainstream, the term “price optimization” was used. Spreadsheets, statistics, and rules-based systems were helpful. The latter operates using a knowledge base of rules/facts about a problem based on domain expertise in the form of if-then statements. The drawback to this approach is that it can only respond to the one environment in which the rules work, so it lacks the agility of AI-enhanced price optimization (dynamic pricing), which adjusts as markets shift and customer demand changes. Furthermore, dynamic pricing can be applied 24/7 and at a more granular level than previously used pricing methods. It can improve company profits SKU by SKU or container by container. How? Its models are built to continually run “what if” scenarios with the goal of selecting the one that maximizes revenue in any given environment, moment by moment. The models can be built to recommend different prices for the same product at different times and store locations and as demand fluctuates. What companies want to avoid is underpricing or overpricing, and dynamic pricing focuses on the sweet spot—the right price that maximizes revenue and profit.
Simulation-Based Price Optimization
Enabling simulation-constrained optimization to drive targeted price bands
Dynamic Data and AI
“Dynamic” refers to the real world and how fast it changes moment to moment. With AI and graph database technologies, context and behavior (behavioral intelligence) of the current marketplace and customer demand can be extracted from data. There are many factors, both internal and external to a business, that influence predictions on the best price for maximizing revenue. When these factors can be represented in data, this is one area where you can differentiate your solution from your competitors. The more data that can be used to define the context, including real-time inventory and demand indicators, the better dynamic pricing models can test what-if scenarios and predict your optimal price for that moment under those conditions. Companies will vary in the internal data that they collect and the accessibility of that data. In general, the internal data available to optimize pricing for one company may differ significantly from another company, industry to industry, or even business model to business model. Developing or purchasing a dynamic pricing AI solution will require business stakeholders to work with data scientists to identify the valuable internal data and make it accessible for testing.
The ability to incorporate massive amounts of relevant external data into analytic models is one of the most powerful attributes of AI-enablement. External data can include anything from geolocation to local events and social sentiment. Because so much information has been digitized, much of the external data that influences our business operations every day, such as weather, traffic, and much more, is readily available. A company’s internal data and these external data sources provide a previously inconceivable amount of data intelligence related to each product or service to be priced. Amazon represents one of the most extreme examples of real-time, dynamic pricing. “The company changes product prices about 2.5 million times daily, meaning the cost of an average product shifts every 10 minutes”(1), whereas Walmart uses dynamic pricing to change product prices over 50,000 times per month (2).
'A big challenge is to determine what data will provide the best pricing insights for a particular business, product, or service while following all of the data privacy laws now in place.'
Developing a data model for a dynamic pricing solution can be challenging for any company, even the Amazons of the world. That is because it is not just about more data, but about valuable data or “intelligent data.” A big challenge is to determine what data will provide the best pricing insights for a particular business, product, or service while following all of the data privacy laws now in place. As Ted Gaubert, CTO at Noodle.ai, notes, “A big part of the AI pricing game is having AI learn everything about what is happening in the market. The goal is to have better information than competitors in order to make better decisions" (3).
Specifically, one of the most important pieces of external data intelligence is competitor pricing data. It used to be that undercutting your competitor down the street by 50 cents could convert a sale. But, today, consumers have many more choices and ways in which to purchase products and services. Price comparisons are just a click away, shifting the competition from hyperlocal to global. Consumers, vendors, and suppliers see your prices and your competitors’ prices in real time. You have to be confident your price is the right one to maximize your profits as well as your market share—also in real time.
Collecting data on competitor prices can be relatively straightforward for some companies using web-crawling techniques; however, not all competitors have online pricing visible to the world. In fact, some business models exist where real-time internal pricing data are not available for analysis. In such cases, data scientists have to get creative to identify data sources that are strong indicators of what is going on in the marketplace. The data required will vary from business to business and product to product, possibly rendering packaged, one-size-fits-all, dynamic pricing solutions inferior to customized solutions. In addition, the differences among businesses and their data highlight the need for data strategies.
Dynamic Models and AI
Ironically, many AI solutions for dynamic pricing that are available today are not themselves dynamic. They have been built to handle only certain types of data and not others. Also, many solutions, whether purchased apps or custom-built solutions, will not automatically recalibrate the AI models as the business conditions change. Over time, AI models can drift or lose accuracy because they are not changing with the internal and external conditions of the company. To address analytic drift, some vendors have automated the continual learning and recalibration of AI models, while others provide monthly or quarterly model updates. Automating model monitoring and recalibration is a beneficial feature found in the more advanced AI platforms on the market.
When you read about the world of ML/AI solutions, you are left with the understanding that you have five paths to acquiring the solution you seek:
• Purchase an AI solution that is proprietary and sold through a vendor.
• Subscribe to a plug-and-play data science platform and build your own solutions (citizen data scientist or data scientist required).
• Use open-source algorithms and datasets and build your own solutions (citizen data scientist or data scientist required).
• Build a data science team to customize your solutions.
• Partner with an AI-as-a-Service company that can help you identify where AI can enhance business decisions, identify use cases, and then build these capabilities for you.
The pros and cons of each path are the subject of another paper, but for those seeking a dynamic pricing solution, here are the challenges:
• If your competitors purchase the same packaged solution as you, then you need to beat them in the data game.
• If your competitors subscribe to the same plug-and-play platform as you, then you have to beat them in your model-building know-how.
• If your competitors rely on open-source components like you, your solutions are completely dependent on the model builder’s ability and experience.
• If your competitors hire a data science team like you, you had better have the smarter team.
• If your competitor partners with an AI-as-a-Service company like you, each of you will receive custom solutions unique to your business.
'How do you price your products and services to maximize revenue, profits, and customer experience while maintaining your business model and goals?'
The point is to have a solution that is the best fit for your business. Each company has its own brand recognition, customer loyalty, and distribution/shipping models, for example, that can affect pricing decisions—even within the same industry segment. So, the question is really how do you price your products and services to maximize revenue, profits, and customer experience while maintaining your business model and goals? For example, it might not make sense for your business to continually undercut your competitor prices if your brand recognition or customer loyalty affords you a higher price point than your competitor. Matching pricing strategy to business goals is a complicated calculation that also involves measuring market share impact to price changes, among other things. The best solutions will be arrived at with cooperation among business stakeholders and AI solution experts.
Acquiring Dynamic Pricing Capabilities
A quick Google search for dynamic pricing software will convince you that not all solutions are created equal. In fact, much of what is available is industry-specific and often restricted in terms of data types that can be ingested, integration capabilities, and the output they provide (pie charts, bar graphs, recommended actions, and more). Whether selecting a packaged solution or building a customer solution, it will be necessary to do your due diligence and prioritize your business parameters and goals to get the best fit.
To acquire and succeed with dynamic pricing solutions suited for your company, we suggest:
• Doing the due diligence on the tools available on the market. A successful AI initiative for any dynamic pricing solution must be built on your goals and a thorough understanding of what is possible with AI for your use case.
• Understanding the monetary and time investments of initial and ongoing costs. Data acquisition, integration, deployment, and user training and adoption, whether for vendor acquired solutions or custom-built solutions, can have hidden costs.
Operationalizing dynamic pricing in a step-by-step approach to (a) reduce risk and costs and (b) improve time to value and adoption. Seek solutions that will allow you to start small and prove value before going enterprise-wide.
Many companies fail to fully appreciate that no matter whether you purchase packaged software or build your own solutions, all of the steps from identifying the most valuable use cases to training your workforce are required to succeed.
Explore Analytics2Go to learn how an AI-as-a-Service partner can help you. We are a team of data scientists, IT professionals, and business consultants with decades of experience. We have developed a quick and inexpensive methodology of working with clients to improve revenue and process efficiencies with AI solutions. We focus on B2B companies in the areas of supply chain management and dynamic pricing across industries including manufacturing, CPG, and retail.
Establishing Your Pricing Model
When considering a pricing model, each business will have its own parameters and guardrails that are consistent with its goals. The individual business parameters will vary within an industry and across industries. For example, cost-based pricing models adjust prices according to business costs and keep profits at a pre-determined level. Competitor-based models take into account competitor pricing decisions. Demand-based pricing models will optimize prices based on supply and demand pressures (4), Companies also should consider whether they sell products and services that are elastic or inelastic. Inelastic products are those where the demand stays the same even when the price goes up. (Examples, electricity or gas). If the demand for your product fluctuates as the price changes, it is elastic because customers can find an alternative product or the same product from a different seller at a better price (4). The models that you build should reflect the type of products you sell and the market in which you operate.
Dynamic pricing solutions should be used when the inventory is perishable or there is a capacity constraint (for hotel rooms, airline seats, and grocery stores, for example) and you have a varied willingness to pay among your customer segments. Another instance where dynamic pricing is effective is with seasonal sales—when customers are willing to pay more for certain items since they are purchased less frequently.
A good example of dynamic pricing where the goal is to balance the supply side and the demand side is Uber. The price goes up as the demand goes up. This assures that consumers can get cars when they need a ride, and drivers (and the company) can get higher fares. It is important to remember that with dynamic pricing, businesses must not allow their analytic models to set prices without pricing limits—the proverbial guardrails. They may lose business if customer sentiment is affected, whether from a sense of unfairness that the price is too high or that the price is insensitive to particular groups, etc. Some businesses may choose not to use dynamic pricing at all and instead set their prices and appeal to those customers who want the certainty of price and service. Your AI solutions should always be consistent with your business model.
Danillo Pereira, Ph.D., is A2Go’s Director of Data Science. In addition to multiple degrees in the computer sciences and teaching professorships, he has published widely in the field of analytics, machine learning, and AI. His main focus at A2Go is creating modern analytics models that aid business planning and decision-making.
Betsy Romeri, Ph.D., is A2Go’s Director of Communications. She was a published author and college professor in the subjects of biochemistry and molecular biology before joining A2Go to help translate some of the concepts of AI into business language.