Managing Director & Partner
San Francisco - Bay Area
Imaginative use cases, rapid technological progress, and boardroom drama have dominated the headlines about generative artificial intelligence (GenAI) since the launch of ChatGPT in November 2022. Largely absent from these conversations, however, is a discussion about how software companies and GenAI application developers should set the right pricing strategy and pricing models, two critical elements that determine the long-term trajectory of any new technology.
The strategic pricing decisions that companies make today will have far-reaching effects that will determine how quickly the adoption of GenAI accelerates, who benefits from it, how much money organizations can reinvest into improvements and competitive advantages, and even the future of human-machine interaction. The less consideration companies devote to pricing strategy and pricing models, the greater the risk that they will artificially limit the potential of their solutions by discouraging customers from experimenting. This will not only limit the reach and network effects that would make those solutions more valuable but will also leave open the door for new entrants with better pricing options.
The pressure to launch a solution compelled many innovators to pick a model and a number quickly for their pricing. As a result, even the most experimental companies have tended either to forgo monetizing their GenAI offers for the time being or have defaulted to one of two simple, familiar pricing models. The first is a token-based pricing model that charges users for consumption and is loosely aligned with the computing power needed. The second is a subscription model that charges a uniform fee per user per month but raises concerns that the underlying costs of computing power will make the offerings unprofitable for the vendor.
We recommend companies take a more strategic approach to GenAI pricing that starts with a thorough understanding of the three information sources for any major pricing decisions: customer value created, the competitive landscape, and the economics of the product or service. An integrated view of those sources gives firms the foundation to choose the pricing model that best reflects their short- and long-term objectives and a pricing architecture that captures and shares value in line with their broader business strategy.
Choosing the most expedient pricing model ignores the reality that GenAI applications are fundamentally different from many other technologies, thanks to their sheer potential for value creation and their evolving ecosystem. Those differences underscore the need for the vendor to think through the following three questions carefully.
How do you create and share customer value? The exponential growth in the adoption of ChatGPT and other GenAI applications shows that it is currently impossible to define the upper limits on the technology’s potential value. While direct value to customers can vary widely, there are six key value drivers that we have observed in the evolving GenAI ecosystem (see Exhibit 1):
Model scale and model accuracy are common and well-known technical indicators of value, while the other value drivers are less technical and should be more visible to customers. A company can feed proprietary data into a GenAI-powered application to improve accuracy and create unique outcomes. For example, Bloomberg reportedly improved the accuracy of its GPT model for finance-specific tasks by up to a factor of four when it incorporated proprietary content.
The major differences between the foundation models, the GenAI-powered applications, and the automation workflows are the range of data that trains the models, the tasks that each application can accomplish, and the degree of human involvement. The foundation models and GenAI-powered applications either augment people’s jobs or free them up for higher-value tasks, while the automation workflows require little or no human supervision.
How differentiated are you within your competitive landscape? The competitive landscape for GenAI is rapidly evolving, too. (See Exhibit 2.) Foundation models, including large language models (LLMs), first garnered the most attention. A few big competitors—early movers like Open AI and hyperscalers like Google (Bard, Gemini), Meta (Llama), and Anthropic (Claude)—will determine the landscape for those large-scale, general-purpose models, in part because they require billions of dollars to train, refine, and operate.
As new and existing technology players enter the market with their own GenAI offers, the landscape is becoming more fragmented in terms of both smaller models and more specialized applications and workflows. Smaller providers have an opportunity to differentiate themselves, provided they either have highly specialized capabilities or can segment their offerings to serve a diverse and fragmented customer base. Mistral AI’s 8x7B model, released in December 2023, beats GPT-3.5 on most benchmarks, although it is much smaller than OpenAI’s models.
How will your costs scale and evolve? The economics of GenAI solutions remain uncertain as the costs continue to evolve. By contrast with software and software as a service (SaaS) applications, whose marginal costs can be very low, the underlying cost to serve of GenAI models can be quite high, depending on the use case and pricing model. By some estimates, a high share of the revenue that Open AI has recently generated has gone toward covering computing costs driven up by a shortage of the GPUs needed to support the models. (See Exhibit 3.)
If a company charges a flat rate for access to an application and the costs scale primarily with the number of queries, then the distribution and cost scaling of queries per user is a major driver of profitability. A large share of heavy users can therefore make the model unprofitable in the near term. Strategic pricing decisions, however, not only require an understanding of current unit costs but also how those costs will scale and change over time. As Exhibit 3 shows, the costs per token for various models have declined. Those expenses are plotted on a logarithmic scale, which means that costs for lower performance models are exponentially lower, both in general and over time. There are multiple forces pushing unit costs down, including Moore’s Law, Huang’s Law, improvements in model training, and investments in accelerators. At the same time, new capabilities tend to result in new use cases that may consume more and higher-cost resources. A deep understanding of the tensions across these forces is essential to the economic lens when making pricing decisions.
There is no one-size-fits-all pricing strategy for GenAI. A model developer facing these economics will instead have several options. They could raise the price per user, keep the current price if they believe costs will decline significantly, redesign to offer a lower-cost model that meets a target price point, or adjust the price in a way that incentivizes customers to align their behavior with costs. Which option is right will depend on the market context (including the customer value and competitive landscape) and the vendor’s business strategy.
The pricing model shapes how a company monetizes and shares the value it creates. The most important factors in setting a model for a GenAI application are the pricing basis (the unit and timing of the price) and the offer architecture (how to package the innovations).
Two predominant pricing bases have emerged for GenAI so far: consumption, which is usually measured per thousand tokens used, and subscription, which is usually measured per user per month. GitHub, for example, prices its copilot at $10 to $39 per user per month. We expect outcomes to emerge as a pricing basis as GenAI models become more specific and more integrated into businesses.
Companies’ choice of pricing model can also reveal information about the fate of their team members. If most successful GenAI applications are priced per user on a subscription basis, they are likely to enhance human performance. If the predominant models are based on consumption per task or per outcome, then the applications are more likely to replace humans or diminish their roles.
Companies can choose their pricing model strategically by integrating key insights from the three information sources (customer value, competitive landscape, and economics) into three parameters: the choice of data source to train the GenAI model, the operating environment, and the perceived value for customers. (See Exhibit 4.) Even then, the choice of model is not prescriptive but rather a conscious and careful decision. Companies must weigh several factors as they choose among the three models: consumption based, subscription based, and outcome based.
After the company has decided on the pricing basis (the unit and timing of price), it can move to the second major decision for its pricing model: the offer architecture, which is how it will package its innovations. There are three offer architectures to consider for GenAI solutions:
The transformative potential of GenAI is so vast that it requires a unique approach to pricing strategies. The market is so dynamic and the solutions are so differentiated that each vendor needs to focus on developing the pricing strategies that work best for their own needs. In this constantly evolving space, we have found three no-regret moves:
Companies will be better equipped to navigate this challenging market when they have a clearer understanding of the pricing landscape and how to set a smart pricing strategy that does justice to the transformative power of their GenAI solutions.
ABOUT BOSTON CONSULTING GROUP
Boston Consulting Group partners with leaders in business and society to tackle their most important challenges and capture their greatest opportunities. BCG was the pioneer in business strategy when it was founded in 1963. Today, we work closely with clients to embrace a transformational approach aimed at benefiting all stakeholders—empowering organizations to grow, build sustainable competitive advantage, and drive positive societal impact.
Our diverse, global teams bring deep industry and functional expertise and a range of perspectives that question the status quo and spark change. BCG delivers solutions through leading-edge management consulting, technology and design, and corporate and digital ventures. We work in a uniquely collaborative model across the firm and throughout all levels of the client organization, fueled by the goal of helping our clients thrive and enabling them to make the world a better place.
© Boston Consulting Group 2024. All rights reserved.
For information or permission to reprint, please contact BCG at email@example.com. To find the latest BCG content and register to receive e-alerts on this topic or others, please visit bcg.com. Follow Boston Consulting Group on Facebook and X (formerly Twitter).