Software buyers are looking for AI capabilities—at the right price. Yet, as AI and generative AI become integral to modern products and services, software producers and suppliers face major challenges in monetizing artificial intelligence effectively without quickly eroding margins.
Technology companies need to understand the demand for AI in order to evaluate the AI capabilities they have and ones they plan on developing in future releases, along with how to market and monetize them. Informed, data-driven decisions are necessary for pricing any AI products and features profitably; the business model also must deliver value to customers. Here’s what to consider.
Prioritize Accurate Pricing to Support Ongoing Success
Accurate pricing of AI requires a comprehensive analysis of buyers’ needs and the producer’s costs. Getting it right, and ensuring profitability, requires more than “a gut call,” as Sam Altman, CEO of OpenAI, made when initially pricing ChatGPT. Perhaps it isn’t a surprise that the company reports losing money on its Pro subscriptions, due to underestimating usage, as Altman stated.
The process shouldn’t be underestimated, as the costs for new AI capabilities are uncertain. There’s a bit of a trial and error approach (or set, measure, and adjust) that’s necessary to get it right. With greater adoption of new AI capabilities, the costs will be better understood and pricing can be further refined to maintain margins and profitability.
With even the biggest players in the field finding it challenging to correctly monetize their AI and GenAI offerings, this much is clear: no matter its size, any software company must have the right processes in place in order to price AI products and services effectively. This is particularly—and increasingly—important at a time when only about ⅓ (36%) of senior executives at global technologies report that price and value are “totally aligned” for their offerings.
Look Beyond Traditional Monetization Models
Software producers quickly realize that traditional monetization models don’t always work for AI. Subscription- and user-based approaches don’t account well for variable costs and often don’t meet the needs of AI buyers. Consider, for example, a customer who is paying $30/month for a software subscription. When AI capability is added by the vendor, the more a customer uses that feature, the more cost the vendor incurs, eroding the margins.
While subscription pricing is great for delivering annual recurring revenue to the producer, a usage-based approach is what can make it sustainable. Extensive reliance on usage-based pricing is more prevalent over the past year, part of an overall trend toward hybrid approaches to monetization of software and IoT devices. As detailed in IDC’s Software Monetization Trends, 2024 report, “CIOs prefer hybrid pricing models,” for GenAI pricing, but “the industry is still measuring the outcome value of GenAI.” In short: knowing how much to charge, when to charge, and what customers will commit to paying, remains a significant challenge.
Usage-based monetization models (sometimes called consumption-based, credit-based tokens, elastic access, metered, or pay-per-use) are defined by usage metrics. For example, metered tokens (prepaid credits) can be exchanged to access technology items at listed cost (per application, feature, capability, etc.) for a period of time. This can be a great option for monetizing AI, as it provides real-time, granular product usage data that supports precise show-back or charge-back to users, supporting cost predictability for buyers.
Outcome-based pricing is also an important consideration for AI. With it, the user doesn’t just pay for access, but for deriving a business benefit out of using an AI capability. Outcome-based pricing typically has a higher unit cost, but lower volume of units, since the user doesn’t pay for trying, but only for succeeding. For example, if the designated outcome is the automatic closure of a support ticket, each closed ticket results in a charge to the user for a certain amount based on successful completion of that process. Identifying the unit price, therefore, is also a challenge.
Take AI Products to Market Thoughtfully
Another major consideration for software producers is how to take the product to market. Common approaches include offering AI as:
- An enrichment model, where the AI capability is added as part of the existing offering, with prices adjusted accordingly.
- An add-on, a new model available alongside an existing offering, perhaps for a different price or with a different monetization model.
- A new standalone offering, with an approach to pricing that needs to be determined.
If, as part of the software supply chain, a software producer incorporates other vendors’ capabilities in its product (such as white-labeling analytics capabilities or reporting), those vendors may already offer AI functionality. If so, it’s necessary to account for the additional expenses owed to those vendors for those capabilities. Failure to do so can erode margins.
Consider Typical Costs Associated with AI
Appropriately pricing an AI offering requires a comprehensive understanding of the costs that go into the AI product or features. A producer’s current cloud expenditure is going to increase. How much?
Begin by gaining clarity into five high-level cost categories:
- AI model training: This, by far, is the most expensive (often prohibitive) cost, running into millions or even billions of dollars. Most organizations will, instead, use an off-the-shelf large language model (LLM). They may also use technologies that allow for augmenting LLMs with their own data, making the model more relevant to their domain and their customers.
- Processing costs: Running models in production requires compute resources. Increased demand—especially for real-time, low-latency applications (like recommendation engines and chatbots)—will drive up processing costs. A general estimate will reveal an expected 20–50% increase on top of current operating cost.
- Storage & data handling: AI models rely on large data sets. In turn, these large data sets increase the need for storage and data processing. Training data needs to be stored, pre-processed, and updated regularly. All of this increases costs for both storage and data transfer. Account for these to cost approximately 10–20% on top of existing cloud costs.
- Specialized AI services: Leveraging managed AI services (for example, from Microsoft Azure AI Services, from Amazon, or from Google) will drive additional premium costs. Again, these typically run another 10–20% on top of existing cloud costs.
- Scaling requirements. Up front, a software producer may not have a clear idea of how much actual use a new service will garner in the market. The service will need to scale rapidly as usage increases. If a product uses AI for personalization or recommendation, every user interaction could potentially trigger AI calculations, leading to higher compute demands. Be sure to have product roadmap assumptions and modeling in place; frequently revise this as adoption of products and services grows.An occasional-use chatbot or small recommendation engine may drive an additional 10–20% cost over current cloud and operation cost; a large scale AI deployment with advanced personalization, deep learning models, and real-time inference requirements may drive costs up by 50–200%.
Prepare to Demonstrate Return on Investment
With an understanding of these expenses in place, how can a software producer demonstrate ROI? How will you provide the C-suite what it wants to model the potential revenue benefits and costs?
Though AI is new and unpredictable, like so many innovations, there are important steps to take to price it effectively and sustainably. One important approach is to leverage statistical experiments to test pricing hypotheses about monetizing AI offerings.
Innovation accounting, which can “quantify the progress of a risky project in an uncertain domain,” helps map the impact and the user journey, create a model, quantify uncertainties, and make decisions about new products and services. As illustrated by innovation coach Tristan Kromer, this can help evaluate variables like user behavior (retention rates, what needs to happen to drive increased customer conversion), what it takes to launch a minimum viable product, and gain clarity into metrics that will help determine accurate pricing for AI.
Additionally, having appropriate entitlement management programs in place can help enforce use rights, guarding against overuse and misuse of products and services. It can also help gather the software usage data and insights required for ongoing decisions about pricing and packaging strategy (including up-sell and cross-sell opportunities), implementation of monetization models that drive recurring revenue and deliver customer value, and that flag churn risk or other operational inefficiencies.
Prepare, Don’t Guess
Any innovation needs to be priced accurately in order to be profitable. That’s true for AI. Appropriate budgeting, flexible monetization options, some level of experimentation, and accurate forecasting are all required for ongoing profitability.
