As artificial intelligence (AI) technology develops rapidly, various AI services are being released on the market. However, determining the price of AI services is not easy. Unlike traditional software or services, AI requires massive computing resources. Therefore, designing an appropriate pricing model is becoming a very important factor in securing profitability for AI companies.
Recently, my favorite newsletter, Unsexy Business, covered a service called Release, which I personally use. I read the interview with Youngwook Kim on Facebook about sales, which was fun to read.
In some ways, it is an important story. So, is it profitable? The question is. Usually, in the IT industry, when designing business development and business models, the axes that are considered are Unit, Tier, and Term. This is not much different for SaaS, PaaS, or IaaS. The problem is the difference in how to set the price.
Each has its own formula as shown below.
Pricing Unit
Requests: Billing based on the number of requests to your AI model. However, each request may consume different computing resources, making consistent pricing difficult.
Tokens: Divide text into small units (tokens) and charge based on the amount of processing. Differential pricing can be applied considering that the number of input and output tokens may be different.
Successes: Billing based on the success of the service provided to the customer. An approach that focuses on customer value regardless of compute usage.
Physical Resource Usage: Billing based on hardware resources (CPU, GPU, etc.) used. Primarily used by AI infrastructure providers.
Tier Pricing
Model Tiering: Apply differential pricing based on the performance and cost of the AI model used.
Subscription Tiering: A method of combining a monthly/annual subscription fee that includes a base usage amount with charges for additional usage.
Terms of Transaction
Prepayment: No receivable risk as customers purchase credits in advance and deduct them based on usage.
Threshold Charging: A method of automatic payment when usage reaches a certain threshold.
Minimum Usage Commitment: Ensure stable revenue by guaranteeing a minimum amount of usage or amount.
So now, how do we actually implement this into a business model? Everyone asks, “What does business development mean?” But business development is exactly implementing this. Just like developers write programs and implement functions.
How to implement (Implementation)
Pay-as-you-go: This is a method where you pay only for what you use. It is divided into the following types depending on whether it is based on a specific point in time or on the amount of usage.
Snapshot Recording: A method of recording usage by aggregating it at regular time intervals. The amount of data is small, but it is difficult to provide detailed information.
Event Recording: A method of recording usage for each individual request. Accurate, but expensive when processing large volumes.
Unlimited method: This is the method used by many SaaS types. It allows use regardless of the number of requests, but recently, with the emergence of costly services such as artificial intelligence, methods are also being introduced that impose restrictions within these.
In addition to the above pricing methods, you can consider new approaches that suit the characteristics of your AI service.
Outcome-based Model: A method of charging based on the business outcomes achieved by customers through AI services.
Value-based pricing: An approach to determining prices by evaluating the actual value a service provides to the customer.
These models can also help build long-term relationships with customers by focusing more on customer value rather than simple usage.
In order to maximize the value of AI services while securing sustainable profits, it is essential to establish a systematic pricing strategy. Each company should closely analyze the characteristics of its services, deeply understand customer needs, and design the optimal pricing model. In addition, efforts should be made to continuously improve the pricing policy in line with the rapidly changing market environment.
Personal thoughts
The cost will continue to decrease, but users will also increase their demands, so I personally don't think this gap will be easily narrowed. For example, mobile chipsets released in 2024 are cheaper than low-power laptop chipsets released in the early 2010s. However, the smartphone market is increasingly moving toward the high end. This may be because consumers value 'ownership' more than whether they use all the functions.
Afterwards, when the era of everyone having an AI model arrives, not everyone will need to have a SOTA-level model, but everyone will want to have one. So, while the cost reduction now is meaningful for the manufacturers, it doesn’t seem to be a great inspiration for businesses or consumers.
This is also the reason why many AI startups overseas are having difficulties with their current business models and are collapsing quickly. It is the same reason why they are going around and around to B2B. However, in the case of B2B, the initial hurdles are high, so even this is not easy. And it is not easy to optimize costs from the perspective of providing B2B services. To do this, you need to introduce it company-wide and hedge usage, but this is not easy either. (Just look at the sessions recentlyreleased bySalesforce , etc.)
How AI will make money in the future is a market that has not yet been answered. I am currently organizing the earnings calls of Microsoft, Google, etc., and I plan to introduce them together when Apple's earnings call (May 2nd) is announced. First, MS and Google announced that their sales increased as they sold AI solutions. However, I think that we need cases from users, not suppliers, now.
Subscribe to 'haebom'
📚 Welcome to Haebom's archives. --- I post articles related to IT 💻, economy 💰, and humanities 🎭. If you are curious about my thoughts, perspectives or interests, please subscribe. Would you like to be notified when new articles are posted? 🔔 Yes, that means subscribe. haebom@kakao.com
