AI costs. What is the cost of implementing AI?

Artificial intelligence is becoming an integral component of the development strategies of many businesses. The question of “how much does AI cost” is not just about the price but also about investing in the future and the adaptive capabilities of the company. In the article, we will analyze the factors influencing the costs of implementing and operating AI. We will also provide specific examples of its applications in business to help entrepreneurs better understand potential expenses. Read on.

AI costs - table of contents

AI costs. What do they depend on?

The costs associated with implementing AI are diverse and dependent on a variety of factors. To understand which elements have the greatest impact on the final price, we have prepared a list of the most important ones:

implementation scope – organizations that allocate at least 20% of their earnings before deducting interest and taxes (EBIT) to AI adoption are considered leaders in AI utilization. According to the McKinsey Global Survey on AI report, they often invest more in these technologies. Thus, a high AI contribution to the company’s profits can raise implementation costs.
access to specialists – the need for specialized positions, such as data engineers, machine learning specialists, or data scientists, can significantly impact the costs of AI implementation. The availability and cost of these specialists in the job market are key factors in the cost of AI for a company.
allowable operating costs – the choice between custom AI solutions and off-the-shelf software affects costs. Custom solutions can cost from $6,000 to over $300,000. While off-the-shelf software comes at a price of up to $40,000 annually.
the breadth and depth of AI adoption – companies that utilize AI across multiple departments may incur higher costs than those that limit themselves to single applications.
future investment plans – companies planning to increase investments in AI in the coming years must anticipate higher expenditures for the implementation and development of this technology. However, this investment will likely be essential for the growth of firms. As many as two-thirds of respondents in the McKinsey Global Survey on AI expect an increase in AI investments over the next three years.

This list highlights that AI costs are complex and require individual analysis. For example, a company opting for the implementation of a data analysis system must consider both the costs of purchasing the software and hiring specialists capable of operating it.

Model training of AI costs

One of the most common costs associated with implementing artificial intelligence that deters people from investing is the cost of training the AI model. This is a process that requires both expertise and financial resources. Above all, however, to train an AI model, you need to collect enough data and perform data analysis.

So when does training a model make sense? Only when a company can expect significant improvements in efficiency or increased profits through the use of AI. The cost of training a model is one of the aspects that is very difficult to estimate. It depends on its complexity, the model’s application, and the company’s requirements.

An example can be implementing an AI system for personalizing the offer of an online store, where a precisely trained model can significantly increase sales by matching products to individual customer preferences. In such a case, the costs of training the model are an investment that brings tangible benefits.

Another AI implementation that requires model training is the optimization of logistics processes. A properly trained model will reduce transportation costs which over time will lead to increased competitiveness and improved delivery time.

Pricing plans

Subscription is a popular option for businesses looking to leverage advanced technologies without the need for significant upfront investments. Here are some example subscription costs:

AI chatbots – they are most commonly used for automating some of customer service tasks; it’s worth looking into solutions such as Drift (monthly cost from $400 to $1500), TARS ($99 to $499 per month), or Intercom Fin (from $39 to $139 per month).
AI content analysis systems for SEO – they can cost around $150 per month, for example, Contadu (from $79 to $297 per month),
AI coding assistants – the prices of the most popular tool Github Copilot, based on the GPT-4 model, which is also the foundation of the paid version of ChatGPT Plus, start at $10/40 zł per month,
ChatGPT Plus or Perplexity – that’s a cost of around $20 per month per user, a free alternative is Google Bard or Microsoft Bing/Copilot.

Before deciding on an AI tool, entrepreneurs should carefully analyze their needs and capabilities. For instance, a consulting firm might opt for a subscription to a data analysis tool to deliver valuable insights to clients more efficiently.

The AI cost of using popular API

Application Programming Interface, or API AI, are tools that enable the integration of AI functions with existing systems, applications, and services. The cost of using popular APIs is usually calculated based on the number of tokens used and the chosen model.

The fees for the most popular models in the OpenAI API:

GPT-4 Turbo costing $0.01 per 1K tokens for input and $0.03 per 1K tokens for output,
GPT-3.5 Turbo – the cost of the previous model, sufficient for most business applications, is around $0.0005 per 1K tokens for input and $0.0015 per 1K tokens for output.

Source: Martian (https://leaderboard.withmartian.com/)

Businesses can also use open access models, such as mixtral-8x7b or llama2-70b. Operating costs are much lower, while APIs are provided by, among others:

deepinfra (https://deepinfra.com/),
Abacus (https://abacus.ai/llmapi), and
Perplexity (https://www.perplexity.ai/).

But how to use APIs to implement AI in your business? A great example would be integrating an API to generate product descriptions in an online store, which can speed up the process of adding new items and improve the quality of presented information. Or creating a tool that can automatically generate personalized responses to customer emails.

Maintaining an AI team or collaborating with external AI specialists?

Who should handle the implementation of artificial intelligence in your company? If you don’t have a team of specialists or enthusiasts – citizen developers, you are faced with a decision between maintaining an internal AI team and collaborating with external specialists. This decision can have a decisive impact on the costs and effectiveness of AI projects.

Maintaining an AI team involves the costs of hiring expensive and experienced specialists, including programmers and data scientists.

Collaborating with external AI specialists can be cheaper and provide access to specialized skills. However, it may make our solution significantly more expensive to maintain later on, as every change will require calling in specialists for help.

The choice between an internal team and external specialists should be driven not only by cost but also by the company’s strategic goals. For example, a small company may choose to work with external specialists to quickly implement AI solutions without having to build an internal team. And then use one of the less specialized employees to support it later.

Not just money – the environmental AI costs

The environmental costs of AI are an issue that cannot be overlooked in a company’s long-term strategy. Fortunately, most business leaders responding to the McKinsey Global Survey on AI are aware of the many risks associated with generative AI, including:

social risks,
humanitarian risks, and
threats to sustainable development, which may imply environmental costs associated with AI.

Organizations should think about ways to manage the environmental risks associated with AI when implementing it. For example, a company using AI to analyze large datasets should consider the impact of its operations on energy consumption and look for ways to optimize it.