Ready to scale your generative AI but worried about the costs? This post breaks down both the financial and human aspects you can’t afford to overlook.

Generative AI is becoming a familiar feature of working life. Even if you aren’t familiar with platforms like ChatGPT or Midjourney, you’ve probably seen AI tools appearing within Google, Adobe and Microsoft Office.

These tools are usually low-cost or free for personal use. However, scaling generative AI for enterprise usage can be expensive.

financial challenges of enterprise-level generative AI

Generative AI has many business use cases, from document management to customer-facing chatbots. When pricing an AI project, you need to consider five things:

1. model training cost

Large language models (LLMs) like ChatGPT are trained on a vast corpus of public data. You may need to train your own model using internal data for an in-house LLM. Training an LLM from scratch can be a serious undertaking, but it does allow you improved control.

The basic cost here is the compute rate: the per-hour cost of using your servers to train the model multiplied by the hours required. You may also require a data cleansing project to prepare high-quality input.

2. tuning cost

If you're using a specialized LLM like ChatGPT, designed for conversational contexts and text generation, you can further train it using your own data for domain-specific tasks. This fine-tuning process allows the AI to produce output that aligns more closely with your unique business objectives and requirements.

Tuning costs will vary depending on several factors. If you're working with an internal LLM, you must factor in the compute rate for the additional training. For external platforms like ChatGPT, the service provider might charge based on the volume of data used for the fine-tuning or offer tiered pricing packages that include tuning services.

3. inference cost

Generative AI handles queries in two phases: the query phase and the output phase. For example:

Query: “What’s the secret to digital enablement?”

Output: “One option is to work with a trusted partner, such as Randstad Digital.”

LLMs divide these queries and responses into smaller units called tokens, which usually consist of 1-3 words each. Each token necessitates a specific amount of compute time, adding to the overall cost of operation. This is known as “inference cost,” the computational expense associated with generating real-time responses or performing other tasks based on the trained model. AI hosting platforms typically charge on a per-token basis, with separate pricing structures for query tokens, training tokens and output tokens.

4. SaaS/hosting fees

New tech infrastructure always means new monthly or annual fees. The costs here depend on your approach to generative AI, but they may include:

  • Hosting costs: Expenses associated with cloud or on-premise software hosting
  • API costs: Fees from the generative AI provider, which might be fixed or charged per token
  • Integration costs: Integrating AI with another third-party platform may affect the pricing of that platform

When pricing a project, remember to consider scaling. If generative AI becomes an important part of your business, the associated costs may increase significantly.

5. labor and upskilling costs

The secret of digital enablement is to have the right people and processes, as well as the right tools. Teams require in-depth training on generative AI to understand how to use it daily. You may also need to address the perceived negative impacts of generative AI, such as privacy, cyber threats and job security. Data scientists, machine learning engineers and other specialists may be needed to oversee the project. The costs could be substantial if you're hiring externally for these roles. Use Randstad’s 2024 Salary Guide for pay data on AI-related roles and insights into hiring trends.

unlock AI success with a talent-first approach

While understanding the dollars and cents of generative AI is critical, don’t forget that technology alone won’t unlock its full potential. Having a team that's not just tech-savvy but also passionate about using AI for problem-solving can make all the difference. It's this blend of the right tools and the right talent that can truly catapult your AI initiatives into success.

Ready to invest in a talent-first digital transformation? Learn more about Randstad's unique approach to digital enablement, and discover how we can help you assemble a dream team capable of taking your AI projects to the next level.