Today, we talk to Zeldeen Müller, CEO of AgendaWorx, an online board pack tool utilising AI, about a real-world case study of how much time AI can save companies versus the surprisingly low cost of AI.
Please explain to our readers how AI is priced.
We will use ChatGPT as an example when discussing the pricing structure, as I think most of our readers have heard about this model. Other AI models such as Claude and LLaMA will have different pricing models.
OpenAI’s ChatGPT operates on a usage-based pricing model, where users are charged based on the number of tokens processed. A token is equivalent to a word or a few characters.*
For example, OpenAI’s ChatGPT charges approximately $0.0005 per 1 000 tokens for input, $0.0015 per 1 000 tokens for output and $0.002 per 1 000 tokens for API usage.
Please provide a real-life example of how much this cost will be for a typical process.
AI can be used for many different applications and uses. Today, we will focus on a case study where a small business used AI to create minutes of meetings, which, for them, was a time-consuming, and therefore non-profitable, service provided to clients.
Their current process:
When creating minutes of the board meetings for its clients, its secretary would listen to the two-hour meeting recording and transcribe what happened during the meeting. This was a slow and meticulous process that took 10 to 16 hours for every two-hour meeting, depending on the complexity of the discussions. After completing the minutes, she corrected grammar and typos, and the manager reviewed them for content and grammar. This cycle of editing and reviewing continued, including layout and font adjustments and PDF conversion, until the minutes were ready for distribution.
The true cost for the company was in the time spent by the secretary and manager, which could be better spent on other tasks that would boost profitability.
Understood. So how did AI streamline this process? What was the AI cost?
The AI process involves transcribing the meeting from the video recording with 99% accuracy for South African accents and converting the transcription into professionally written minutes.
A rough estimate of the AI costs per set of minutes is:
Process one:
Getting the transcription of the video recording: R1.37**
Process two:
Converting the transcription to a set of minutes: R0.87. Time-saving is 10 to 16 hours for the secretary and one to three hours for her manager.
This company saved 19 hours of work with one process at a meagre AI cost of only R2.24.
Yes, that is quite a saving. And the company does five sets of minutes for clients every month. That is 100 hours saved per month at a cost of only R11.20. It is a no-brainer.
That sounds great; you mentioned the API costs. Why should companies use an API as a secure platform rather than allow their staff to chat with AI directly?
It ensures better control over data privacy and security. When employees interact directly with AI, they can enable history and allow AI to train on their prompts and responses, potentially exposing sensitive information. On the other hand, when using a secure API, the prompts and responses are kept confidential. They are not used to train other AI models, providing a safer environment for handling sensitive data. This approach ensures that the company’s information remains secure.
