How businesses can choose AI tools that improve productivity
Machine-assisted translation of the archived article. Read the Chinese original.

In the Internet age, CTOs, as a key position in Internet enterprises, have always been the ones whose founders must choose with caution.
By the time of AI, code and product logic were being compromised, the functions and capabilities that had been achieved in the traditional way of writing codes, large models could be dry, no need for product managers to decompose, programmers to write codes, fix bugs and so on! Large-language models have brought about a very simple but almighty interactive era. The IT technology that we now know is out of date, and the various computer languages look back on an ancient technology like a manual machine typewriter.
CAIO (Chief Artificial Intelligence Officer) was created. In the case of MSMEs, CAIO's location and integrity have been greatly liberated and strengthened, and in the future there will be a large number of CAIOs at the heart of the company, especially as first-hand works!

ChatGPT, OpenAI and MyGPTLife
Let's take a look at the A.I.D. scene!
MyGPTLife AI use cases | ||
Use case | Purpose | Material to provide |
AI customer service | Offer customers round-the-clock support and reduce time spent on routine enquiries. | Company and product information, after-sales policies and examples of customer conversations |
AI strategy adviser | Help management with business planning, market analysis, product planning and strategic decisions. | Company information, industry research, competitor information and commercial reference material |
AI sales adviser | Help sales teams plan conversations, develop scripts and analyse performance. | Industry and product information, customer profiles and sales scripts |
Internal company knowledge base | Support staff training, assessments and questions using departmental material. | Company information, industry standards, internal policies and departmental records |
AI marketing assistant | Help prepare marketing copy, campaign plans, slide decks and visual material. | Marketing copy, campaign plans, slide outlines and visual references |
AI work assistant | Help staff prepare reports, meeting summaries, bids, translations and email. | Reports, meeting notes, bid documents, translation references and email |
How can a company choose an appropriate AI service?
First, it should be made clear that AI at this stage is a tool for supplementing and promoting human effectiveness and cannot replace key posts, which, in this case, do not refer to ordinary jobs, some of which can indeed be replaced.
Then there should be value and price as an effective tool. So what would be the price of using AI for an enterprise, because it is a new product, a new scene, and market prices are very volatile, and businesses are afraid of being wronged and cut.
I've got a few chestnuts here, and we all know. "Data is about, don't dig in to the point of the numbers."
The article compares AI spending with broadband. At RMB 1,500–2,000 a year per line including maintenance, it estimates around RMB 4,000 a year for two lines serving a 30-person business. It separately quotes a rough benchmark of RMB 200 per person per year.
2. AI, as a software expense, corporate software expenditure, in the case of our usual office software, Microsoft office software, for example, whose value to the enterprise is incalculable and unquantifiable, and AI, for example. If the value of creation was estimated, would it be for enterprises to spend 20 per cent of their budget on software? Such an algorithm would certainly be ridiculous: in most small and medium-sized Chinese enterprises, more than 2 per cent of the cost of software is an indication that the company is very interested in the software.
Thus, by two analogy, the per capita cost of broadband use is closer to the cost of AI.
Don't use the wrong algorithm to figure out how many employees AI can bring to you, and pay AI according to the value it creates.
The author recommends budgeting for AI as a software expense, based on expected usage.
Historical individual pricing examples for GPT in mainland China
1. ChatGPT - Free version - Personal registration OpenAI, requiring magic and virtual cell phone numbers
ChatGPT Plus cost $20 per month, approximately RMB 145 per person at the exchange rate used in the article.
ChatGPT Enterprise pricing was available by contacting sales.
Comparing GPT model prices and capabilities
1 | Model name | Price in US dollars per 1,000 output tokens | Relative price | Illustrative bottled-water equivalent | Rate limit | Length limit, including context |
2 | gpt-3.5-turbo | 0.002 | 1 | C’estbon | 90K / min | 4K |
3 | gpt-3.5-turbo-0613 | 0.002 | 1 | C’estbon | 90K / min | 4K |
4 | gpt-3.5-turbo-16k | 0.004 | 2 | Ganten | 180K / min | 16K |
5 | gpt-3.5-turbo-16k-0613 | 0.004 | 2 | Ganten | 180K / min | 16k |
6 | gpt-4 | 0.060 | 30 | Evian mineral water | 200K / min | 8K |
7 | gpt-4-0613 | 0.060 | 30 | Evian mineral water | 200K/ min | 8K |
8 | gpt-4-32k | 0.120 | 60 | Fine Water, drawn from 600 metres beneath Mount Fuji in Japan | 150K/ min | 32K |
9 | gpt-4-32k-0613 | 0.120 | 60 | Fine Water, drawn from 600 metres beneath Mount Fuji in Japan | 150K/ min | 32K |
The article likens GPT-3.5 and GPT-4 to Intel Core i5 and i7 processors, and their 4K, 8K, 16K and 32K context windows to different amounts of computer memory.
This example assumes a heavy user making 300 calls and consuming 300,000 tokens per month.
GPT-3.5 4K — estimated monthly token cost of $0.60.
GPT-3.5 16K — estimated monthly token cost of $1.20.
GPT-4 8K — estimated monthly token cost of $18, which the author compared with the price of ChatGPT Plus.
GPT-4 32K — estimated monthly token cost of $36.
The article also looks at image-generation costs, using Midjourney as its example.
Midjourney’s published prices at the time

Annual Midjourney subscription plans
How should a chief AI officer choose among these plans?
The author suggests estimating how many people will generate images at once. Using an office with six shared facilities as a rough analogy, and an e-commerce team in which about one in three employees uses image generation, the example arrives at two concurrent jobs.
The Standard plan included 15 hours of Fast generation, three concurrent jobs and unlimited Relax generation. At the author’s estimate of 30 seconds per generation on the V5.2 GPU cluster, 15 hours would allow up to 1,800 Fast generations per month. For a company of 30–50 people, the article estimates RMB 30–60 per person per year.
For heavier image workloads, the article suggests the Pro plan and estimates roughly RMB 50 per person per year.
The following comparison draws on more than six months of MyGPTLife deployments.
The author models different blends of GPT-3.5 16K and GPT-4 8K, taking account of both cost and output quality. The comparison sets aside GPT-3.5 4K for its limitations and GPT-4 32K for its price.
No. | Plan including unlimited Midjourney image generation | Annual cost in RMB per user | Recommendation, out of five stars |
1 | 80% of tasks to GPT-3.5 16K and 20% to GPT-4 8K | 513 | 5 stars |
2 | 50% of tasks to GPT-3.5 16K and 50% to GPT-4 8K | 949 | 3 stars |
3 | 100% of tasks to GPT-3.5 16K | 223 | 4 stars |
4 | 100% of tasks to GPT-4 8K | 1675 | 2 stars |
As an essential productivity factor for an enterprise, in the future besides computers, networks, and offices, the company CAIO must choose to have AI consider these aspects.
1. As an enterprise’s Content asset and productivity tool, AIGC must be firmly in the firm’s hands.
2. The good AI model must be a fee in the future, and enterprises need to develop copyright awareness and pay awareness, not to allow their employees to use free AI tools outside, to create legal risks for copyrights, and to face the risk of loss of staff and an AI-Content competition between businesses and employees!
3. Corporate procurement must be aimed at improving effectiveness and reducing costs, not being a major source of injustice, especially in the case of AI’s new products, with too many herring-cutting programmes, seeking professional advice and choosing the most appropriate option for an enterprise.
The author’s estimate for AI spending at a small business was RMB 200–300 per employee per year. The article compares that amount with a modest employee benefit and argues that the productivity gains can be substantial.
