In the past two years, almost all companies have been talking about AI, but there are still a few that have truly put AI into effect. Observing cases of 'using AI but not using it', the problem often lies not in the model itself, but in the three hurdles: data, organization, and expectation management.
The first hurdle is data. Large models require high-quality, structured, and retrievable business data to realize their value, but most enterprise data is scattered across different systems, with inconsistent formats, confusing calibers, and unclear permissions. Many AI projects perform impressively in the POC stage, but degrade as soon as they enter the production environment, often due to data not keeping up. Data governance may not sound sexy, but it is the hardest foundation for AI implementation.
The second hurdle is organization. AI projects are often led by IT or digital departments, but the real demand side is on the front line of business. If the business department does not participate, pay the bill, or change the workflow, AI tools will ultimately be put on hold. I have seen many companies buying AI platforms, but no one is willing to bring in core business data - because the processes have not changed, incentives have not changed, and trust has not been established. The essence of AI implementation is organizational change, and technology is just one part of it.
The third hurdle is expectation management. Many companies have either high expectations for AI ("using AI can reduce costs and increase efficiency") or low expectations ("AI is just a chatbot"). The former leads to projects being halted due to low ROI after launch, while the latter leads to insufficient investment and superficial experimentation. The pragmatic approach is to start from one or two high-value, low-risk scenarios, set measurable indicators, run real results within 6 to 12 months, and gradually expand.
AI will not automatically change a company, what changes the company is the management decisions made around AI. Thinking clearly about data, organization, and expectations is much more important than choosing which model or platform to buy.