The difficulty of assessing and proving the effect of AI implementation is a myth; often there simply is no effect. Why? AI provides an effect in saving employee time or increasing their efficiency. This only works if a company has a huge number of people working on identical tasks—for example, a call center. But as soon as we come to a typical oil company, this doesn't happen. There aren't 100 people working on one task, neither in the office nor at the field.

A typical average AI solution proposes to solve 5% of one employee's tasks faster, so we take a department where the employee's tasks will be solved faster. What does the department or company get from this? Nothing. The employee will spend 5% more time drinking coffee, smoking, and the company will spend money on AI. But we can load them with other tasks! No—the business process is currently running and moving without these tasks, there is profit and everything is fine, they don't need new tasks. Therefore, AI implementation will be profitable only when a company has a bunch of programmers and they can roll out services to production faster, increase conversions and sales, not in a year but tomorrow—here the effect is measurable, which is why Codex, Claude code are being adopted everywhere. Software that allows writing letters faster is not needed by corporations.

If you are making an AI product for large companies, calculate what effect you give the company besides increased employee break time?