At Imprima, we believe no one should rely wholly on AI without careful thought, regardless of its accuracy. Even the most accurate AI can’t fully see issues the way you can. Why? Simply because the AI does not have the specific context and responsibilities that you have. At the end of the day, you own whatever decision gets made with AI’s assistance – not the AI itself.
That said, we still aim to maximise accuracy. Contrary to what you may think, the goal isn’t to replace human involvement. It’s to save you time and effort while keeping you deeply involved in decisions.
Consider the following example
Suppose we had two AI engines that extract key information from documents:
- Engine 1 has a low precision rate of 20% – so 4 out of 5 extractions are incorrect (“false positives1“). You’d waste a lot of time digging through inaccurate data to find the good stuff!
- Engine 2 has an 80% precision – so 4 out of 5 extractions are correct, and only 1 out of 5 are “false positives”. Compared with Engine 1, there are 4 times less AI results you have to check.
The infographic below shows what this means in practice, and may make the difference between these two results even more obvious:
Higher accuracy means less wasted effort for you. We don’t pursue it assuming you’ll trust the system blindly, but rather so you can work faster with better information while avoiding distraction. We aim to assist, not replace.
Interesting thought experiment? Actually, it’s more than that. At Imprima we have developed a new AI algorithm, based on the latest LLM tech, custom built and trained, which allows you to achieve very high recall2 and precision at the same time.
We will discuss this algorithm and share test results in a future blog post.
- False positives are when the AI predicts an item but it is incorrect. For example, if a user wants to identify the date the contract renews and instead the AI picks up the date the effective date.
- 100% recall means that every item you are looking for is identified.