What is Guaranteed Accuracy?
If a mistake is found on a job, we credit you with 10X the cost of that job.
In other words: if we do not provide an accuracy of at least 90%, the service becomes effectively free of charge*.
The problem with accuracy ratings
Accuracy is a measure that gets thrown around by some companies as a marketing claim: “94% accuracy!”. On the other hand, accuracy may not be mentioned by other companies at all.
In reality, “accuracy” is only relative to the dataset tested against. One can easily put together a test set of ninety-nine images which yield accurate results and one image which doesn’t. This test set would produce an impressive accuracy score of 99%.
Even a “random” test set could be deceiving, as the model (AI) being tested could be trained with lots of similar data. Such a model would produce good results with this particular “random” test set, however it might perform much more poorly when a fresh test set from very different source is used. This is called “overfitting” in machine learning.
DIY accuracy test
The only way to get an accurate accuracy measure on your own “real world” traffic is to test on a sample of your production traffic. This is not easy to do beyond doing a small anecdotal test. To achieve statistically significant findings, a large sample set needs to be first collected, then manually labelled correctly. This will form your ground truth test set and it should be thousands or even tens of thousands in size. This process of collecting and labelling for ground truth needs to happen repeatedly to ensure that the results you are getting meet your accuracy requirements over time. Just because a model works for you today, does not mean that it won’t be stale a month from now.
The process of scientifically testing the AI models you pay for therefore becomes a burden. This causes many to just assume that they are getting good results from AI when in reality the big claims promised by AI providers end up being a ROI disappointment.
Image Intelligence eliminates this burden and risk by providing customers with Guaranteed Accuracy. Customers can trust that they are getting a high level of accuracy because of our money back guarantee.
* Terms and conditions apply. Contact us for terms and conditions.