Startup of the day: a platform for identifying faces on the photos of Ever

Automatic identification of faces has become an ordinary function. New suppliers in this sphere appear every month, there is nothing interesting in them. But the American startup Ever demonstrates two features of this business.

First, the technologies are good enough that the quality does not bring significant advantages. On the promotional page Ever published test results, according to which the startup takes a confident first place. Based on the results of this study, it becomes clear that the technology of products in the test is virtually indistinguishable.

It is impossible to say how it is better from technologies if one of them receives 98.37% of correct answers in one test and 99.7% in the other, and the other 98.67% and 99.3% respectively (or vice versa). And even if some gets "only" 98.0% and 99.0%, there are no representable scenarios in which this will significantly affect the client's business. The technology with 99.999 percent accuracy will find new applications. But while everyone is hanging out in the range of 98-99.5%, their capabilities are identical.

The second feature - the quality of the product depends not on the genius of the engineers. Three years ago, the startup developed as far as possible from the artificial intelligence and pattern recognition application - the store of photographs.

The offer is quite common, there are many alternatives, but Ever had an audience, and in the cloud, the startup accumulated 12 billion photos - based on their service and coached its own neural network, hence the good results came.

After changing the direction of business, Ever received $ 16 million investment. After that, he changed the technical director - the old "overtook" Google with Microsoft, but still did not fit.