EverString focuses on data accuracy with its new Data Platform
The predictive analytics firm is claiming accuracy rates for its business data of more than twice the standard rate using its new techniques.
Predictive analytics firms like EverString use computer models to make predictions about which businesses might buy your B2B products and services, based on patterns found in different kinds of data.
As CEO and co-founder JJ Kardwell told me, his company and its competitors had been focused on creating the best computer models they could, and the data they all used was something of a “commodity.” In EverString’s case, it got data about businesses by web crawling, by buying data from providers like Dunn & Broadstreet and by creating data through machine learning, such as classifications.
But, he said, the accuracy left something to be desired. At the scale of millions of accounts, he said, firmographic data generally was only about 42 percent accurate. Here’s the kind of account data EverString needs to generate, in this case about Marketo:
So, about 18 months ago, EverString decided to get into the data business, and this week it’s announcing its Data Platform.
The result, Kardwell claimed, is that EverString is now creating the cleanest source of account-based data on the planet, more than 90 percent accurate across millions of companies.
Its Data Platform has three main components. First, EverString built a platform for directly handling an unlimited number of crowd-sourced data-checkers, either directly or through firms in underdeveloped countries that provide such services.
The problem with this kind of low-cost human data checking, Kardwell noted, is determining the accuracy of the results. Common practice has been for a human reviewer to randomly check 1 percent of the data to determine an accuracy rate, he said, but that is obviously a hit-or-miss approach.
Instead, EverString set up its platform so that profiles of accounts are broken down into small bits of data tasks, like checking a company’s industry classification, and then at least two human data checkers will verify the same bit of data. If there’s a discrepancy, that data task is pulled for review by a higher-level worker.
Second, the new EverString Data Platform employs the most accurate data tasks from the crowdsourced human checkers to train its deep learning AI engine, which can then complete similar data tasks for millions of companies. Kardwell said the deep learning amplifies the human-based data processing by 10,000.
Finally, Kardwell said, the ETL process of moving data from one database to another has been entirely automated.
He pointed out that the 90+ percent accuracy rate EverString is claiming is for millions of companies. Some companies claim to have 95 percent accuracy for their business data, Kardwell noted, but that’s for, say, 75,000 companies.
“We’ve broken the [inverse] relationship between accuracy and scale,” he said, adding that some unnamed Fortune 1000 companies using the new Data Platform over the past six months have reported their growth rates have doubled because of the cleaner data.
“The predictive analytics world of two years ago is dead,” he said, and data is the new battlefront.