Amperity is coming out of stealth mode today with an eponymously-named product that solves a problem which vexed CEO Kabir Shahani for years: how to connect data at large scale and create a unified view of the customer.
Shahani founded Seattle-based Amperity with CTO Derek Slager in 2016 in order to solve that problem.
The two had worked together before, selling their last company, Appature, to healthcare giant IMS Health in 2013. It was during his time at Appature that Shahani noted the problems which beset the process of implementing marketing automation.
“We had to bootstrap data to do what we wanted to do, we had to write a bunch of ETLs, we had to figure out how to get data connected,” he told CMSWire. “And you know — it still broke all the time. There were always issues with scale, matching data and inaccuracy.”
Today the company revealed its solution to those problems.
Amperity rapidly delivers unified customer data that has been distilled from a brand’s disparate and raw customer data sources — think email campaign actions, in store transactions, online behavior, browsing history, mobile interactions and more. It does this by continuously ingesting this raw customer data, without the need for ETL or schemas, freeing technical teams from data mining and saving months of implementation time.
The databases the platform creates can then be segmented or downloaded to any technology using Amperity’s visual or SQL segment editors.
Shahani freely admits — and told his early investors — that he wasn’t sure he would be able to crack this nut when he started the company. “We just didn’t know,” he said. As it turned out, it only took about eighteen months to go from “We don’t know” to product roll out.
A look back at his journey is instructive if only to understand the issues that Amperity’s intelligent customer data platform addresses.
The Journey to Intelligent Customer Data
The first problem the team tackled was how to handle large scale data in a way that didn’t require ETL and analysis up front. Their solution means users now have a pipeline in which they can dump their raw data.
Another problem was how to truly assign probability across the data sets instead of just, as Shahani puts it, “determining that something is 85 percent accurate and calling it a fact.” The team says it solved that problem too, figuring out how to create probabilistic connections between every pair in a cluster.
In real life scenarios this meant that a company could request a data set and have it be deterministically connected at 100 percent accuracy.
Here Shahani noted that 100 percent accuracy is not always necessary. “You can turn down the probability and still get a tremendous amount of accuracy. Let’s say I were to launch a campaign for all of my customers that had spent $1,500 with me but hadn’t bought one of seven products I wanted to sell. The cost of being wrong about some of those customers’ attributes is not very high. We had early adopters find that by reducing the accuracy from 100 percent to 97 percent — it would be wrong three out of 100 times about some attribute — they could exchange that for an additional 20 percent of customers to target for the campaign.”
Engineering Help From Facebook Alumnus
Amperity had basically solved everything it wanted to solve: except for one key thing. For a while it couldn’t get the application to stay up for very long. If it wanted to be a commercial success it would have to meet the enterprise SLA expectation of 99.999 percent availability.
Around this time Dave Fetterman, one of the first engineers at Facebook — he reportedly bunked with Mark Zuckerberg when the company was in its early days — joined the team and helped stabilize the product for commercial use.
“He knows scale and stability,” Shahani says. “You can find maybe 20 people in the world with that kind of experience.” As for the answer to Amperity’s stability issue, that was just old fashioned rigorous operational processes, Shahani said.
Amperity got its final nudge to move out of stealth mode a few months ago when some early customers approached the company asking for a multi-year license agreement. By that point the system had been stable for five months.
“And so we decided to go beyond being a product shop and actually launch the product into the market,” Shahani said.