Gretel AI raises $50M for a platform that lets engineers build and use synthetic data sets to ensure the privacy of their actual data

Increasingly, conversations about big data, machine learning and artificial intelligence are going hand-in-hand with conversations about privacy and data protection. Now, a startup that is building tools to make it easier for engineers to implement the two simultaneously is announcing a round of growth funding to continue expanding its operations.

Gretel AI, which lets engineers create anonymized, synthetic data sets based on their actual data sets to use in their analytics and to train machine learning models has closed $50 million in funding, a Series B that it will be using to get the company to the next stage of development. The product — which is built as a SaaS product but can also be accessed via APIs — is still in beta but aims to be open to general availability later this year.

Anthos Capital is leading the round with Section 32 also participating alongside Greylock and Moonshots Capital. Greylock led the company’s previous round in 2020, and the startup has raised $65.5 million to date.

From what we understand, this latest round puts the company’s valuation at between $320-$350 million.

The idea behind using synthetic data sets is that it lets an organization remove the risk of data leaking that might contain personal information or other kinds of sensitive data. There are other solutions to address the same issue involving data encryption, although this can be a costly, time-consuming and resource-intensive approach that faces scaling challenges.

The germination for Gretel.ai came out of the direct experiences of the three co-founders in work they did as cybersecurity specialists at a range of organizations including IBM, AWS, Netscout and the U.S. military and over the years.

“We always found that using the right permissions with data was always the bottleneck,” said Ali Golshan, the CEO who co-founded the company with Alex Watson (CPO) and John Myers (CTO). They could see that the longer-term issue would be a growing need and priority for data privacy. “As the world moves from the web to the immersive world of sensors and IOT we are transitioning into a world where people will share their data unconsciously or unknowingly. But humans are not meant to be mined.”

As data engineers, their priority is to be able to work with data easily and quickly, but as citizens of the world, they were unhappy with data protection implications.

“Removing the bottleneck of compute is the problem we’ve solved, and we have created high-velocity development,” he said. “But now we are running into the bottleneck of the data. AI is on a collision course with privacy. At this collision course, we should create tools” to fix that.

Gretel’s opportunity is one that many companies targeting the enterprise market have taken in the world of digital transformation: many organizations now have large engineering operations working on applications to run their businesses, but they still do not have the firepower of the world’s largest technology companies. So Gretel set out to build a toolkit that would let any company build anonymized data sets for themselves, similar to what big tech companies use in their own data work.

The advantage of anonymized data goes beyond simply replacing a synthetic data set for an actual one; they can also be used to augment a data set, or to fill in the gaps where the real-world data might be lacking. Both of these are critical components, especially in cases where the data is needed to train systems, such as in the case of autonomous services, where you seemingly can never have enough teaching data.

Watson had previously worked at AWS (fun fact: we scooped when Amazon acquired his previous startup, harvest.ai), and he says that to date Gretel.ai has secured early customers in areas like life sciences, financial services and gaming. In more basic use cases, it can take as little as 10 minutes to create a synthetic data set. In more complex applications — for example in a genomic database, it could take several days. 

Relatively speaking, this represents “very low friction” for engineers, Watson said, both compared to other approaches such as data encryption using techniques like homomorphic encryption, or indeed the analogue approach of contacting third parties and getting permissions to use datasets. The latter can take six months or longer, too long in cases where time is of the essence.

“This significant Series B investment is a direct reflection of Gretel’s ambitious vision, swift growth, and strength of position in the AI industry as the standard-bearer of tools that enable privacy by design,” said Emily White, president of Anthos Capital, in a statement. “Gretel’s ease of use, the extendability of its services, and the superior accuracy and quality of its synthetic data are much-needed solutions to simplify the exceedingly complex legal and technical barriers companies face due to data privacy concerns.”

“Gretel gives data teams working in any framework or language the tools they need to build privacy by design into their existing workflows and data pipelines, greatly simplifying this process,” added Sridhar Ramaswamy, partner at Greylock. “Time and time again I hear from software engineers and data scientists about the value Gretel offers. Its developer-first, tech-agnostic approach to solving privacy issues is incredibly valuable to every business sector.”