The funds arrived just as ChatGPT maker OpenAI released a new Privacy Filter product that matches some of DataMasque’s abilities, for free – on the face of things, drawing the Kiwi company into the “SaaSpolcapse” scenario now facing so many software companies in so many sectors.
First, a quick re-cap of what DataMasque does.
When companies build new applications, train AI models, or run analytics, software developers need high-quality data to test their systems.
Traditionally, companies either had to use real customer data (which creates massive security and regulatory compliance risks) or hand-craft fake data (which is time-consuming and often lacks real-world complexity).
DataMasque co-founders Grant de Leeuw and Greg Daniel.
DataMasque’s latest tool not only masks PII information in structured data – say, my patient ID in a medical database – but also the unstructured data that lies scattered around any organisation; say, my name as a patient in an email, Word document or a call transcript.
And if Chris Keall was masked to, say, “John Smith”, it would be John Smith in every mention.
DataMasque founder Grant de Leeuw says other solutions will simply blank out a name, breaking connections, and OpenAI’s PrivacyFilter only has shallow masking options.
Cheaper than the ‘free’ option
To appeal to the bean counters as well as the IT crowd, he points out that no corporate-grade AI is free.
Many large organisations have plans where they pay by the token – and data masking can rip through them.
DataMasque recently posted: “A 10 GB workload through Claude APIs (which can be run on AWS Bedrock for privacy) costs around US$37,500 ($62,800) for synthetic generation (US$15 [$25] per million output tokens, since the model is mostly producing) or US$45,000 ($75,320) for detection ($IS18 per million blended, since input and output volumes match).
“The same 10 GB through DataMasque’s Local AI Engine on a single $0.80/hour AWS GPU instance is roughly $1300, about $0.50 per million tokens – operating in your environment, with the same compute and with no data leaving your network.”
And bear in mind experts say we still don’t know the true cost of AI.
The industry is still in a “land grab” phase, similar to the early days of the streaming wars, or the halcyon lockdown era when services like DoorDash and Uber Eats were heavily subsidised by venture capital.
On-prem option
De Leeuw also points out that as long as you’ve got a GPU (graphics processing unit, or the main chip used for AI training and testing), you can run DataMasque entirely on-premise (that is, on your own company’s computers) using no tokens at all, “even for very large databases”.
De Leeuw stresses he’s “enabling AI” through DataMasque being able to test AI apps and services. And DataMasque has partnerships with the three biggest hyperscalers: AWS, Microsoft and Google.
But his point is that if you want to do everything in-house, you can.
Need for speed
The final part of his slapdown is that DataMasque is much faster.
He says his product uses algorithms to identify and mask data.
“We know a credit card is a credit card because it’s passing checksum,” he says, referring to the formula, which should produce a number ending in “0″ for a legitimate card, that banks and websites use to assess if a credit card number is kosher or mis-typed.
“Or we know it’s a social security number because it’s passing checksum.
“Whereas with AI, it’s still a probabilistic system. So where AI can de-identify a row every two seconds for a table with eight columns, we can do 50,000 rows every three seconds.”
That matters if you’re a client with billions of fields, de Leeuw says.
On DataMasque’s books, those include “a top four bank in Australia”, “a top four bank in New Zealand” and the South Carolina Department of Health and Human Services (which wrangles the US state’s Medicaid programme).
Why they’re back for more
“When we first invested in 2023, few enterprises had recognised the sensitive data problem that the DataMasque team saw,” OIF partner Isabella Rich says.
“Today, it’s blocking nearly every serious AI rollout.
“Grant and the team are building the data infrastructure layer for enterprise AI, and we’re proud to be doubling down.”
Icehouse Ventures principal Bex Gidall,says: “With a lean team based largely in New Zealand, DataMasque has managed to win highly competitive enterprise deals against far better-resourced global competitors.”
And from the newcomers, Wavemaker managing partner Paul Santos said: “[DataMasque] has helped customers to mask data seven times faster, reducing traditional masking workflows from days to hours. The strength of its value proposition has enabled the company to win multiple customers in BFSI [banking, financial services and insurance], government and telecommunications globally.”
What’s next?
The new funds will be used “predominantly for US expansion,” de Leeuw says.
The US$4m ($6.7m) is effectively a seed extension.
Wavemaker (which now holds a 12% stake) approached his firm.
Its Series A round will be a lot higher, and likely US-led, the co-founder says, as the start-up seeks to raise its profile in North America.
“We’ve the best product in the market, just no one’s heard of us.
“We need to get the word out,” De Leeuw says.
Chris Keall is an Auckland-based member of the Herald’s business team. He joined the Herald in 2018 and is the technology editor and a senior business writer.