☎️ Interview: Jordan Brandt, CEO of Inpher on the State of Privacy-Enhancing Technologies
On why monolithic incumbents are going to be left with all the data risk; why taking a vertical go-to-market strategy is sub-optimal; and why data federation is an inevitability
Collaborative computing is the next trillion-dollar market. We are at the beginning of fundamentally reshaping how data is used in the economy. When data can be shared internally and externally without barriers, the value of all data assets can be maximized for private and public value.
To explore the concept of Collaborative Computing further, I spoke with Jordan Brandt, CEO of Inpher, a company pioneering privacy-preserving machine learning (PPML). Highlights include:
Why monolithic incumbents are not only missing performance and revenue opportunities, they are now going to be left with all the data risk;
Why taking a vertical go-to-market strategy might be sub-optimal;
Why data federation is an inevitability
"Every company wants to be more secure and leak less data. Every company wants access to more data or just uses all the data it already has. Really it's a case of looking specifically at the jobs data scientists are doing today and making their lives easier."
When you pitch customers on better using their data, what benefits do you lead with?
As you say in the collaborative computing paper, we sell solutions, not technology. So the reality is that it depends on the specific problem the customer needs to solve. Different benefits persuade different customer stakeholders. The user is usually a data scientist and so, accessing more data is compelling, as is getting to work on data faster. Those benefits are of course significant, but less so to the economic buyer, who obviously cares about cost and ease of integration and those sorts of things. Then you need sign-off from the CISO who cares about security guarantees. And then you have legal who obviously need to make sure they remain compliant, and they understand the risks and all those things.
What is the most common barrier you encounter when selling your services?
The barriers are definitely falling. A few years ago the most common barrier was just market education. We spent a lot of time educating our customers and other decision-makers within the organisation. That's why we went with the term 'secret computing', because it's easy to understand what you are getting. We made that market education piece easier. Then we spend a lot of time making it easy for users to persuade economic buyers within the organisation. Everything from video explainers, case studies, factsheets, and all that stuff. Our job is getting easier, as we see more public work in the space. But now we are finding the whole process of selling is easier and privacy-preserving tools are more common.
So now really the barriers are closer to typical SaaS selling. It's all about how easy it is to spin up and use. How well does it connect with existing data pipeline tools? This is where privacy-enhancing tools still have the disadvantage because the fact is, it's not as easy as a standard SaaS tool. We are using advanced cryptography and so we need to be much more careful about how it integrates with other systems. This process is much easier if our customer is already using a micro-services architecture, but it's harder for organisations that are still using legacy systems. A banking executive said at a conference once that it's "easier to build a new bank than to re-engineer the existing tech stack". So I suppose this is a barrier for us, but more than that it's a huge problem for companies using these legacy systems. In a few years they will be left not only with a slower, more expensive and less secure tech stack, but they won't be able to protect their own and their user's data either.
Enterprise data infrastructure sits across the entire company, who is the best person to sell to in your experience? Compliance seems like an easy way in but you could get stuck selling a commodity product competing on cost?
Yes that's right, compliance was the place they started. They have a budget and a strong pain to solve with regulation. The next step is the CISO. They generally love our stuff because it reduces the risk they have to manage, we are shrinking the attack surface and limiting their data liabilities. So it's powerful for them. But downside risk only gets you so far. Once companies start using Inpher, it becomes obvious the people getting the most value are data scientists and, by extension, the commercial teams that are applying the algorithms for business value. So this is where we are now.
Which use cases do you think are the lowest hanging fruit for data collaboration tools?
I don't know if it's a case of finding use cases as such. Look, data collaboration or computing on encrypted data has widespread value. Every company wants to be more secure and leak less data. Every company wants access to more data or just uses all the data it already has. Really it's a case of looking specifically at the jobs data scientists are doing today and making their lives easier. For us it's about giving data scientists the tools to easily integrate into their workflows and then learning what algorithms and data types they want to work with. It's more granular than market specific or even use case specific.
I’ve found financial services and healthcare to be relatively early adopters of data collaboration tools mainly because of the regulation around privacy and data security. Have you found the same and what other verticals do you expect to be the next adopters?
The obvious answer here is to say the best opportunities are in heavily regulated industries like financial services or healthcare. But that's a little too crude. Financial services do have a greater technical understanding and budget to find ways to share data internally and externally. Healthcare like financial services has the need, but the sharing environment is just so complex. That said, pharma seems to be ahead of the game here. There is the obviously huge need and potential value unlock in sharing clinical trial data across the industry. And then the advertising industry has a relatively new but major need to find ways to target users in a private way. The advertising industry is not regulated to the same extent as financial services or healthcare, and so can experiment faster, which is a good thing for adoption.
All of this said, I don't think vertical-by-vertical is the best way to uncover and target opportunities for data collaboration adoption. Just like use cases, the best way to think about this is from the data scientist perspective. What are there problems working with data today and what tools do they need to better do their job?
What cultural, technical or social change would be required for demand in data collaboration to increase 10/100x?
A 100x change can only be driven by consumer demand. I think the key is when users realise it's a false trade-off between personalisation and privacy. It's sort of framed today that you have to give up your privacy so you can get these personalised services. And people generally like personalised services and so the trade-off is worth it for most people.
But it's no longer a trade-off we have to make. Secret computing and PETs generally mean that we can get personalisation without giving up our privacy. Now of course, we need some performance gains in the next few years so that plaintext and ciphertext processing is close to parity, but we will start to see that narrative. So people will ask: if I can get the same quality service without giving up my data then why not? You wonder at that point why all companies that use algorithms to deliver personalised services just won't do private computing. There might be some small performance or cost trade-off, but the benefits in terms of PR, compliance, security and risk management will make it a no-brainer.
When thinking about helping companies utilise their data, a sensible framework is: governance, sharing, and monetisation. It feels like 95% of companies investing in their data infrastructure are still on data governance, maybe 5% are finding ways to share internally, and <1% are even thinking about monetisation yet. Does this sound right to you?
At a high level yes. Data monetisation has been a promise since the advent of commercial PETs, technically we know this is where we need to end up to widely deploy this stuff. But the missing link has been the business model. And it's not as simple as saying, let's bundle up this dataset and set up some permissioning system to provide paid access. There are a whole bunch of complex questions around data control and ownership which aren't simple. More pragmatically, no company is dedicating resources to systematically addressing these questions yet. We will know we are getting closer when we start to see new business units with general managers with teams set up to take data assets to market. This is where forward-thinking CDOs are going, but still a long way to go.
We are seeing hundreds of startups looking to solve enterprise data problems, what do they need to know that you have learned the hard way?
Well I think your paper says it well: sell solutions not technologies. This is a particular challenge in cryptography because people who really understand the crypto obviously have to deeply care about the technology. So obviously technology is what matters. Teams can understand that clients might not care as much as they do about the tech, but the primacy of the technology will come through in sales and marketing. So it is really important that culturally the organisation empathises with clients and helps them sell internally.
It feels like the data consolidation model that has been at the forefront of data utilisation strategies has perhaps reached its limitations in terms of efficacy. With the emergence of “Data Mesh”, Collaborative Computing, and, more generally, customer centricity, do you see a horizon where a data federation model plays a more significant role in the lifecycle of data estates?
I haven't personally come across the term data mesh yet, but as it relates to this trend towards decentralisation and data federation, yes I see this. These architectural shifts are inevitable as new technologies emerge to enable new things. The reality is that the Cloud solved a whole bunch of challenges with distributed management of computing, storage and networking in a cost-effective way. Once it was all in one place it was much easier to analyse it and do statistics. We just didn't have the tools to query or learn from distributed data until very recently. So we now have the tools, and the push is from regulation and national data sovereignty policies. Now it's not inevitable to just bring everything together in one place to process. For many organisations, actually not centralising data is a net benefit. So yes, a data federation model makes sense. But again, it's important to note that the analyst doesn't care about data federation. They care about availability, speed and time-to-deployment. So data federation will win in so much as it solves the everyday problems for customers.