Betterdata: Engineering Data Privacy with Generative AI

The idea for me actually came from my Ph.D. My last project involved ethically hacking the Ethereum blockchain, where I managed to reverse-engineer how user wallets are created. In blockchain systems, instead of a password, you have a private key. I was able to reverse-engineer Ethereum's wallet-generation framework and gained access to 670 user wallets. This allowed me to access those wallets and take funds out without anyone even noticing (of course I never did that and reported my findings together with defensive measures in my Ph.D. thesis).

This experience made me realize that when we generalize the concept of using encryption for data sharing, there must be better alternatives. At that time, DeepFakes were also making waves, with people substituting faces with celebrities like Tom Cruise in humorous clips on YouTube. DeepFakes are essentially a kind of synthetic data generation that uses real data (e.g., celebrity face images) to generate synthetic media, such as altered videos or images, that appear to be real. Our idea was to use this same kind of technology, but instead apply it to tabular data to solve a real business problem: using data with compromising privacy.

We conducted market research by identifying a significant opportunity stemming from the General Data Protection Regulation (GDPR) in Europe, which was enacted in May 2018. The GDPR affected over 28 million businesses globally, creating a huge market for privacy tech solutions. While this statistic validated the market size, it didn’t directly guide our product development.

The key for us was speaking to 50 potential customers to find convergence in their needs. These customers ranged from startups and SMEs to large MNCs. Through these discussions, we realized that companies with substantial private customer data, typically large MNCs or tier-one enterprises, were facing significant challenges around data access and data sharing.

The "magic number" of 50 isn’t arbitrary. Some suggest 100, but this is difficult for most founders starting out, so 50 strikes a practical balance. When speaking to customers, the goal is to find convergence. For example, out of 10 customers, if three mention the same problem or problem space, it indicates a pattern. By speaking to 50 customers, you’re likely to identify 10–15 customers with overlapping issues, which is a strong foundation on which to build a prototype.

In the early stages, we built synthetic data generation algorithms and a basic UI for users to upload test datasets, generate synthetic versions, and compare the two. Using five open-source datasets, we secured PoCs and pilot programs with banks.

By testing our early prototype with customers early on, we soon realised that banks would never let us work with their actual customer data outright because of privacy concerns. So we had this gap—on one side was the bank, and on the other, us, without any knowledge of how their real data looked. To bridge this, we installed our solution on-site, gathering feedback through consultative discussions.

Through this process of continuous consultation, we then stumbled upon another problem. While open-source data is clean and well-structured, real enterprise data is messy and often has missing values and inconsistencies. Our initial pipeline broke when handling such data, prompting refinements. Addressing these real-world challenges became key to evolving Betterdata’s capabilities.

Even if we provide the best results today, whether it works on customer data is subjective to their private datasets that we have zero visibility on. The gap between what we know and what they know continues to evolve. As we move forward, the goal is to keep bridging this gap. Over time, this gives us more differentiation because if someone else comes up with the same idea as ours, they can’t claim the same insights or capabilities that we have. They wouldn’t know, for example, what a typical bank’s data looks like, which gives us a significant competitive edge. In other words, the more customers we tested on, the more our "data advantage" grew because our algorithms were prepared for a wider variety of use-cases.

When it comes to iteration and refining our product, my approach is simple: listen to your customers, and your customers only. It's crucial to block out the noise—whether it's external advice, trends, or hype—and focus entirely on what the customer needs since they're the only ones actually paying you.

I emphasize that everyone on the team must be constantly asking one simple question: Which customer is asking for this feature or solution? As long as there’s a clear answer, we can move forward. The feedback we get directly from customers is our one true guide. External noise or tech trends, while potentially helpful, often bring more noise than clarity. This is especially true when there's so much talk about "hot" technologies like AI. Differentiating between signal and noise is crucial. You might make a quick buck by jumping on a trend, but long-term success comes from solving real customer problems.

While customers don’t always know exactly what they want, they are very aware of the problems they face. Our job is to identify those problems and find ways to solve them in a manner they have never seen before. This process often involves some back-and-forth where customers will often express a desire for certain features or capabilities, but we have to stay focused on solving the core problem rather than addressing their exact request since customers aren't usually in the best position to know how to solve their own problems.

We evaluate whether we’re on the right track by whether customers keep coming back to us. If they return with more feedback, more requests, or new problems, that’s a good sign that we’re addressing their pain points well or not-so-well (either way is fine). If they just say, “goodbye,” and walk away, we know we haven’t hit the mark. This ongoing feedback loop is what drives iteration.

I think the logic here is that when customers feel that you're helping meet their needs, they're more willing to keep working with you because it benefits them. On the other hand, if you're not doing a good job, it's simply easier to part ways than to take the time and effort to guide you towards a better product.

In the early stages, it’s important to ship products that solve a problem, even if they’re not perfect. Speed matters. Our first version is about delivering something that solves the issue quickly. We then let customers decide whether it's good enough. If they say it’s good, we consider it a job well done. If they say it’s not, we go back to the drawing board and iterate.

However, not all customers are disciplined or clear in their feedback. As a company, we must maintain more discipline than our customers. If a customer isn’t giving useful feedback and is only wasting our time, we need to be willing to move on and focus our efforts on those who truly value our work. This is difficult but necessary to avoid wasting valuable time and resources.

As we continue to evolve, it’s essential to stay focused on building our niche. Tech trends can be enticing, but we must remember that differentiation is what will sustain us in the long run. By staying true to the problem we're solving and being mindful of our customer needs, we can create something unique that stands out from the noise.

For our product, the core use case became generating synthetic data for privacy compliance. We decided to focus on the highly specific needs of industries like banking and government institutions, which deal with sensitive customer data and require secure, compliant systems. By zeroing in on this niche, we were able to build a solution that really catered to their pain points. This focus laid the foundation for our scaling strategy, which we are now using to expand beyond our initial markets and into the broader Asian market.

When it comes to assessing product-market fit, we focus on two key metrics. These help us understand whether we are truly resonating with our target market and if we're headed in the right direction.

Metric 1– Customers' Willingness to Pay

The first metric we track is customers' willingness to pay. This is a straightforward way of gauging whether customers see enough value in what we're offering to part with their money. Here's how I approach this:

• Meeting Quantity: I keep track of how many meetings we can book each month, as well as the quality of those meetings. The more meetings we have, the more we know that there’s genuine interest in what we’re offering.
• Meeting Quality: It's not just about the quantity of meetings, but also about the type of people we are speaking to. Are they decision-makers or key stakeholders? This helps us validate whether we’re talking to the right audience.

At Betterdata, we applied this principle early on by focusing on getting as many meetings as possible with the right people—mainly key decision-makers within banks and large enterprises.

Metric 2– How Much Customers are Willing to Pay

The second metric we look at is how much those paying customers are willing to pay. This is a bit more nuanced, as prices can fluctuate during the early stages, but ultimately, we needed to identify our sweet spot. When customers are willing to pay a premium for what we're offering, it signals strong potential for product-market fit.

For us, we realized early on that large enterprises, especially those handling sensitive customer data (such as banks), were more willing (and able) to pay a premium for our synthetic data platform. We focused on targeting these companies because they had the budget and the need for privacy tech, especially in light of GDPR and ever-increasing data protection regulations globally. As we moved forward, we found that companies were not only willing to pay but were also willing to pay premium prices for solutions that helped them innovate using their data while maintaining compliance. Once we had identified the top use cases where customers were both willing to pay and willing to pay higher ticket sizes, we focused on building our platform around these specific niches.

Coming from an academic background, my early passion was primarily about educating people on the privacy space and the potential of synthetic data. So, if a customer told me they were using anonymization and didn't need synthetic data, I'd dive right into an hour-long educational seminar (passion has no price right). However, we weren’t making any sales. We were just educating people, wasting a lot of time, and seeing no return on our efforts.

Over time, I learned a valuable lesson: sometimes, you need to let the customer decide if they have a problem that your solution can solve. Today, if a customer says they’re using anonymization and don’t need synthetic data, I simply tell them, “You don’t have any problem. You can solve your issues with what you have.” This shift in mindset helped us better qualify customers and avoid wasting time on those who weren’t ready to invest in our solution.

As we began to refine how we would charge customers, it became clear that pricing had to align with the specific needs and questions of each customer. Early on, we spoke with a wide range of decision-makers—data scientists, Chief Data Officers, Chief Digital Officers—and they would ask all sorts of questions. Some were things we hadn’t even thought of or features we didn’t yet have in our platform. Initially, we tried to address every question and feature request, which created a lot of friction and confusion.

Eventually, after speaking with almost 100 customers, we created a product roadmap that categorized our offerings into different AI models. Each model had its own set of features. So, the next time we spoke to a customer, we would map their requirements to one, two, or all models, depending on their needs. This gave us a clearer way to justify pricing. For example, if a customer needed Model 1, they paid a specific price. If they needed Model 2 as well, the price would increase, and so on. This approach allowed us to manage customer expectations and prevent them from asking about features they didn’t need. If they wanted all the features, they would know they’d have to pay a higher price.

This also introduced focus into our customer conversations. We were able to clearly communicate that if they only needed one model, they didn’t have to ask about features from the other models. And if they were interested in the full suite of features, they would be expected to pay a premium for it (you see people get serious about what they want when they know about pricing implications, otherwise it’s just window shopping).

At Betterdata, our customer engagement process unfolds across three distinct phases, each with its own approach to pricing and payment.

Phase 1: Trial and Benchmarking

The first phase begins when we reach out to potential customers. During this stage, we aim to demonstrate the effectiveness of our solution, typically using open-source datasets to show benchmarks and results. This serves as a trial period for the customer, allowing them to evaluate the feasibility of our solution without committing to a full contract.

Phase 2: Pilot or PoC (Proof of Concept)

Once the customer finds the trial results attractive, we move into the second phase, which involves conducting a PoC or pilot with the customer’s private data only. 

During the pilot, we work closely with the customer to refine the solution and ensure it fits their specific needs. However, we initiate the PoC with the understanding that, if successful, the customer will enter into an annual contract with us. This ensures that we are not just providing a temporary solution but establishing a long-term relationship with the customer.

Phase 3: Multi-Year Contracts

Once the pilot is completed successfully, we move into the third phase, where we transition the customer engagement to a commercial contract that includes all the requirements, models, and services the customer needs. Our goal is to move towards multi-year contracts, which provide stability and long-term partnership opportunities.

We focus on cracking the code for marketing in a non-conventional way because I don't always find traditional frameworks useful. Instead, my approach revolves around answering a very basic yet essential question: Which customers are willing to pay, and how much? Once we know this, we can craft a message that clearly shows how our solution fits their needs.

Our marketing strategy revolves around two primary components:

1. Building a strong network of lead magnets to facilitate warm introductions
2. Building our brand from a thought leadership perspective through
3. 1. Customer traction,
   2. Technology evidence, and
   3. Regulatory guidelines.

This holistic approach has helped us effectively navigate the diverse market conditions in Asia and build strong relationships with our B2B customers. By focusing on these areas, we’ve been able to drive business growth and foster trust with our stakeholders.

One of the most effective marketing channels for us has been building a network of lead magnets. These are influential figures in the industry, such as senior advisors, thought leaders, and strategic partners. For instance, we engage with prominent institutions in Singapore like IMDA and SUTD, as well as our investors like Plug and Play Tech Centre. These entities help us stay informed and allow us to directly reach the right people.

The focus is on nurturing relationships and generating warm introductions. Instead of cold outreach, which is least effective in B2B sales, we aim for a healthy number of intros each month. These intros help us quickly assess whether a customer needs our product and shorten our sales cycle significantly.

Building our brand has been another critical aspect of our marketing strategy. Our approach focuses on three core elements of credibility:

• Customer Traction: We highlight our traction by working with high-profile customers and showcasing case studies. For example, we worked with the US Department of Homeland Security this year, and we aim to secure more prominent clients in the future. Publicly announcing these partnerships builds credibility and trust among potential clients.
• Technology Evidence: Our R&D team plays a key role here. Everything we build is backed by science, and we publish papers that undergo peer review by AI experts. This ensures that our solutions are technically sound and provides credibility within the AI community.
• Regulatory Guidelines: We actively engage with regulatory bodies to ensure our technology complies with data privacy guidelines. For instance, we have advised PDPC Singapore on two guidelines that share how businesses can comply with PDPA (Personal Data Protection Act) Singapore using data anonymization and synthetic data. This involvement demonstrates our commitment to compliance and reassures customers that we understand and deeply respect data protection laws.

Personally, I dislike receiving cold calls, so I don’t believe in outbound strategies for enterprise sales. If I don’t believe in it, I can’t build a company that relies on it. Instead, we used three primary methods to engage potential customers.

1. Account-Based Marketing (ABM): Our users are technically savvy and mature, so we target them with content that directly addresses their specific problems. While many know about AI and synthetic data, it’s the niche problems that pique their interest. For instance, banks may require a platform capable of handling a billion customer transaction records. Open-source models often can’t scale to that extent. By presenting solutions tailored to these specific challenges, we attract interest.
2. Targeted Content Marketing: Regularly speaking to customers informs us about their challenges, which in turn helps us create relevant content. This content is then used to target ideal customer profiles to reach out to, while also helping to establish ourselves as thought leaders in the privacy engineering and AI space.
3. Raising Awareness Through Partnerships: There is a significant lack of awareness about privacy, especially in Asia, let alone using synthetic data to address privacy challenges. To tackle this, we collaborate with partners like Infocomm Media Development Authority in Singapore, the National University of Singapore, Singapore University of Technology and Design, and VC firm Plug and Play to share knowledge. We also contribute to advisory guidelines on data privacy with PDPC (Personal Data Protection Commission) Singapore, which helps build credibility and raise awareness for the problem and solution we are trying to validate.

For Betterdata, social media has not been a major driver of new business, especially since we operate in the B2B space and engage with tier-one enterprises. Currently, all of our ongoing projects are with large enterprises, and none of them came through social media. So, at least for us, we haven’t seen significant success directly from social media channels.

However, content marketing—specifically publishing technical papers—has played a crucial role. For example, one of our projects in Hong Kong started because a potential customer read one of our technical papers published in IEEE. They reached out to us referencing the paper, which led to a swift pilot project and a growing relationship. This shows that while social media hasn’t been a major source of leads, the technical content we produce can still generate high-quality inbound interest.

We’ve recently focused more on the quality of the content we create, aiming to produce pieces that address real problems our customers face, rather than generic content. This shift has led to a noticeable increase in high-quality inbound leads. For example, one recent lead came from a senior architect at eBay in the US, who reached out after reading a blog we wrote about an overview of synthetic data.

While social media specifically hasn’t been a primary tool for lead generation, high-quality content that is well-informed by customer needs has been a valuable asset. Over time, we’ve seen a clear correlation between high-quality content and inbound interest, suggesting that content marketing—especially when it speaks to real customer challenges with depth and length—can still be an effective strategy even in a B2B context.

For sales in B2B software, the most critical factor is case studies. In Asia, especially, there's a strong focus on risk aversion, meaning potential customers often want to know who else is using the product before making any commitments. If you don't have case studies to show, you need to build credibility through other means, such as demonstrating customer traction, providing proof of your technology, and adherence to regulations. This triangle of credibility—customer traction, technology, and regulatory validation—helps customers feel more comfortable with your product and your company's reliability.

When it comes to pricing, there's a clear correlation between the size of the deal and the level of evidence you need to present. For smaller deals ($10,000), basic feedback may suffice, but for larger ones ($100,000), more evidence is needed to justify the cost. Customers will ask about your product's differentiators, frameworks, and users.

Balancing credibility and pricing is tough—you can’t charge $100K with no evidence, but you also can’t build credibility by offering the product for free. Finding the right customers requires persistence and research, and often, luck. Out of 100 prospects, only one might truly believe in you, but that one can make a big difference.

The curve from zero to $100K is more exponential than linear. Sales aren’t straightforward; they’re driven by both logic and emotion. Even with strong evidence, stakeholders may remain unsure. Evidence must align with guiding the customer through their decision-making, building their confidence until they’re ready to commit long-term.

We closed our seed round in April 2023, a time during which we weren’t actively raising. Founders should always have a fundraising mentality. I wouldn’t recommend waiting until you have just a few months of runway left. This is cutting it too close. Instead, stay in fundraising mode actively, and when the timing is right, execute. How do you know when the timing is right? It is when investors are reaching out to you, not the other way around. 

For us, our investment came through a mix of referrals and inbound. We didn’t start fundraising in August 2022, but we kept conversations going, and eventually, things moved forward. Even our lead investor initially rejected us. They said we weren’t the right fit, but a few months later, we simply re-engaged again and things fell into place. I cannot emphasize the importance of staying humble and treating everyone you meet with humility, because you never know, they can always come back and make things work.

The key here is to always have a fundraising mindset. You don’t need to be in panic mode with only a few months of runway left. Instead, keep your pipeline open and always be building relationships and networking. This way, when the time is right, you can execute quickly.

For fundraising, one of the most effective strategies we used was creating a comprehensive, easily digestible investment memo. We built this memo on Notion, which covered everything: the problem we’re solving, the solution, the market, the team, risks, challenges, and opportunities. It also included a data room with all the necessary details in one place. The goal was for any investor to review the memo and data room within 30 minutes, getting a clear overview of everything without needing to chase down additional information. This approach made it easier for investors, and we received positive feedback because we made their jobs easier by providing all the necessary details upfront, saving them time.

We also focused a lot of attention on our financial model, especially since investors in Asia really value numbers (and I mean really!). At the seed stage, creating a 5-year financial forecast is overkill, can be tricky and is needed by almost every investor I met, but we did put a lot of thought into our assumptions and made sure the numbers made sense. In retrospect, I still believe it is an overkill because I have yet to meet a startup in Asia who said they hit their projections, but the financial model can still be used as a proxy to understand what kind of business you want to build. 

Another key aspect of our approach was the format of the data room. We ensured that all the information was well-organized and easy to navigate. Instead of bombarding investors with numerous files that would take hours to go through, we focused on making each file concise and clear, so they could quickly grasp the key details. As the old adage goes, brevity is the soul of wit and that is exactly what we did.

When it comes to crafting and conducting a pitch, my approach is to keep things simple. At the seed stage, I focus on four main things: problem, solution, team, and market—that's it. The key litmus test for me is how well I can summarize each of these aspects in just a few words, eliminating unnecessary details while still communicating the full picture.

The goal is to find the few words that convey the essence of the message. Being able to communicate everything effectively in under a minute is an excellent stress test. It forces you to find words that are meaningful and precise, especially when you’re pitching.

For later stages, you can expand on things like market segments, traction, acquisition rates, retention rates, and logos you’re working with. But for seed stage, focus on those four key areas. I would say you need to be particularly strong in at least three of those four areas. If you can do that, you’ll paint a compelling picture.

When it comes to growth and scaling using the funds that you've raised, I believe it’s essential to have a clear goal or sense of direction. Let me explain it like this: Imagine I’m hungry and need to find a restaurant. There are restaurants all around me—360 degrees, the full circle—but I have to choose one. Once I decide, I need to start walking toward it. Every step brings me closer.

Two critical factors here: First, choosing the right restaurant (or, in business terms, setting the right goals). Second, understanding the timeframe and the steps required to reach those goals.

This is where discipline comes in. What you can’t afford to do is set a goal, take two steps, then change your mind and start walking toward another goal. If you keep jumping from one thing to the next, you’re chasing your own tail—you’re not progressing in any specific direction, and that is super-extra-very unproductive. The secret to growth is consistency, not necessarily hard work. It’s the consistency of doing something over time that allows you to gain insights and understanding that others can’t easily replicate. This only happens if you focus on something for a prolonged period.

This is crucial, especially in today's world being constantly distracted by things like TikTok and YouTube, where we can skip or speed up content. We think we control everything, but when it comes to goals, that’s not how it works. Especially when things aren’t going as planned, you have to stand your ground. You can’t just change targets every time things get tough. When things get tough, the tough get going.

I met my co-founder, Kevin, during the Entrepreneur First program. At that point, we were both in teams that had fallen apart. We had six weeks to come up with something to pitch, so there was urgency. We discussed several ideas, and I liked how my co-founder approached problems from multiple perspectives.

After a few meetings, I just had this gut feeling that he was someone I could work with. On the flip side, Kevin, came to me with a framework—a list of 100 questions to ask your potential co-founder. That’s the opposite of how I operate. I told him straight away that it wasn’t my style. I don’t need to answer 100 questions. I’ve made my decision that we can work together, and that’s it.

It was a moment that really highlighted the contrast in how we approach things. My approach is more non-linear, while his is more structured and methodical. It’s a great fit for us because it allows us to approach problems from different angles, and that’s been invaluable in running the company together.

First, uncertainty. I always ask candidates how they handle uncertainty and for a specific example. This can be personal or professional. The way they walk me through their process gives me insight into how they would handle the uncertainty that comes with working at a startup. In a startup, you're often figuring things out on your own, with no roadmap. You need to be independent and self-reliant, believing that no matter what, you'll figure it out. So how someone handles uncertainty is a key indicator for me in determining whether they’re a good fit to be our team members.

Second, I believe skills can be learned over time, but culture is what keeps people at a company. People don’t usually leave because of a lack of skills; they leave because of cultural misalignment. So, it’s really important for me to ensure that there’s alignment between an employee’s personal goals and inspirations and the environment we provide. I want to create a space where every team member feels appreciated for their contributions and secure in their livelihood. I don’t want anyone distracted by concerns about their salary, equipment, or benefits. Once those basic needs are met, they can focus fully on their work.

The key is to commit to your decision and see it through to completion. This is what leads to real results. Many people confuse hard work with grinding away endlessly without taking time to learn and reflect. If you’re working 24/7 but not gaining insights, then I assure you, you are on the wrong path. So, it’s crucial to find your goal post and align with your team in that direction.

To give you a personal example: I’m a Muslim, and we pray five times a day. When we pray, we face the Kaaba, a place of worship in Mecca, Saudi Arabia. This direction is essential—without it, everyone would be praying in different directions, creating chaos. This very alignment, the sense of direction, is what brings unity, both in our personal and professional lives. By extension of logic, in business, alignment with your team on the goal post ensures everyone is moving in the same direction, leading to compounding and multiplying results. That is the true secret of growth.

Once you define your goals and focus on your strengths, you’ll either achieve them or you won’t. If you hit your goals, great—you can push beyond them. If not, step back, reassess, and refine your strategy. You can always adjust your approach, but the key is to keep moving toward that specific goal post.

If I had to draw an org chart for the team right now, this is what it would look like:

1. Business
2. Product
3. 1. Software Engineering
3. Machine Learning Engineering
4. R&D

We have the C-suite at the top, followed by team leads, and then the executors. The key here is that we take market expectations and distill them down into actionable tasks.

The process and workflow looks like this:

1. Market expectations are received.
2. Tasks are defined and assigned to the team.
3. Execution happens with no communication barriers between team members. Everyone can talk to anyone, and the focus is always on getting things done effectively.