Subscribe | Advertise | PRO Member Hub

TOGETHER WITH INTEGRAL

Perspectives from the people building the future of health AI…

We sat down with Shubh Sinha, CEO and Co-Founder of Integral, to unpack how the company is approaching one of the most important but under-discussed parts of the AI stack: safely activating sensitive real-world data.

Integral started in healthcare, working with pharma companies, payers, and healthcare data providers to make highly regulated data usable under HIPAA. That meant preserving the signal in claims, pharmacy, provider, demographic, and other real-world datasets while reducing privacy risk and producing the defensibility artifacts required for enterprise use.

Now the same problem is spreading well beyond healthcare. AI labs, data platforms, labelers, vertical AI companies, and enterprises all want access to proprietary real-world data because it contains genuine behaviors, decision-making patterns, and human complexity. But that data often comes with regulatory, contractual, privacy, IP, and business constraints that make traditional sharing slow and risky.

Shubh shared why healthcare data enablement was the blueprint to later expand, why the easy data era is ending, what most people get wrong about the privacy vs. utility tradeoff, and why the future of AI may depend on a trusted privacy layer for activating real-world data safely.

PS. Shubh also joined TBPN to discuss Integral, their recent raise, and why proprietary data is becoming such an important layer of the AI stack - and even got the first double-gong (!) in TBPN history.

Let’s start from the top. What’s the founding story behind Integral, and why start with privacy as the product?
My background has always been in data businesses in regulated industries.

Before Integral, I was at LiveRamp, where I worked on new verticals. LiveRamp’s core product was about identity resolution and data connectivity. If you think about a brand like L’Oréal, they might interact with a customer through their own website, Sephora, Target, an ad platform, and other channels. Before LiveRamp, those touchpoints were fragmented. LiveRamp connected the data siloes so companies could understand a more complete customer journey.

That same problem exists in healthcare, but on steroids.

Large healthcare organizations like pharma companies, payers, pharmacies, and providers all touch pieces of a patient journey, but none of them owns the full picture. That matters because the more intelligently healthcare companies can understand patient journeys, the better they can develop medicines, identify underrepresented populations, and deliver care-changing products to the right people.

At LiveRamp, I helped build healthcare data infrastructure to ingest and analyze sensitive healthcare data and combine it with consumer attributes like geography, income, race, ethnicity, and other social factors that shape care. That work exposed the core bottleneck: the manual process of cleaning, packaging, and approving regulated data was slow, expensive, and fragile.

It could take months to get a dataset live, and then you had to keep doing it continuously because datasets change as people’s behaviours change. The idea behind Integral was to turn that manual privacy and compliance process into infrastructure.

You started in healthcare and pharma, one of the hardest data environments in the industry. What did that teach you?
Integral today is a platform for privacy engineering, remediation, and documentation for proprietary data going into AI. But the reason we can apply what we do across industries is because we started in healthcare under HIPAA, which is one of the earliest and strictest privacy frameworks in the US.

Healthcare data has always been proprietary data. Your doctor is not posting your medical records on the open internet. Your pharmacy claims, lab results, and care history are highly valuable, but also highly sensitive and locked away. We spent the first few years of the company helping pharma companies, payers, data providers, and other healthcare organizations unlock that data in a way that was both compliant and useful.

That was the hardest possible starting point - which is exactly why it became the training ground for the next frontier. We were working with some of the most regulated data, for some of the largest companies, in an environment where privacy failures have real consequences. But that is also what made the company stronger.

We learned how to get datasets live in days instead of months, while preserving more signal and maintaining compliance. We also learned that the challenge was not unique to healthcare. Healthcare just felt it first.

Now, as AI companies seek real-world data in other domains like finance, energy, enterprise software, codebases, and customer interactions, the same problem is spreading. There may not always be HIPAA, but there are regulatory constraints, contractual obligations, trade secrets, IP concerns, and business risks. The same privacy engineering problem shows up again.

Why is proprietary data becoming so important for the next generation of AI?
The easy data era is ending.

The first generation of AI was trained on public data, open internet data, and heavily curated human data. That was initially enough to build very powerful general models. But if AI is going to get better inside specific industries and real-world workflows, it needs the data that captures how the world actually works.

That means proprietary data. Health records, claims, financial transactions, customer interactions, CRM data, Slack messages, codebases, operational systems, and other datasets that are not freely available online.

The edge now is in that real-world complexity. A model trying to understand healthcare, finance, enterprise workflows, or consumer behavior needs the messy patterns that people generate every day. That data can come from a broker, a mid-market hospital, a pharmacy, a financial institution, or any company producing valuable data exhaust.

The opportunity is that enterprises can monetize data they already produce, while AI companies get the signal they need to build better models, evals, benchmarks, and products. But that only works if the data can move safely, quickly, and continuously. That is the infrastructure problem we are trying to solve.

Most teams think privacy and utility are in tension. Either strip the data to protect privacy, or preserve signal and take on risk. Why do you think that framing is wrong?
That framing is exactly what we are trying to move past.

A lot of teams today take a very destructive approach. They strip fields, mask values, redact aggressively, or generate synthetic substitutes. That can make people feel safer, but it often destroys the value of the dataset. You strip the noise, but you also strip the signal.

What the market needs is a scalpel, not a hatchet.

Take a name like Shubh Sinha. That is identifiable. A blunt process might simply remove the name. But sometimes the fact that the person has an Indian-origin name carries useful signal for a downstream model or analysis, especially if you are studying demographic patterns or access issues. Instead of just deleting it, the right approach resynthesizes a realistic name that preserves the statistical meaning without preserving the identity.

That is the difference between privacy work and privacy engineering. It requires understanding the regulatory constraints, the privacy risks, and the technical value of the data. Then applying the right transformations surgically so the data remains useful while reducing risk.

The goal is not to make data useless in the name of safety. The goal is to make high-quality data usable in a way that is defensible.

In plain English, what does Integral actually do for an AI team?
Integral is the independent privacy layer for real-world data.

If an AI team purchases or receives sensitive real-world data, we help make sure that dataset satisfies the regulatory, contractual, and business constraints tied to it. That might include HIPAA, CCPA, internal legal requirements, seller restrictions, or another framework that governs how the data can be used.

The output is higher-quality data, available faster, with a defensible privacy posture.

Our customers started in pharma and payers, and we still work with top pharma companies, top payers, and healthcare data providers. But we also now work with AI labs, labelers, data platforms, and vertical AI builders that are trying to use sensitive proprietary data across different modalities.

The common pain point is always the same: how do I get the data I need in the most compliant way possible, as quickly as possible, and in the highest-quality state possible so I can actually use it for my business or AI use case?

You describe Integral as Forward Deployed Privacy Services. What does that look like in practice?
The core problem is getting high-quality proprietary data quickly and compliantly. In practice, that means embedding both software and privacy expertise directly into the customer’s data pipeline.

There is a technical side. We sit where the data needs to go and help engineer it into a state that is useful and compliant. We are not forcing customers into an artificial workflow. We meet them where the data already needs to move.

Then there is the privacy side. A lot of the datasets we see are being used or sold for the first time. That means the applicable rules are not always obvious. The data may cross countries, regulatory regimes, contractual frameworks, or internal risk standards. Our Forward Deployed Privacy team includes people who understand both regulation and technical implementation, so they can translate those constraints into privacy configurations that actually work in the pipeline.

The important part is that privacy and engineering are not separated. The technical output needs to be high quality, but it also needs to pass the regulatory or commercial bar set by HIPAA, CCPA, a data seller, a data buyer, or another governing body.

That combination is what turns privacy from a final review step into an operating layer.

Why is independence so important here?
The proprietary data economy runs on trust.

You have sellers who want to monetize valuable data without giving away sensitive information, IP, trade secrets, or creating regulatory exposure. You have buyers who want as much signal as possible without inheriting radioactive risk. Those two sides need a trusted layer in the middle.

That is where independence matters.

If the buyer assesses the risk themselves, the seller may not trust the conclusion. If the seller does it alone, the buyer may not trust the utility or the methodology. Integral can sit between both sides as an independent layer that helps ensure the exchange is above-board and defensible.

That independence matters for product, go-to-market, and trust. It lets us work with some of the largest companies in the world, but also with smaller data owners taking a chance on monetizing a dataset for the first time.

Another way to say it is that Integral is Switzerland. We are not trying to own every party in the data ecosystem. We are trying to be the infrastructure layer that allows the ecosystem to move more safely.

There are already tools for masking, tokenization, redaction, and synthetic data. Where do those approaches hit their limits?
Those tools are useful, and we work with them when customers already have them embedded. We are not trying to rip everything out. If a customer has a tokenization partner or redaction provider they like, we can work alongside that.

But masking, tokenization, and synthetic data are not the whole problem.

The harder problem is end-to-end compliant data logistics. You need to pick up the data, process it, transform it correctly, preserve utility, assess residual risk, produce the right documentation, and deliver it continuously. That is not just a point solution problem. It is an infrastructure problem.

Our secret sauce is combining privacy policy expertise with data infrastructure engineering. We did not approach the market as a set of compliance rules to follow. We approached it as a set of parties that need to exchange data safely.

That mindset changes the product. Compliance is embedded into the conveyor belt. Data scientists and AI teams get the inputs and outputs they need, while the system is doing the privacy and regulatory work underneath.

Can you walk through a concrete healthcare example where Integral changes what is possible?
Take a top pharma company launching or scaling a prescription product.

Developing a drug can take years and cost enormous amounts of money. Once it is approved, the company needs to understand whether the right patients are actually getting access to it. That becomes a data problem.

A patient may see a TV ad, talk to a doctor, receive a prescription, go to a pharmacy, and then continue through a messy journey with many middlemen. Pharma companies want to understand that journey so they can tailor outreach, deploy reps, identify underrepresented groups, and make sure the medicine reaches the people who need it.

To do that, they may need to combine ad data, provider data, pharmacy data, social demographic data, and geographic data into a 360-degree view. That has to happen continuously, not once a year after a long consultant-led review.

Integral helps make those datasets usable in a high-fidelity, compliant way. Instead of waiting months to get a dataset live, data scientists can get what they need in days. And it is not just faster. They often get better data because we can preserve or engineer useful attributes rather than simply remove them.

The outcome is speed, quality, and defensibility. The pharma company can move faster, the data scientists get richer inputs, and the data can be used in a way that stands up to scrutiny.

You just announced $25M in total funding. Why this round, and what does it unlock?
Integral’s success is predicated on scale. We’ve proven out our initial value, and now we want to make sure we invest to see every piece of real-world data that requires sanitization. 

Patient data and proprietary enterprise data become more valuable when you can preserve longitudinal linkage and behavioral complexity over time. The real value comes from being able to trace journeys, understand changes, and continuously improve decisions.

With our earlier funding, we built a robust product and proved the minimum viable network across data providers and data purchasers. We showed that there is a real flywheel: more high-quality data leads to better decisions, better decisions create more value, and that value supports more demand for data.

This round lets us put more fuel on that fire.

The goal now is to scale across verticals: more data sources, more purchasers, more data types, more verticals, and more use cases. That means expanding the team, investing in product, growing go-to-market, and continuing to build the privacy engineering and statistical methodology needed to support more complex real-world data flows.

It also signals something bigger about the market. Enterprises are starting to understand that proprietary data is becoming one of the most important inputs for AI. The funding is a proxy for that appetite.

If Integral works, what does the data supply chain feeding AI look like in a few years?
Today, a lot of people are still unsure what they can do with sensitive proprietary data. Can I use it? Can I sell it? What needs to be removed? What risk remains? Who assessed it? Those questions slow everything down.

At scale, Integral should make those questions feel much simpler.

The idea is that when Integral is involved, sellers know they are not giving away secrets, buyers know they are not receiving radioactive data, and both sides trust the privacy, quality, and compliance integrity of the exchange.

That makes more of the data supply chain usable. More companies can activate valuable proprietary datasets, and AI builders can access the real-world signal they need to build better models and products. Our goal is for Integral to be the independent layer that enables more sensitive data because the trust is efficient and high quality. A determination or defensibility opinion the exchange can stand behind. Not because privacy disappears, but because the risk assessment, transformation, and defensibility work are built into the infrastructure.

If we get that right, the data economy becomes more trusted, more active, and more useful.

What stood out, what’s tricky, and why it matters…

Integral is tackling one of the less flashy but increasingly important parts of the AI stack: how sensitive, proprietary data actually moves.

The first generation of AI was built on public data, open internet data, and human-curated datasets. That was enough to create powerful general models. But the next wave is different. If AI is going to work inside healthcare, finance, enterprise operations, life sciences, customer support, codebases, and other real-world settings, it needs data that captures how those worlds actually operate.

That data is usually not public. It is sitting inside health records, claims files, pharmacy transactions, CRM systems, support logs, operational tools, financial systems, and other proprietary environments. It is valuable because it is messy, longitudinal, and real. It is also hard to use because it carries privacy risk, contractual restrictions, regulatory obligations, trade secrets, and IP concerns.

This is where Integral fits. The company is not selling data. It is trying to become the independent privacy and compliance infrastructure that lets data move safely between sellers and buyers. In healthcare, that means helping pharma companies, payers, data providers, labs, and AI teams use sensitive data in a way that preserves utility while reducing risk. Outside healthcare, the same playbook applies to any proprietary dataset that needs to be activated for AI.

Integral recently announced an $18M Series A ($25M total raised) and the funding will help accelerate its Forward Deployed Privacy Services, which embed privacy engineers, statisticians, software engineers, and methodologists directly into customer data pipelines.

What stood out in the conversation was Shubh’s point that healthcare was not just a starting market. It was training on hard mode. HIPAA forced Integral to build around real regulatory scrutiny, high-stakes privacy risk, and data that has always been proprietary. As the rest of the world starts to look more like healthcare from a data perspective, that foundation becomes more relevant.

The company’s broader bet is that proprietary data becomes one of the key bottlenecks for AI. Models are proliferating. Compute is being bought up. Public data has already been mined. The edge increasingly comes from the private, real-world datasets that capture how people, companies, and systems actually behave. If Integral can become the trusted layer that makes those datasets usable, this becomes a much bigger infrastructure story.

What stood out

What’s tricky

Final thoughts

Integral is building in a part of AI that probably deserves more attention.

A lot of the AI conversation still focuses on models and compute. But as general models become more available, the next bottleneck is the data that lets those models become useful in specific industries. In healthcare, that means longitudinal patient journeys, claims patterns, pharmacy behavior, demographics, care access, and the many messy signals that do not live in one clean system.

The challenge is that this data is valuable precisely because it is sensitive. You cannot just scrape it, dump it into a model, and hope for the best. Someone has to make it usable, defensible, and safe enough to move. That is the layer Integral is trying to own.

The healthcare roots matter here. Healthcare has been dealing with proprietary, regulated, high-stakes data for a long time. In a strange way, the rest of the AI economy may now be catching up to healthcare’s data problem. Enterprises in every industry are realizing they have valuable data exhaust, but they need a way to monetize or activate it without creating new privacy, legal, or commercial risk.

If Integral works, the company becomes more than a privacy vendor. It becomes part of the trust infrastructure for the real-world data economy: the layer that lets sellers safely monetize proprietary datasets, buyers access high-quality signal, and AI builders move beyond public data into the messy domains where the next generation of models will need to operate.

That is a big swing, but the timing makes sense. AI needs more real-world data. Enterprises want new revenue streams. Regulators and customers are asking harder questions about provenance, risk, and defensibility. Integral sits right in the middle of that shift.