Subscribe | Advertise | PRO Member Hub

TOGETHER WITH VI

Perspectives from the people building the future of health AI…

We sat down with Omri Yoffe, CEO and Founder of Vi, and Spencer Honeyman, President of Global Growth, to unpack how the company is building what it calls the AI execution layer for healthcare and life sciences.

Vi works with large enterprises across health plans, health systems, pharma, biotech, and wellness. Its platform combines a clinical and behavioral Data Web with specialized models and agents that sit on top of the systems customers already use, including EHRs, CRMs, cloud infrastructure, and other enterprise tools. Rather than replacing those systems, Vi is designed to make them more predictive and turn their data into next-best actions.

Vi’s core applications are Activate, Engage, and Operate. Activate helps identify and reach the right patients or physicians, Engage determines how and when to interact with them, and Operate applies agents to workflows like care navigation, patient communications, research, and operational optimization. Across 100+ enterprise customers, Vi says it supports more than 190M lives and orchestrates roughly 500M interactions each day.

A big part of Vi’s approach is accountability. The company works with customers to underwrite the business case before deployment, measures results against control groups, and structures pricing around the value created. Spencer said more than 90% of Vi pilots convert to scaled deployments, while the company says its platform has generated more than $2B in measurable value for partners and helped support the development and commercialization of 50+ drugs.

Vi recently completed a $145M transaction at a $1.64B valuation to further invest in the platform, new products, and talent while strengthening its balance sheet.

Omri and Spencer shared how Vi evolved from its roots in predictive health into a broader enterprise AI platform, why proprietary data may matter more as frontier models commoditize, what “owning the outcome” means in practice, and where they think agents can have the biggest impact across care navigation, clinical trials, and the broader health enterprise.

Let’s start from the top. What led you to start Vi?
Omri: My path into healthcare was unusual because I came from aerospace.

We lost one of our best friends in the Air Force. He was flying an F-16, pulling high Gs, passed out, and crashed. That experience made precision and predictive health very real to us. We started building biosensing technology that could extract signals like heart rate, oxygen saturation, motion, and perfusion in high-risk environments, and eventually sold that technology for use in advanced fighter helmets.

That exposed me to a much larger problem. Healthcare still struggles to connect the dots for the people it is supposed to protect. Patients miss the right next action, families struggle to get the right second opinion, and drugs that science has already made possible take too long to reach patients.

We experimented with direct-to-consumer healthcare, but eventually realized the biggest impact would come through enterprises. Health plans, providers, pharma companies, and other large organizations already touch millions of people. The mission became to make health more precise, predictive, and efficient, and to be accountable for whether we actually create that impact.

For readers hearing about Vi for the first time, what does the company do today?
Omri: Our broader mission is what we call health abundance in our lifetime. We believe every person should ultimately have access to precise, predictive, and affordable care.

The way we execute on that is by becoming the AI execution layer for healthcare and life sciences. We help enterprises turn their data into next-best actions, whether that means accelerating a clinical trial, identifying the right action for a patient or physician, improving care navigation, or making an operational workflow more efficient.

There are three things underneath that. First is the Data Web, which combines clinical and consumer data. Second is specialized agents and models that deploy on top of the systems customers already use, whether that is Epic, Veeva, Snowflake, Databricks, AWS, or something homegrown. Third is accountability. We underwrite the outcome, prove it through a control group and pilot, and then scale if it works.

How do Activate, Engage, and Operate work across healthcare and life sciences?
Omri: Take pharma as an example.

With Vi Activate, we can connect into the company’s systems, understand the disease and trial criteria, and use the Data Web and models to identify and activate patients who are more likely to qualify for a clinical trial. The goal is to find those needle-in-the-haystack patients faster and accelerate enrollment.

Once a drug is approved, Vi Engage helps determine the next-best actions for physicians and patients. Which HCP should receive which information? Which patient should receive which intervention? Through what channel, and when?

Then Vi Operate applies agents to the internal work of running the enterprise, including patient communications, supply chain forecasting, clinical analysis, research, and other operational workflows.

Healthcare follows a similar pattern. A chronic care organization might use Activate to identify eligible patients, Engage to improve enrollment and utilization, and Operate for areas like care navigation or case management. The common thread is moving from identification to engagement to execution.

Health enterprises already have huge technology stacks. How do you avoid becoming one more tool?
Spencer: We learned pretty early that trying to replace major systems makes everything much harder from a change management perspective.

Our goal is to be the “Intel inside.” Our customers already have existing technology and EHR stacks, like Epic, Veeva, internal risk engines, CRMs, outreach teams, and clinical programs. We are not trying to rip those out. We sit behind or between those systems and make them more proactive, precise, and personalized.

One of the most common objections is, “I already do something for predictive AI today.” Often our answer is: great. Here’s how Vi is different and how, together, we’ll deliver better results. The important question is whether Vi can make the existing system smarter without requiring the organization to rebuild its technology stack or operating model, whether that’s accelerating a clinical trial, capturing more specialty demand for a health system, or driving script lift in life sciences.

The Data Web seems central to Vi. What is it, and why is it difficult to replicate?
Omri: The Data Web has three main components.

The first is first-party data accumulated through years of enterprise relationships, representing more than 190 million de-identified patients and members across the platform. The second is longitudinal clinical data across areas like claims, prescriptions, eligibility, EHR and EMR data, and other clinical signals.

The third piece is consumer behavior, what we sometimes call understanding the “other 23 hours.” That includes socioeconomic signals, web behavior, mobility, and other patterns outside the traditional healthcare system.

When you combine those types of information, you can better understand what intervention is likely to work for a particular population and where to meet people.

The hard part to replicate is not simply buying another dataset. It took us eight or nine years to build the first-party relationships, contractual structures, data infrastructure, and expertise required to make that information usable. We also have a dedicated team whose entire job is expanding the Data Web and bringing in new signals.

Can you share an example of the kind of insight the Data Web can uncover?
Spencer: One of my favorite examples comes from a regional pediatric autism clinic we work with that was looking to capture more market demand.

They treat children with more advanced forms of autism through an intensive program. We were looking at signals that could help connect families that might benefit. Some were fairly intuitive, like searches around managing outbursts or purchases of over-ear headphones associated with sensory overload.

But we also found that families in this population were visiting trampoline parks and similar entertainment centers at more than 10x the rate of the general population. We initially thought it might be a bug. When we showed it to the clinic’s CEO, he immediately understood it. For some children with autism, intense physical activity can be an important form of sensory regulation.

That is only one signal among roughly 70,000 data points Vi leverages across more than 96% of U.S. households, but it is a good example of why broader behavioral data can matter. Sometimes the useful signal is something you would never think to put into a traditional clinical model.

What problem are healthcare clients usually trying to solve when they come to Vi?
Spencer: Vi serves both health plans and health systems. Health systems use Vi to capture proactive signals and reduce leakage among existing patients who do not have a clear referral pathway within their system. Today, they can be blind to that demand.

Health plans already have sophisticated systems using medical data to predict what care members may need. The challenge is that much of that data is lagged. By the time a claim or other medical signal shows up, the opportunity to intervene and prevent avoidable costs or improve health outcomes may already have passed.

We augment their existing next-best-action engines with near real-time behavioral intent models. If we see a cluster of de-identified signals consistent with someone potentially heading toward a spinal procedure, for example, we can return a propensity flag for action to the health plan. They can then check in and potentially offer a second opinion, physical therapy, a digital MSK program, or another intervention before a potentially preventable or suboptimal health event occurs.

For one client with roughly 600,000 members, about 9% ultimately needed a procedure over six months. We were able to predict that group around 68 days before the first medical signal appeared, at roughly 94% accuracy. The point is not that those procedures were inappropriate. It is giving the plan time to intervene when there may be a better path.

Lots of companies are pitching themselves as enterprise AI platforms. Where do you think Vi’s moat actually comes from?
Omri: I do not think there is one magical technology that nobody else could ever build. The moat is the combination.

You start with the data edge and the years of enterprise relationships required to build it. Then you need technically strong teams that can deploy specialized agents and models inside large healthcare organizations. And finally, you need a business model that forces everyone to care about whether the customer actually gets the result.

That last piece matters. Our commercial teams are not rewarded simply for signing a customer. We need to identify the right use case, underwrite it properly, prove the value, and then expand.

Any one of those pieces can be copied. Combining the data, vertical expertise, deployment capability, and outcome accountability is much harder.

What does the technology stack look like under the hood?
Omri: At the bottom is the infrastructure layer, which for us is largely built on AWS with a sophisticated AI Ops and ML Ops environment. That allows us to work with a mix of open-source and open-weight models, our own specialized models, and frontier models from companies like OpenAI and Anthropic where appropriate.

Above that sits the Vi platform. The Data Web is one part of it, along with inference, harnessing, memory, tokenization, model and agent feature stores, and integration tooling. Memory is especially important because the system needs to understand the context and value created over time rather than treating every interaction independently.

We have also built a large integration library across systems like Epic, Veeva, Salesforce, CRMs, and other enterprise platforms. On top of that sit Activate, Engage, Operate, and Pulse, supported by technical squads that combine engineering depth with enough customer understanding to deploy these systems in practice.

AI has a notorious pilot graveyard. How do you structure pilots and pricing so they actually make it into production?
Spencer: We underwrite every client from pre-launch. We treat the pilot as a validation period, not an experiment where we figure out the business case afterward.

We start with the KPI that actually matters. For a health plan, that could be cost of care. For a health system, it could be enhanced specialty demand capture. For pharma, it might be clinical trial acceleration or script lift. We usually take what we have achieved elsewhere, cut it in half, and ask whether even that result would create enough value to matter. If the answer is no, we walk away.

If the business case works, we agree on the economics and path to scale upfront. We maintain a representative control group, the customer measures the incremental result, and if we hit the agreed outcome, the deployment scales. That structure is why more than 90% of our pilots convert to scale. We are not finishing the pilot and then starting a new procurement process or asking what the results mean.

At full scale, we generally map our economics to roughly a 5:1 ROI for the customer. That can be a fixed licensing fee derived from expected value, value-tiered pricing based on the level of impact, or a value share tied to the incremental value created. The structure varies, but the principle is the same: we want our economics aligned with the outcome we actually produce.

What separates customers that get significant value from the ones that stall?
Spencer: The biggest thing is starting with a problem where our data and vertical models are the best solution, instead of starting with AI as a solution looking for a problem.

There is so much excitement around AI that some organizations effectively have AI budgets and then go looking for places to use them. We think about it the opposite way. If the best solution is pen and paper, a carrier pigeon, or the most advanced AI system in the world, first agree on the problem and the outcome you are trying to drive, then have a clear rationale for why AI is needed to get the result.

The second piece is collaboration with the internal team. We work in healthcare and life sciences, and the workflows are sometimes life or death. You cannot drop a model into a complex enterprise and expect it to immediately operate at full optimization. There is a glide path and a lot of iterative feedback.

If the dynamic becomes Vi versus the internal team, the client is likely to struggle. If it is Vi plus the internal team, the results tend to be much stronger.

Where do you think AI agents will have the biggest impact across healthcare and life sciences?
Omri: Two areas stand out to me.

The first is care navigation and case management. There is an enormous amount of human labor involved in figuring out what a patient needs next, helping them navigate the system, coordinating those steps, and making sure they actually happen. Agents can make that process much more predictive, precise, and scalable.

The second is getting molecules to market. Anything that can accelerate patient identification, clinical trials, risk analysis, and ultimately the path from a promising drug to the patients who need it can fundamentally change healthcare.

Those are the two areas where I think agents can create some of the biggest real-world impact.

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

Vi is tackling a familiar enterprise healthcare problem: organizations already have enormous amounts of data, software, and domain expertise, but that information does not consistently turn into timely action.

That helps explain where the company sits in the stack. Vi connects into the systems its customers already use, combines enterprise data with its broader Data Web, applies specialized models and agents, and pushes the resulting actions back into existing workflows. Depending on the customer, that could mean finding patients for a clinical trial, deciding which physician should receive which intervention, identifying a member who may benefit from earlier outreach, or automating an internal operational process.

The Data Web is probably the most distinctive piece. Vi combines longitudinal clinical information with behavioral and consumer signals covering what the team calls the “other 23 hours.” Claims and EHR data tell you a lot about what has already happened inside healthcare. Search behavior, mobility, socioeconomic signals, and other patterns can sometimes surface intent earlier. In one health plan example Spencer shared, Vi identified members who would ultimately need a procedure roughly 68 days before the first medical signal appeared, at around 94% accuracy.

Vi has also built a disciplined path from pilot to production. The team starts with a specific KPI, underwrites the expected value, establishes a control group, and defines the economics of scaling before the proof of value begins. Spencer says more than 90% of pilots move into scaled deployments. That process gives Vi a clear filter for enterprise AI projects: if moving the KPI does not create enough value to matter, the project shouldn’t start.

As frontier models become easier for enterprises to access, more of the differentiation shifts to proprietary data, integration, deployment, vertical expertise, and the ability to turn predictions into measurable results. Vi has spent years building around that combination.

There is a useful idea underneath Vi’s “health abundance” language. Healthcare already has an enormous amount of science, treatment capability, data, and infrastructure. A meaningful part of the remaining gap is execution: recognizing the right patient, choosing the right intervention, reaching them early enough, and making sure the action actually happens.

Vi has spent years building around that layer. The Data Web helps it see signals that traditional healthcare systems may miss, while its models and agents sit on top of the customer’s existing stack and push those signals toward action. The outcome-based commercial model then puts a fairly hard test around whether any of it worked.

The company also seems appropriately unsentimental about models. Omri does not argue that Vi needs to own every model or that one proprietary algorithm creates the moat. As foundation models improve, more of the valuable layer may shift toward data, integrations, domain-specific infrastructure, and the ability to deploy AI against a business or clinical outcome at enterprise scale.

The difficult part is maintaining that advantage across a very broad set of use cases without becoming overly custom, while also handling the governance burden that comes with increasingly rich behavioral data.

If Vi can keep proving that its Data Web produces meaningfully better decisions and that those decisions reliably translate into measurable action, it has a credible path to becoming an important execution layer across health enterprises. The ambition is broad, but the company has built a fairly disciplined mechanism for testing whether that ambition produces results, and the high rate of pilots moving into scaled deployments suggests customers are seeing the value to keep expanding.