From a beachhead in healthcare, firm aims to apply its venture studio model and own pool of data scientists to more industries
Navid Alipour is co-founder and managing partner at Analytics Ventures, a global venture studio providing end to end infrastructure and to plan, form, launch, fund, and validate brand new ventures in Artificial Intelligence. Navid has co-founded and invested in multiple AI based companies including CureMetrix and CureMatch. He looks to identify top tier scientists, academics or corporate partners, and then to work with them to turn their R&D into viable businesses. He recently took a few minutes to speak to AI Trends Editor John P. Desmond about his work.
AI Trends: Thank you for talking to us today and to the readers of AI Trends, Navid. Tell me about your firm and how you got involved with artificial intelligence.
Navid Alipour: Analytics Ventures was started about six years ago. I was a co-founder along with my French partner and my German partner. We invested our own capital for the first four years, then we were approached by the head at the plasma astrophysics department at UC San Diego. He is one of the world’s top experts in AI. We ended up starting our first company, CureMetrix, under the venture studio model. We also started our own AI lab and raising funds.
How early stage a company do you get involved with, and how do you decide if they are actually using AI?
Per the venture studio model, we literally are as early as possible, in that we help incorporate and launch and found the companies, with other domain experts. We’re industry-agnostic. In any industry, you can apply AI and machine learning to unlock value from the data. We look for the domain experts, scientists, academics, entrepreneurs, or corporations increasingly, to start joint ventures with.
In regards to vetting out the technology, we formed our own AI lab with our head scientist. We give a technical test to anyone we hire. You’d be shocked at whose resumes we’ve seen that have not passed our test. So any scientist we hire does truly have an AI background, and not just using glorified business intelligence. And because we start the companies, the other co-founders don’t need to have to have an AI background. They just need the domain expertise to say how they would want to apply AI to the data to make a recommendation or prediction or forecast, or detect anomalies that don’t belong. And then our guys on the team can build the AI model.
Can you describe AI Labs?
After we started CureMetrix and we had hired good data scientists, and we wanted to start new companies in different verticals–agtech, fintech, education and healthcare–we did not want to hire data scientists for every one. So we decided to form our own AI lab and attract scientists to work on current and future companies. It’s one of our competitive advantages as a firm that we have this pool of data scientists to help build the AI applications.
What are some of the specific challenges you see around applying AI in healthcare?
One challenge is around who controls the data. We have HIPAA privacy laws that we have to be very cognizant of, and the hospitals have to be cognizant of, and the doctors so they don’t violate those laws. Then we have to marry the domain expertise with the AI experts. Then we have the regulatory environment and the business model, and whether it needs FDA approval. Even if something is all software, if it’s a diagnostic per the FDA, it’s considered a medical device. So then you have to go through FDA approval, and that could take several years.
You could have an amazing piece of software that works and is using AI to detect cancer, but if you don’t get that FDA approval, you can’t sell the product. And so, you have to spend the time and money and years to go through those hurdles to get that FDA approval. So healthcare is definitely a more challenging industry because of those regulations. There are reasons for them to be there, and it has its own set of challenges.
Are the companies you invest in able to get access to the data they need to do their work?
Within healthcare, it takes a bit of time. CureMetrix has millions of mammogram images that they trained algorithms on, but it took a lot of time. You have to go to those hospitals and imaging centers, show them the value of your technology and then sign the partnership agreements. It’s a chicken or egg thing. How can you train the algorithms if you don’t have the images, the datasets? So that continues to be a challenge, and that’s why there’s value to these companies, just based on the images they have in their database to train the machine learning algorithms.
And then on CureMatch, our other healthcare portfolio company that’s a decision support tool for oncologists, you have to collect the 31-page PDF that is produced by Foundation Medicine or Guardant Health or another company that sequences the cancer biopsy. You have to work with the oncologist, not just the patients, to get that information and to then use the algorithms to recommend the top combination of drugs out of four and a half million combinations.
That sounds complicated.
It is. It takes a lot of really smart doctors and scientists to do the work. It’s a perfect example of the convergence of the life science and software industries. And so you have doctors and PhDs that need to work with the data scientists and software engineers and developers to be able to build a company that can do that. You’re not just building a dating app or an app to deliver pizza.
You said CureMetrix enters into partnerships with hospitals to get the data they need. How does CureMatch get the data it needs?
CureMatch gets the data from the oncologist, who gets it from the patient. Because it’s a software company, we can’t practice medicine. So we provide our recommendation to the oncologist, who uses CureMatch as a tool in the fight against cancer. Oftentimes, it’s the patient that finds us, but then we have to work with the oncologist.
In the US, what’s crazy is that oncologists benefit and make money off selling or administering chemotherapy. This is illegal in Europe and a lot of other parts of the world. While oncologists should have the patient’s best interest in mind, if they’re not precision medicine experts, they just kind of stick to the standard protocol and what they learned in medical school. Even if 98% of the patients of a particular cancer like stomach or liver cancer die, they say, “Well, that was the standard protocol. These were the drugs that I was trained on in medical school or they discuss at conferences to give.” I think that needs to change from the regulatory level on down, to say that before any drugs or any chemo or any treatment is given, to get that sequenced biopsy information and use a tool like CureMatch to say, “Hey, for this patient’s specific cancer, these are the top three drugs out of four and a half million combinations,” because no human brain can process that.
Why is this illegal in Europe?
It doesn’t make sense that oncologists should profit from giving drugs to their patients. If you’re a cardiologist, you shouldn’t make money off giving more prescriptions for blood pressure medicine. No other doctor, no other specialization profits from administering drugs. It’s only oncology, to my understanding. We do feel like this needs to change.
So you’re saying with your tool, with CureMatch, the doctor gets a list based on the patient’s specific diagnosis of what the most successful drugs are, so that’s what should be used. Is that the case you’re making?
Yes.That’s accurate. Look at it this way, if you were sick today, you go to the doctor, they order some lab work. You go get some lab work done. They send that to your doctor, and then your doctor calls you or sees you and says, “This is what’s wrong with you, and because of that, we need to give you these prescriptions or these medications.” You don’t just go to the doctor and they look at you and say, “Okay, we’re going to do XYZ.” But with oncology, if you have, let’s say, stomach cancer or CLL, leukemia or lymphoma, you go to the doctor and once they determine you have that, your primary care sends you to an oncologist, and they then say, “Okay, come back next week. We’re going to put you on chemotherapy.” And that’s wrong.
In this day and age, companies like Foundation Medicine or Guardant Health will take a physical sample of that cancer tumor, or if it’s a blood cancer, a liquid biopsy. They sequence it, and they put out what I like to call the 23andMe or the Ancestry.com of that specific cancer. So it’s a 31-page report from Foundation Medicine, for example, that is the genetic-sequence information of that specific person’s cancer.
CureMatch will take that PDF produced by Foundation Medicine and we will match it to our database that includes all the FDA-approved drugs. We’re just software. We don’t touch any physical sample. We match and say if the patient wants a three-drug combo based on this patient’s specific sequenced cancer information and the drugs available, out of four and a half million combinations, it produces a report with the top three recommended drugs.
The oncologist looks at that and either agrees with it, or makes another decision. While people think of cancer as based on what body part, stomach cancer, liver, lung, breast, there’s only just cancer. And if you take that genomic information, you might find that the cancer in the stomach is actually very similar to a lung cancer. We had a particular case where the recommendation was two drugs plus KEYTRUDA, which is a Merck drug, for lung cancer. The patient who had the stomach cancer said, “I would have never thought of that, but that makes sense.” And so that’s the power in what CureMatch is doing with using AI.
So, once the oncologist gets the recommendation from CureMatch, they still sell the drugs, correct? The software doesn’t change that part?
No, but they at least then, they have a scientific reason for why they’re giving which drug to which patients, or which drugs, instead of just pushing the chemo because they have it in the office because they bought it.
And you’re saying a human doctor cannot process 4.5 million combinations that your system considers.
Exactly. No human brain can process that. I don’t care how smart someone is. So AI is not meant to replace the doctor. It’s meant to empower them so they can do their job better. Just like Excel did not replace your CPA or my CPA. If anything, there’s more CPAs around. But it empowers them to do their job better and more efficiently. And that’s the same with AI in healthcare. It’s not meant to replace the oncologist.
And on the other side with CureMetrix and breast cancer, it’s not meant to replace the radiologist. It’s meant to help them do their job better because the machine doesn’t get tired. It doesn’t need a coffee break. It doesn’t have a food coma after lunch. It doesn’t get distracted. And by the way, there’s a shortage of doctors, and so they’re not going away, nor is that the purpose of AI. It’s meant to empower them and to help patients get better outcomes.
Is AI delivering? What is the payback?
That’s definitely the multi-billion dollar question. The payback is huge. As AI is applied to different industries to increase revenues or decrease cost by bringing operational efficiencies, and as Mark Cuban was quoted on saying on CNBC a couple years ago, “If you’re not thinking of how to apply AI and machine learning to your respective business, you know, you’re going to go the way of Barnes & Noble and Blockbuster Video and disappear,” just like the dinosaurs.
We agree with the premise that every industry and every professional should be thinking of how AI could be applied to remain competitive.
It’s both an opportunity and a scary proposition.
And there’s an ethical part to it, too, because some jobs will disappear. But you can’t stop innovation. AI is a tool. There will be bad applications of AI; some will use it to cause harm. But we feel the good is going to far outweigh the bad, and we just need to have the law catch up with technology.
Are there any other companies you are involved in that you want to highlight?
Dynam.AI is our AI as a service company. Where we initially built the lab and hired data scientists just for ourselves, we ended up being contacted by other VC firms and private equity firms and corporations to help them on due diligence around AI investments. We saw a demand, so we spun off Dynam.AI as our AI-as-a-service company. We now have Fortune 500 clients that leverage our AI bench and to help them apply AI to their data or to do due diligence on investments they want to make.
And Kazuhm is our company that recaptures unused compute capacity from desktops and servers and helps enterprises reduce their cloud spend by as much as 90%. So we help companies to reduce their cloud spend and their carbon footprints, because not as many data centers have to be used.
Kazuhm has been around for two and a half years. They were heads down, building the technology for about two years, and now they are live and have a product and are selling. One statistic that blew us away is that over 70% of the capacity of a computer is not being used. You’re never using the full compute capacity of your computer. And at the enterprise level, every few years they have a refresh cycle where the CFO has to spend the money to buy new computers, and they didn’t even use 70% of what they had. So that’s where Kazuhm helps recapture that unused compute capacity, and reduce cloud spend for companies.
What are the challenges you see to a successful adoption of AI in companies, in general?
One of the key challenges is having clean, structured data to train the algorithms, because otherwise, as they say, “It’s garbage in, garbage out.” So having access to clean, structured data is critical. And then of course, then having the talent, the scientists to develop the algorithms. Many open source tools are good for ABC but not XYZ and so to truly optimize, to solve a problem with an AI solution, you need to custom-build the algorithms for the specific problem. That takes having the right talent.
Are you able to find and hire the people you need to help the companies that you get involved with?
To date we’ve been very fortunate; the data scientists love working at our lab and the companies we’ve created. As I mentioned earlier, we have a high bar for the technical test. What we found is good data scientists want to work with other AI players, and for them, it’s not just about the money. They’re not going to go to another company just for more money. They want to solve problems. Of course, they’re not allergic to money. They want to be compensated, but they’re not going to make a change, the good ones, just for monetary purposes.
We found in corporate America, or corporations at large globally, it’s hard to attract the top data scientists long term, because if you’re a good data scientist, and you go home to your spouse, do you really want to say you work for a soda manufacturer or a motorcycle manufacturer or a medical device company? That doesn’t excite them. Innovation often happens outside of large companies, and that pattern is continuing with AI development. That is why we are deploying this venture studio model where we look to work with corporations to identify ways to unlock value from their data by applying AI. And if it makes sense, we do it internally for them, and that’s where Dynam.AI works with the corporate Fortune 500 on down. And sometimes, it makes sense to do a joint venture and to spin out a company. And so, we’re always looking for those opportunities.
Do you have any advice for young people trying to get into the field of AI? Whether they’re college students or early career people, what would you advise them to either study or focus on?
One thing we really advocate for is at the high school level on up, there should be a data science 101 class for everyone. Just like when I was in high school, I had to take wood shop. I’m certainly not a carpenter. There are many online classes, the Courseras of the world, where someone could get educated. Obviously to be a data scientist, you have to start early and a focus on physics and math is critical. And then, there are obviously classes and whole programs being formed around data science.
An important point to make is also not everyone has to be a data scientist and write algorithms to be an AI expert. There are data scientists that develop algorithms. Then there are the engineers that productize those algorithms to build a piece of software. And then there are strategists and consultants that look for the opportunities and problems to solve with the AI. That does not require a technical background. It requires someone to understand what an AI tool or machine learning applications can do, and identify the business purposes where it could be deployed. And so, whether you’re a lawyer or a CPA or a real estate agent or a consultant, having an understanding of AI not only impacts your own business, in a competitive sense, but also being able to better serve your clients.
Learn more at Analytics Ventures.