Ten Principles for Building a Health or Life Sciences Venture

Most new ventures in healthcare and life sciences fail for reasons that were visible early. Ten principles help founders, investors, and directors test a young company before the capital is committed, and the same ten apply to every AI program a board approves.

In Brief

The Problem. New ventures in healthcare and life sciences carry commercial, clinical, regulatory, and data risk at the same time, and the failures that follow were usually visible at the first serious review.

Why It Persists. Founders, investors, and boards judge the idea, the team, and the market as one bet, rather than testing the technology, the user value, the business model, and the dependencies separately and in sequence.

What Boards Should Do. Use ten principles as a review discipline for any new venture, and apply the same discipline to internal AI programs, which carry the same risks as a startup inside an established organization.

Why Principles Matter More in Health and Life Sciences

The success of a new business in healthcare, healthcare analytics, or life sciences depends on its ability to adapt in a complex and often unpredictable environment. The stakes are higher than in most industries, because patients' lives and well-being are in the balance. A venture needs a small set of guiding principles that keep its decisions about value, resources, and direction honest.

The ten principles below are not new. They are the questions I would expect a board or an investment committee to ask before each round of funding: validation, durable competitive barriers, staged risk, efficient customer acquisition, realistic projections, objective evaluation, customer engagement, simplicity, managed dependencies, and the balance between intellectual property and early entry. My argument is that they belong on more agendas than the startup's. Every AI program that an established health system or life sciences company approves is a small venture inside the organization: unproven in its setting, dependent on a vendor, and expected to pay for itself. Directors who would test a startup this way should test their own AI approvals the same way.

When the first version of this article appeared in January 2023, several of its examples were presented as successes. Some have since failed or changed hands, and I have kept them, corrected, because they teach more than the original versions did.

Validate the Technology, the User Value, and the Business Model

A venture must prove three things separately. The technology must work. It must meet a real, unmet need for the people who use it. Someone must pay for it on terms that sustain the business, which usually means economies of scale, healthy margins, and a large or growing market. Each needs its own evidence, and success on the first two does not prove the third.

23andMe was once cited, including in the original version of this article, as a company that had validated its business model by giving consumers useful genetic information at scale in a large market. In March 2025, it filed for Chapter 11 protection to begin a court-supervised sale of its assets (Mehta, 2025). On July 14, 2025, TTAM Research Institute, a nonprofit affiliated with co-founder Anne Wojcicki, completed the purchase of substantially all of its assets for USD 305.0 million in cash plus assumed liabilities (23andMe Holding Co., 2025). Consumer demand had been validated. A sustainable business model had not.

Build Barriers That Competitors Cannot Easily Cross

A new business should build a competitive advantage that is difficult for others to overcome. It can come from proprietary technology, from data that others cannot assemble, from regulatory clearances, or from deep integration into a customer's operations. The barrier should be specific and testable. "We are first" is not a barrier.

Cerner, the electronic health record company in the original article, built its position through proprietary technology embedded in hospital operations. That position also made it a target. Oracle completed its acquisition of Cerner in June 2022 (Oracle, 2022). Founders should decide early whether they are building to remain independent or to be acquired, because the two paths call for different choices about data, contracts, and capital.

Stage Milestones and Financing Around Risk and Uncertainty

Risk and uncertainty are not the same. A risk can be named and tested. An uncertainty cannot yet be measured. A well-run venture stages its milestones so that each round of financing retires a specific risk before the next commitment, which allocates capital more effectively and makes each decision easier to defend.

Flatiron Health illustrates staging from the investor's side. Roche had worked with Flatiron for two years and held 12.6 percent of its equity before it agreed, in February 2018, to acquire the company for USD 1.9 billion on a fully diluted basis (Roche, 2018). The buyer tested the value of oncology real-world data through a partnership and a minority stake before it committed to ownership.

The same discipline applies inside established organizations. A board approving a new AI system can approve it in stages and attach conditions that management must meet before each expansion: a validation result, a monitoring report, or a defined error rate.

Prefer Models With Low Acquisition Cost and High Lifetime Value

The cost to acquire a customer (CAC) should be well below the lifetime value of a customer (LTV). A model with that profile supports higher profitability and growth that the business can sustain. In healthcare, the calculation is harder than in consumer markets because the customer is often not the user. A hospital buys, a clinician uses, and a payer may decide whether the service is reimbursed. Each of these parties belongs in the calculation, and a venture that wins the user but not the payer has not yet found its customer.

Project Realistically

Cost estimates, milestone dates, prices, and adoption rates should rest on evidence, not hope. Sales cycles in health systems are long, and regulatory timelines move. Overly optimistic projections do more than miss targets. They consume the cash a venture needs to survive the delay, and they distort the price of acquisitions made on the strength of them.

Teladoc Health completed its USD 18.5 billion acquisition of the chronic-care company Livongo in late 2020. For 2022, Teladoc reported a net loss of USD 13.7 billion, driven by a non-cash goodwill impairment of USD 13.4 billion (Pifer Parduhn, 2023). An impairment of that size is an accounting statement that the projections behind the price did not hold.

Build Objective Evaluation Into Governance

A venture should review its performance and strategy on a set schedule, identify what is not working, and change or stop it. This takes courage and objectivity as much as analysis. Someone in the room should be free to say the plan is not working. A small board with at least one independent director usually serves a young company better than a board of investors alone, because investors who led the last round have a stake in the last decision.

A decision to stop is a legitimate outcome of an objective review. On April 30, 2024, Walmart announced that it would close all 51 of its Walmart Health centers in five states, stating that "escalating operating costs create a lack of profitability that makes the care business unsustainable for us at this time" (Price, 2024). The company reviewed its results and stopped rather than continue to fund a model that did not work.

Engage the Customer Early and Often

Understanding and addressing customer needs is essential. Early users who find fault with a product are a source of information, not a problem to manage. Early engagement with clinicians, patients, and purchasing staff shapes the product and shortens the sales cycle. It also surfaces safety and workflow problems while they are still inexpensive to fix.

The Permanente Medical Group applied this discipline to an internal AI program. When it enabled an ambient AI scribe for 10,000 physicians and staff in October 2023, it paired the rollout with peer support at the point of care, patient-facing materials, rapid-cycle upgrades with the vendor, and ongoing monitoring. Within ten weeks, 3,442 physicians had used the tool in as many as 303,266 patient encounters (Tierney et al., 2024).

Keep the Business Simple

Complexity is a cost. A business model that a director, an investor, or a hospital buyer can grasp in one conversation is easier to finance, sell, and govern. Complexity in the product may be necessary. Complexity in the business model rarely is.

Zipline shows how a plain model can carry a difficult technology. On October 14, 2016, Rwanda launched a national drone delivery service with Zipline that delivered blood for emergency transfusions to 21 facilities in the western half of the country, filling orders in around 30 minutes (Gavi, 2016). The engineering was complex. The proposition was one sentence long.

Manage Dependence on Third Parties

Dependence on a single supplier, platform, data source, or distribution partner limits a venture's control over its costs, its innovation, and its direction. The original version of this article said that CRISPR Therapeutics had minimized its dependencies. The record shows a more useful lesson: it chose a dependency deliberately and governed it by contract. In April 2021, CRISPR Therapeutics and Vertex Pharmaceuticals amended their collaboration on the sickle cell and beta thalassemia program. Vertex took the lead on global development, manufacturing, and commercialization; CRISPR Therapeutics received USD 900 million up front and kept a 40 percent share of program costs and profits (Vertex Pharmaceuticals, 2021). On December 8, 2023, the resulting therapy, Casgevy, became the first approved treatment to use CRISPR gene editing (Berkowitz & Simkin, 2023).

The cost of a dependency that no one has governed became visible across the industry in February 2024, when a cyberattack took down Change Healthcare, a UnitedHealth Group subsidiary that handles about 15 billion health care transactions a year. In a survey of nearly 1,000 hospitals, 94 percent reported a financial impact, 74 percent reported a direct impact on patient care, and 67 percent said that switching to another clearinghouse was difficult or very difficult (American Hospital Association, 2024).

Some dependence is unavoidable, so it should be chosen and written into the contract. For AI products, the terms that matter most are use of the customer's data for training, notice of material changes to the model, audit rights, and indemnification.

Protect Intellectual Property, but Do Not Wait on It

Intellectual property (IP) protection matters, especially in therapeutics and devices. It is not a substitute for reaching the market. Early entry can establish relationships, evidence, and loyalty that a later competitor will find hard to match. Gilead Sciences' hepatitis C drug Sovaldi, approved by the Food and Drug Administration (FDA) in December 2013, was described at approval as the first pill to treat certain types of hepatitis C without injected interferon, at a list price of USD 84,000 for a 12-week course (Pharma Manufacturing, 2013). Being first is still no guarantee. The venture must also be right, and its price must be one that payers will sustain.

Applying the Principles to AI Approvals

A board that approves an AI program is making a venture decision. The table translates each principle into the question a director should ask before approval.

Principle	Question for an AI approval
Validate technology, value, and model	Has the system been validated in our setting, do users want it, and who pays for it?
Build barriers	Does this create an advantage of our own, or the same capability every competitor can buy?
Stage around risk	What condition must be met before the next phase of deployment?
Acquisition cost and lifetime value	What will it cost to bring clinicians and staff onto the system, and how long will the value last?
Project realistically	What return did management project, and how will it be measured after go-live?
Evaluate objectively	Who, independent of the sponsor, will report on performance, and when?
Engage the customer	Which clinicians and patients shaped the design and the testing?
Keep it simple	Can management explain what the system decides, and what it does not, in one paragraph?
Manage dependencies	What do the vendor terms say about training on our data, change notice, audit, and indemnity?
IP and early entry	Who owns the models, outputs, and improvements built on our data?

The Strongest Objection

The strongest objection is that startup principles do not belong on the agenda of an established health system or pharmaceutical company. Its board oversees a going concern, not a seed-stage bet, and it should not micromanage internal projects. I agree that the board should not run the project. Nose in, fingers out. But the risks that sink young ventures, unvalidated technology, optimistic projections, and a single critical supplier, are the same risks that an AI program carries into an established organization. The difference is that a startup's board sees them at every financing round, while an established board may see an AI program only once, at approval. The principles give directors a short, repeatable test for that one moment, and a reason to ask for a second look.

For the Board

• Which of the three validations (technology, user value, and business model) has the venture or AI program proven, and with what evidence?
• What specific risk will the next round of financing, or the next phase of deployment, retire, and how will the board know it has been retired?
• Who is the paying customer, who is the user, and how does each affect acquisition cost and lifetime value?
• Which projections support the price or the budget we are approving, and when will we compare them with results?
• Which third-party dependencies could stop the business or the program, and what do the contracts say about them?

To discuss how these principles apply to your board's AI approvals, book a conversation.

References

23andMe Holding Co. (2025, July 14). Current report (Form 8-K). U.S. Securities and Exchange Commission. https://www.sec.gov/Archives/edgar/data/1804591/000119312525158551/d11473d8k.htm

American Hospital Association. (2024, March 15). AHA survey: Change Healthcare cyberattack significantly disrupts patient care, hospitals' finances. https://www.aha.org/node/692736

Berkowitz, B. A., & Simkin, M. M. (2023, December 8). FDA approves two gene therapies to treat sickle cell disease including a CRISPR-based therapy. Foley and Lardner. https://www.foley.com/insights/publications/2023/12/fda-gene-therapies-sickle-cell-disease/

Gavi, the Vaccine Alliance. (2016, October 14). Rwanda launches world's first national drone delivery service powered by Zipline. https://www.gavi.org/news/media-room/rwanda-launches-worlds-first-national-drone-delivery-service-powered-zipline

Mehta, I. (2025, March 23). DNA testing company 23andMe files for bankruptcy protection, CEO resigns. TechCrunch. https://techcrunch.com/2025/03/23/dna-testing-company-23andme-files-for-bankruptcy-protection-ceo-resigns

Oracle. (2022, June 7). Oracle completes acquisition of Cerner [Press release]. https://www.oracle.com/news/announcement/oracle-completes-acquisition-of-cerner-2022-06-07/

Pharma Manufacturing. (2013, December 9). FDA approves Gilead's potential hep C cure; $1000 per day pricetag. https://www.pharmamanufacturing.com/industrynews/2013/fda-approves-gileads-sovaldi

Pifer Parduhn, R. (2023, February 23). Teladoc reports historic net loss of $13.7B in 2022. Healthcare Dive. https://www.healthcaredive.com/news/teladoc-historic-net-loss-137b-2022/643213/

Price, E. (2024, April 30). Walmart shutters 51 doctor clinics after just a few years amid struggle to make healthcare profitable. Fast Company. https://www.fastcompany.com/91115790/walmart-closing-health-care-clinics

Roche. (2018, February 15). Roche to acquire Flatiron Health to accelerate industry-wide development and delivery of breakthrough medicines for patients with cancer [Press release]. https://www.roche.com/media/releases/med-cor-2018-02-15

Tierney, A. A., Gayre, G., Hoberman, B., Mattern, B., Ballesca, M., Kipnis, P., Liu, V., & Lee, K. (2024). Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catalyst Innovations in Care Delivery, 5(3). https://doi.org/10.1056/CAT.23.0404

Vertex Pharmaceuticals. (2021, April 20). Vertex Pharmaceuticals and CRISPR Therapeutics amend collaboration agreement for CTX001 [Press release]. https://investors.vrtx.com/news-releases/news-release-details/vertex-pharmaceuticals-and-crispr-therapeutics-amend