You Can See Everything, the documentary that follows Elizabeth Holmes and her partner Billy Evans in the 34 days before she began her federal prison sentence, has earned more than $64 million since its October 16 release [1]. I see Holmes's story as a cautionary tale for life science companies. It shows what happens when hype and optics crowd out the principles that decide whether a technology works.
Theranos raised more than $700 million from investors, according to the SEC. It did it with false and misleading statements in investor presentations, product demonstrations and media articles [2]. In truth, its own analyzer could run only a small number of tests, and most patient tests ran on modified commercial machines made by other companies [2]. The money it raised said nothing about whether the product worked.
That lesson is worth more today than it was a decade ago. If you lead strategy, BD or R&D portfolio decisions, AI partners are pitching you every week, and you have to decide which ones belong in your strategy.
The AI boom has created a gold mine of opportunity
This is an exciting time for life science companies. Venture firms have put more than $40 billion into AI drug discovery over the past decade, according to PitchBook data, and Isomorphic Labs raised $2.1 billion in a single round this year [3]. AI labs now run their own drug programs. Anthropic acquired Mithril Medicine, a one-person company that worked with 11 outside labs to take a muscular dystrophy treatment from an idea to promising results in mice in less than 18 months [4].
More partners, more capital and more tools mean more ways to move faster. The frenzy also makes it tougher to tell what's real, making the same terrain look like a potential minefield. I see two types of mines in it.
The first type of mine: partners that aren't what they seem
Theranos isn't the only well-funded health technology company that turned out to be something other than it claimed.
Babylon Health raised $550 million in a single round and went public at a $4.2 billion valuation [5]. In 2018, a UK advertising authority told it to withdraw a claim that its triage tool gave sound advice 100% of the time. The study behind that claim was an internal evaluation in which actors followed scripted scenarios, and former employees said there was no system to follow up on what happened to patients [6]. After Babylon collapsed, its UK business sold for about $620,000 [7]. I wrote about what Babylon's partners missed in my Babylon Health case study.
Nate raised more than $40 million on the claim that its app used AI to complete online purchases on its own [8]. Federal prosecutors allege its actual automation rate was effectively zero, with contract workers in a call center placing the orders by hand [8]. These are allegations, and the founder is presumed innocent until proven guilty.
Each of these companies cleared investor scrutiny and raised serious money. Funding is a surface signal. It tells you someone else was persuaded. It doesn't tell you the product works.
The second type of mine: honest partners solving the wrong problem
This mine is quieter and more common. The partner is honest, the technology does what it promises, and the investment still goes nowhere.
The drug pipeline shows the pattern clearly. A new review of 129 AI-enabled assets from 69 companies found a 77.8% probability of success in Phase 1 and 19.5% in Phase 2 [9]. The authors note that the Phase 2 figure looks more like conventional drug development. Insilico, one of the most prominent AI-native developers, acknowledged the same pattern in its own 2025 paper in Nature Medicine [10]. Experts writing in Nature Reviews Drug Discovery called AI's clinically relevant impact "disappointingly limited" [11].
The reason is structural. Phase 1 mostly tests chemistry, pharmacokinetics and tolerability, where models have plenty of data to learn from. Phase 2 asks whether a drug works in human disease, and that data is sparse [10]. Lack of efficacy accounts for 40 to 50% of clinical failures [10]. AI is helping with the early steps. It hasn't yet changed the step where most programs fail.
Nothing in those numbers points to deception. The technology improved steps that weren't the constraint.
Both mines hide behind surface signals
In both cases, decision makers judged what they could see: the funding round, the demo, the test score, the Phase 1 win. Fewer traced whether the partnership would change outcomes where the business needed it.
Strong due diligence on a partner matters, and it's how the first type of mine gets caught. Strategic value comes first, because it's the only protection against the second type. You assess it by asking one question at each part of your value chain: if this partner delivers exactly what it promises, what happens to the outcome here?
Walk the chain from the step the partner touches to the result the enterprise cares about. In R&D that might run from discovery through validation, trials, regulatory review, launch and access. Eliyahu Goldratt made the underlying point in The Goal four decades ago: a system's output is set by its constraint, and time saved anywhere else doesn't raise it [12]. One recent review in drug development calls the modern version the "new productivity paradox." It's the risk that upstream acceleration only moves the bottleneck downstream when validation capacity and operational readiness don't keep up [13].
The same logic holds outside R&D. A faster medical writing partner won't shorten your cycle if MLR review is the constraint. A better site-selection tool won't help a protocol that can't enroll.
Expectations are the other foundation hype tends to erode. When every competitor announces an AI partnership, it's tempting to promise the board a breakthrough. Leaders who set expectations step by step, with clear measures of what should change and when, make better calls about what to fund next. Anthropic did this with Mithril. It said plainly that AI "can't make cells grow or mice age any faster" and labeled its later timelines "targets, not commitments" [4]. I made a related argument in Beyond the Hype: What 25 Years of Genomics Teaches Us About the Future of AI. Sequencing became fast and cheap long before it changed how many drugs succeeded.
Starting from foundations also makes the first type of mine easier to spot. A leader who knows what outcome the partnership must prove will ask for evidence at that point (Babylon), ask where the capability shows up in daily operations (Nate) and ask who outside the partner has confirmed that it works (Theranos).
Five gates to clear before you commit
So is it a minefield or a gold mine? That depends on where you start. Diligence tells you whether a partner can deliver. These five gates tell you whether delivery would matter. I'd run them before diligence starts, because the answers show diligence which claims deserve the closest look.
Strategic fit. Name the enterprise objective or capability gap this partnership addresses, and how much it should move. If no one can name it, stop. The trade-off: a tight target can end an exploratory bet before it has time to show value. If you want to explore, label the bet as exploration, with a smaller budget and a learning goal.
Value chain impact and the bottleneck. Trace what changes at each step between the partner's work and the objective, then find the step that limits the result. If the bottleneck sits downstream, the gain will wait in a queue. The trade-off: the answer often points to a fix inside your own organization, such as review capacity or data standards. No partner can sell you that, and it's a less exciting announcement than an AI deal.
Dependencies. List what has to be true elsewhere for the value to land: downstream capacity, fit-for-purpose data, governance and the readiness of other partners such as CROs. The trade-off: you may need to invest downstream first, which delays the partnership and the visible win.
Expectations and proof. Agree on what the partner must prove at the bottleneck, how you'll measure it and when you'll review it, then set expectations with leadership to match. The trade-off: outcome evidence takes longer than a pilot cycle, and a measured plan can look modest next to competitors' announcements. It protects your credibility when results arrive slower than the hype.
The decision. Choose to back, reshape, stop or sequence, and name the conditions that would change the call. Sequencing often wins: fix the bottleneck first, then bring in the partner. The trade-off: waiting can cost you access to a partner competitors also want, and stopping means accepting what's already been spent.
What this looked like in a platform decision
I worked with the leadership of a global biopharma company that had a growing number of filings and limited capacity to manage them. They were deciding whether to keep investing in an immature platform. The vendor's claims on time, cost and quality looked strong on paper.
We tested those claims against key metrics around effort, rework, usability and implementation progress. The question was whether the platform reduced workload or just reframed and moved it onto the employees the company couldn't afford to lose. We also checked four foundations the platform depended on: process standards, content governance, reliable upstream data and expert capacity to run it. Then we compared continued investment with faster alternatives, including process changes and AI-assisted drafting.
The recommendation was to stop until performance and those foundations improved, with clear conditions for restarting. An estimated $2 million to $6 million in first-year licensing and implementation spend was at stake. Just as important, the decision protected resource time for critical filings. Diligence on the vendor alone wouldn't have surfaced that, because the constraint sat inside the client's own organization.
What this means for BD and clinical development
For BD and enterprise strategy leaders, run the first three gates at the portfolio level before anyone sees a demo. A partner that improves a step that isn't limiting your results can wait, with no budget committed for now. Mapping your value chain also turns diligence into a focused test of the claims that matter most.
For clinical development leaders, you own the step where much of the value from early-pipeline AI stalls. Ask to be in the room when discovery partnerships are scoped. Your capacity, your protocols and your evidence plans decide whether those partnerships pay off.
Questions for the leadership team
For our largest AI partnership, which enterprise objective is it supposed to move, and has that number changed?
Where is our true bottleneck between discovery and approval, and how many of our AI partnerships touch it?
What have we told the board to expect from our AI partners, and would we still stand behind it today?
The partnerships that pay off are usually decided long before the demo, at the level of objectives and value chains.
Facing a consequential strategy decision? Bring it to Talon Group Consulting.
Talon Group Consulting's Hidden Readiness Gap study looks at how life science leaders judge external partner readiness, investment discipline and the data behind their priority initiatives. It's confidential, takes 8 to 10 minutes and closes December 6. Early findings will be shared in December, with the full report in January. Take part in the study.
References
McArdle, Tommy. "Elizabeth Holmes' Partner Billy Evans Denies He 'Controls' Her Amid Doc's Success." People, 3 Nov. 2026, people.com/elizabeth-holmes-partner-billy-evans-defends-himself-amid-docs-success-12211267.
"SEC v. Elizabeth Holmes, et al. and SEC v. Ramesh 'Sunny' Balwani." Litigation Release No. 24069, U.S. Securities and Exchange Commission, 19 Mar. 2018, sec.gov/enforcement-litigation/litigation-releases/lr-24069.
"ByteDance AI Drug Discovery Spin-off Raises $290 Million." The Insight, 15 Oct. 2026, theinsight.asia/zhang-yimings-ai-drug-discovery-spin-off-hits-1-5-billion-valuation.
"Early Progress on Accelerating Drug Development with Claude." Anthropic, 27 Oct. 2026, anthropic.com/research/early-progress-drug-development.
"The Fall of Babylon Is a Warning for AI Unicorns." WIRED, 2023, wired.com/story/babylon-health-warning-ai-unicorns.
Olson, Parmy. "This Health Startup Won Big Government Deals, But Inside, Doctors Flagged Problems." Forbes, 17 Dec. 2018, forbes.com/sites/parmyolson/2018/12/17/this-health-startup-won-big-government-dealsbut-inside-doctors-flagged-problems.
Jennings, Katie. "Bankrupt Digital Health Company Babylon Sells UK Assets for Just $620,000." Forbes, 19 Sept. 2023, forbes.com/sites/katiejennings/2023/09/19/bankrupt-digital-health-company-babylon-sells-uk-assets-for-just-620000.
"Tech CEO Charged in Artificial Intelligence Investment Fraud Scheme." U.S. Attorney's Office, Southern District of New York, 9 Apr. 2025, justice.gov/usao-sdny/pr/tech-ceo-charged-artificial-intelligence-investment-fraud-scheme.
Haque, Waqas. "AI-Discovered Drugs Are Reaching the Clinic, But Are They Reaching Patients?" Oncology Data Advisor, 3 Nov. 2026, oncdata.com/ai-discovered-drugs-are-reaching-the-clinic-but-are-they-reaching-patients. Summarizing Haque, Waqas, et al. "Artificial Intelligence-Enabled Drug Discovery: A Growing but Largely Early-Stage Clinical Pipeline." Cancer Treatment Reviews, vol. 151, 2026, p. 103223.
Starsiak, Casimir, III. "Is AI Drug Discovery Looking in the Wrong Places?" Drug Discovery News, Nov. 2026, drugdiscoverynews.com/is-ai-drug-discovery-looking-in-the-wrong-places-17618.
Leventhal, Rajiv. "AI Drug Discovery Attracts Billions, But Clinical Impact Remains Limited." EMARKETER, 18 Sept. 2026, emarketer.com/content/ai-drug-discovery-attracts-billions--clinical-impact-remains-limited.
Goldratt, Eliyahu M., and Jeff Cox. The Goal: A Process of Ongoing Improvement. North River Press, 1984. Summarized in Irons, Ann. "Throughput Accounting and the Theory of Constraints." ACCA Student Accountant, Oct. 2011, accaglobal.com/content/dam/acca/global/pdf/sa_oct11_throughput.pdf.
Herdiana, Yedi, et al. "Artificial Intelligence as a Disruptive Force in Pharmaceutical Innovation: Transforming Discovery, Development, and Manufacturing." Drug Design, Development and Therapy, 2026, pmc.ncbi.nlm.nih.gov/articles/PMC12994533.

