AI Is Not Ready to Be the Decision-Maker — And Pretending Otherwise Is the Real Risk
August 3, 2026
No Comments
Read More »
Close
At BIO 2026, a panel on AI and data partnerships in biopharma spent an hour describing something that sounded, on the surface, like generosity.
Eli Lilly’s TuneLab platform has signed up more than 100 companies. Demand to join is high enough that it functions as a waitlist business. Lilly has struck arrangements with partners like Axelleaf and a Japanese CRO specifically so it can bring more datasets into the TuneLab ecosystem, and it is offering platform access to smaller companies in exchange for their data, with, as it was described on stage, “no fiscal comingling.” Translation: no cash changes hands. The biotech gets tools; Lilly gets data.
Nobody on the panel called this an acquisition strategy. They called it ecosystem enablement. Aliza Apple, who heads TuneLab, framed it as a way to let smaller companies commercialize their models and data without needing to build a large commercial team of their own. That’s a real and generous-sounding value proposition, and it may even be true. But generosity and strategy are not mutually exclusive, and the industry has a bad habit of taking the friendliest available framing at face value.
Then Jack Castle, CBO of Ochre Bio, said the quiet part out loud. Discussing why more biotechs are now willing to partner around their data, he noted that “the maximum benefit is to have the data in their hands and a perpetual license,” particularly valuable for training runs. Read that sentence again: not access to the data, not a multi-year license with defined scope, but a perpetual license, held by the partner, to material generated by your company. That’s not a footnote to the deal. For a data-hungry AI program, it may be the entire point of the deal.
This paper argues that TuneLab-style platforms, and the broader category of “free” AI-for-data arrangements now proliferating across big pharma, should be evaluated by biotech BD teams with the same rigor as an M&A term sheet. Not because they’re predatory, but because they’re structurally identical to an acquisition of the asset that increasingly matters most: proprietary biological and clinical data, minus the acquisition premium, the change-of-control provisions, or the shareholder vote.
Acquiring a biotech is expensive, slow, and comes with integration risk, headcount, and a board that has to approve the price. Acquiring the data a biotech generates, without acquiring the biotech, its liabilities, or its people, is comparatively cheap. It requires a platform, a marketing story about democratizing AI tools for smaller players, and patience.
TuneLab’s structure does exactly this. A biotech gets access to compute, models, and infrastructure it likely couldn’t build alone. In exchange, its data becomes part of a pool that a company the size of Lilly can use to build tools with far more leverage than any single contributor could manage independently. Lilly’s collaboration with BigHat Biosciences, explicitly pursued because of BigHat’s data quality, illustrates the point cleanly: the relationship exists because of what BigHat has, not because of what BigHat needs. The access-in-exchange-for-tools framing obscures which party is actually the acquirer here.
None of this makes Lilly a villain. A rational actor building a moat around AI-driven drug discovery should want exactly this kind of arrangement. The problem is that biotechs are signing on the other side of the table as though this were a partnership among equals, when the more accurate mental model is: you are trading a long-dated, possibly irrevocable asset for a short-dated convenience.
Here is the detail that should worry BD teams more than the license language itself: nobody in the room, including the panelists building these platforms, agreed on how to measure whether the AI tools on offer are actually good.
The panel raised, without resolving, the question of whether the industry will ever converge on a standard for benchmarking model performance. Right now, it doesn’t. Success is measured differently by different players, and Lilly, notably, still validates its own models with wet-lab work rather than trusting model outputs on their own. If the platform’s own architect feels the need to independently verify results in a lab, that should tell every biotech considering a data-for-access deal something important: there is no external, agreed-upon yardstick you can use to confirm that what you’re getting back is worth what you’re giving up.
This is the part of the bargain that should feel uncomfortable. In a normal transaction, both sides can at least agree on what’s being exchanged and roughly what it’s worth. In this one, the currency biotechs are handing over (proprietary data, in perpetuity) is precisely priced and irreversible. The currency they’re receiving (access to tools of unverified, unbenchmarked, self-assessed quality) is neither.
The incentives are understandable. Small and mid-sized biotechs often lack the capital, headcount, and infrastructure to build competitive AI capabilities on their own. Ochre Bio’s Castle noted that more companies are now open to partnering around their data than in the past, a shift driven partly by need and partly by a growing recognition that data sitting idle has no value at all. TuneLab’s pitch, that not every company needs to build a large commercial team, resonates because it’s true. Most don’t.
But “we couldn’t have done this alone” is a description of leverage, not a justification for terms. Every biotech that has ever sold itself cheap in a difficult financing environment has used some version of that logic. The fact that a deal solves a real capability gap doesn’t mean the price is fair; it means the seller had limited alternatives, which is exactly when a counterparty’s terms deserve the most scrutiny, not the least.
Treat any data-for-platform-access arrangement as a de facto acquisition of your data asset, and diligence it accordingly. Before signing, a BD team should be able to answer each of the following with specificity, not with a vague sense that “it’s probably fine because everyone else is doing it too.”
None of this means biotechs should refuse these deals. Many will remain the smartest available option for
companies that need capability now and can’t build it alone. But smart BD teams should stop treating “no
cash changes hands” as a synonym for “no cost.” The cost is just denominated in a currency-data rights
in perpetuity-that most deal teams aren’t used to pricing. Until benchmarking convergence gives the
industry a shared way to measure what these platforms actually deliver, the safest assumption is that the
terms were written by the party who understood the asset’s value better. Usually, that isn’t the biotech.
At Scimitar, Akira Robinson serves as Partner, Commercialization. He operates at the intersection of commercial strategy, launch execution, and market access, advising biopharma executive teams at critical moments where launch readiness directly determines asset value and time to market. With 20 years of experience across life sciences, diagnostics, biologics, and digital medicine, his work tackles complex commercial challenges for teams across the US and globally. His expertise spans the full commercial value chain—including launch planning, licensing, market access, pricing, analytics, marketing, sales distribution, and patient services across therapy areas, including: oncology, radiopharmaceuticals (RLT), rare disease, CNS, cardiology, and gastroenterology.
