Kill, pivot or partner: the two questions you keep answering as one

In December 2023, Bristol Myers Squibb agreed to buy Karuna Therapeutics for $14bn [1]. The asset at the centre of that cheque was a muscarinic M1/M4 agonist called xanomeline, and Eli Lilly had it first, having shelved it back in the 1990s [2].

Lilly had taken xanomeline into Alzheimer's, chasing a cognitive signal. The signal was real, and so were the side effects: nausea, vomiting, diarrhoea, sweating, a cholinergic profile no patient wanted to live with [2]. So it went on the shelf, until Karuna licensed it in 2012 [3], paired it with trospium to strangle the peripheral side effects, and carried the combination all the way to FDA approval as Cobenfy for schizophrenia in September 2024 [4].

Read that sequence back. The receptor biology was never the problem; Lilly's fit for that biology was. Two questions were hiding inside one kill, pivot or partner decision, and Lilly, like almost everyone, answered them together.

The value/fit collapse

Every kill/pivot/partner call is really two questions wearing one coat.

The first is about the molecule: does this asset still have real scientific or commercial value? The second is about you: does this company have the capability, the capital and the strategic focus to be the one that realises it? One question interrogates the science, the other the organisation holding it.

Here is what teams do with those two questions: they fuse them. Runway gets tight, or the board loses its nerve, and "we can't afford this" quietly answers "it doesn't work" two dials that measure completely different things, soldered into a single switch. Call it the value/fit collapse: the most expensive mistake in portfolio management, precisely because it feels like discipline while you are making it.

Jekunen's 2014 paper treats development as a set of real options rather than a linear march. Most teams reach for licensing only once the go/no-go has already failed, as a consolation prize; Jekunen has it live at every review, a standing option you weigh every single time. In his words, "there are decisions to be made on the licensing of new compounds and licensing-out existing compounds, and on whether to continue ongoing projects or terminate them" [5]. That is four standing options on the table at once, then. And real-option valuation is a proper peer-reviewed methodology built for exactly this kind of decision, whatever the people who wave it off as academic hand-waving might tell you [6].

So separate the two questions on paper before every go/no-go call: score value in one column, fit in another, and never let one column quietly fill in the other.

The four quadrants: the kill, pivot or partner matrix

Cross the two axes and the muddle resolves into four boxes.

Low value, any fit: kill. Take crenezumab: Roche ran two large Phase III trials, CREAD1 and CREAD2, and stopped both on 30 January 2019 after an interim analysis found the primary endpoint, CDR-SB, was unlikely to be met [7]. Look closely at the texture of that failure, though, because the mechanism plainly engaged: in the earlier Phase II ABBY and BLAZE studies, crenezumab lowered CSF amyloid-beta oligomers in the large majority of treated patients [7], and yet the Phase III clinical benefit never came [7]. Mechanism engaged, clinical benefit absent. That is a real and distinct kill signature, worth naming because teams routinely mistake the biomarker for the value: engagement tells you the mechanism fired, the endpoint tells you whether it mattered to a patient. When the molecule does exactly what you designed it to do at the molecular level and the patient is no better, killing is the correct call rather than a failure of nerve.

Real value, poor fit: partner. Lilly abandoned daptomycin when it downsized its infectious-disease division, and the antibiotic then sat on a shelf for four years before Lilly out-licensed it to Cubist. In the words of Krishnamurthy and colleagues' systematic review, "Eli Lilly has received over $333 million in royalties on the product sales to date" [8]. A shelved molecule can keep its biology long after its owner walks away, ending up in a house its owner cannot live in while someone else still can. Partnering works from strength just as readily as from distress: Daiichi Sankyo licensed trastuzumab deruxtecan (Enhertu) to AstraZeneca for $1.35bn upfront and up to $6.9bn in total potential value, kept Japanese rights and manufacturing, and co-sponsored the pivotal DESTINY-Breast03 head-to-head throughout [9]. None of that is a verdict on the molecule itself. Poor fit just means the asset needs a different owner, and the moment to find one is while the data is still fresh.

One caveat on that Enhertu number, and on every biobucks headline you will ever read: announced deal value and realised deal value are very different animals. Recon Strategy's analysis of the deal databases found preclinical deals realise under 20% of their announced value even nine years out, and Phase 1 deals around 32% [10]. The upfront is real money; the milestones are a lottery ticket. Price the asset accordingly.

Value contingent on a testable redesign: pivot. This is Karuna, whose trospium fix carried a specific, falsifiable hypothesis: bolt on a peripherally-restricted anticholinergic to block the gut and sweating effects without it crossing into the CNS and blunting xanomeline's central action [4]. You could have written that hypothesis down in a single sentence in 2012 and known in advance exactly which result would kill it. Sildenafil shows the same discipline a decade earlier, moving from angina to erectile dysfunction to pulmonary hypertension, each step tracking the same nitric-oxide/cGMP smooth-muscle-relaxation mechanism rather than a marketing hunch [11].

Real value, good fit: continue. This is the only box most teams genuinely consider, a hangover from a linear playbook built for a different kind of company. Its shadow is the value/fit collapse running in reverse: a low-value asset kept alive because it lives in the wrong home and nobody in the room will say so out loud. Sunk cost wearing a "continue" badge.

Why nobody ticks the partner box

Here is the structural hole. Kill and continue are on the table at every portfolio review, but partner almost never is, and no gate forces it onto the agenda at the same moment as the other two.

Some years ago I sat in a board meeting where a perfectly good early-phase asset was talked into the ground over a single afternoon. The molecule had a clean mechanistic story and a plausible second indication; what it did not have was a champion inside a company whose whole attention was fixed on its lead programme. Nobody argued the science was dead. They argued we couldn't afford it, and then, having settled that, quietly filed the thing under "doesn't work." The word "partner" was never once spoken, so in the language of the review that asset was killed, when in reality it had merely been orphaned.

The literature has a name for the reflex. Krishnamurthy and colleagues call it the "not sold here" syndrome: "the unwillingness of companies to out-license compounds that may be promising for other indications... Business units argue that if they do not sell a product, no one else should" [12]. The same review notes that pharmaceutical companies "were generally described as employing few, if any, staff to aid in out-licensing shelved compounds" [12]. There is no team, no gate, no owner, so the option does not exist on the org chart and therefore does not exist in the room.

Underneath sits ordinary behavioural economics: sunk cost, escalation of commitment and champion bias, all now well documented in the portfolio-management literature as forces to design against rather than personality flaws [13][14]. It remains a live concern in current deal-making, not settled history [15]. (Of course, the people quickest to treat out-licensing as surrender are usually the ones who have never had to make a payroll run against a dwindling balance.)

So the fix is boring and structural: a partner column on the same review that decides kill versus continue, triggered at the same moment, before the data ages and the bidders drift away.

A pivot in name only

Pivot is the most abused word of the three, because it lets you keep spending while sounding decisive. The test is simple and falsifiable: can you name, in one sentence, the specific mechanistic reason the new approach avoids the old failure mode? Karuna could. Axovant could not.

Axovant bought intepirdine, a 5-HT6 antagonist, from GSK for $5m in December 2014 after it had shown "preliminary efficacy" in Phase II [16]. The company then raised $315m and reached roughly $3bn in market value before generating a single new pivotal data point [16]. The Phase 3 MINDSET trial, roughly 1,300 patients with intepirdine added to donepezil, missed its primary endpoint in September 2017, and the share price fell more than 70% in a day [16]. So what did Axovant do next? It put the same, unmodified molecule into a different indication, dementia with Lewy bodies, with no new mechanistic hypothesis for why it would work this time, and that trial also failed, in early 2018 [16]. Same jar, new label. That is not a pivot.

That said, a real hypothesis is no guarantee of a win. Phenserine was mechanistically grounded and still underperformed [17]; the falsifiable hypothesis is what makes an attempt legitimate, and it does nothing to ensure the attempt succeeds. A deliberate expansion run from strength, such as semaglutide moving from diabetes into obesity on a well-understood mechanism, is not a rescue pivot at all, because nothing failed, and confusing the two is its own kind of error. (Post-hoc subgroup reanalysis after a futility stop is the genuinely contested middle ground here, and we have written about one famous example of it elsewhere.)

"The matrix is a luxury I can't afford"

When the runway is four months and the board wants a clean story for the next raise, a two-axis matrix can look like theatre, and an out-licensing process that takes the better part of a year threatens to pull a lean team's focus off the one asset that can still raise money. Killing fast and pouring every remaining pound into the lead programme can be the disciplined move, whereas nursing a shelf-asset fantasy that no acquirer has actually bid on rarely is. All of that is fair.

And all of it is the value/fit collapse in motion, where the problem is timing rather than the frame. Jekunen again, and this is the line to pin above the desk: "If development is terminated based only on a panic reaction to poor study results, that decision has been made far too late. By that point, too much time and money have been wasted, which calls into question not only the decision process but can even jeopardize the very existence of the company" [5]. Run the two-question test before the crunch, not during it, so the partner conversation happens while the data is fresh and there are still bidders in the market, rather than as a fire-sale after the cash position has already cast the deciding vote.

What to do:

1. Two columns, never one: score value and fit separately, and ban any sentence in the room where affordability or board sentiment is smuggled in as evidence that the molecule does not work.
2. Add a partner column to the kill/continue review: this is the structural fix for the missing gate, so trigger it at the same moment as the other two, not after a kill has quietly been decided.
3. One sentence for any pivot: write the falsifiable hypothesis for why this approach avoids the prior failure mode, and if you cannot write it, you are funding a dead thesis wearing a pivot's clothes.
4. Timestamp the decision: decide before the cash-crunch decides for you.

None of this needs a consultant. What it needs is an honest read of the molecule and of your own company, kept in separate columns, and that is the kind of discipline that matters most when resources are tight. Whether that valuation happens in-house or with outside help, that objective read is what a platform like InovaSight is built to support, something we have written about at length. The judgement itself stays yours.

Somewhere out there is a molecule that works, killed in the wrong home, that reached no patient it could have helped. The receptor biology was never the question. Make sure you are answering the one that is.

References

1. Bristol Myers Squibb to acquire Karuna Therapeutics for $14bn. Announced 22 December 2023 (widely reported public record; BMS press release, "Bristol Myers Squibb Strengthens Neuroscience Portfolio with Acquisition of Karuna Therapeutics").
2. Xanomeline: muscarinic M1/M4 agonist developed by Eli Lilly for Alzheimer's cognition in the 1990s and shelved over dose-limiting peripheral cholinergic tolerability (nausea, vomiting, diarrhoea, hypersalivation/sweating). Public record / drug-development history (Bodick et al., multicentre xanomeline AD study; Alzforum KarXT profile).
3. Karuna Therapeutics in-licensed xanomeline from Eli Lilly in May 2012 (fact of licence; public record).
4. FDA approval of Cobenfy (xanomeline–trospium, KarXT) for schizophrenia, 26 September 2024. Trospium co-administered as a peripherally-restricted anticholinergic to block peripheral cholinergic effects without CNS penetration. FDA / Bristol Myers Squibb public record.
5. Jekunen A. (2014). "Decision-making in product portfolios of pharmaceutical research and development – managing streams of innovation in highly regulated markets." Drug Des Devel Ther;8:2009–2016. PMID: 25364229. https://pubmed.ncbi.nlm.nih.gov/25364229/
6. Graham E, Jaki T, Harbron C. (2020). "A comparison of stochastic programming methods for portfolio level decision-making." J Biopharm Stat;30(3). PMID: 31825729. https://pubmed.ncbi.nlm.nih.gov/31825729/
7. Hoffmann-La Roche. "A Study of Crenezumab Versus Placebo in Participants With Prodromal to Mild Alzheimer's Disease (CREAD)." ClinicalTrials.gov: NCT02670083 (TERMINATED). https://clinicaltrials.gov/study/NCT02670083 — and "CREAD2." ClinicalTrials.gov: NCT03114657 (TERMINATED). https://clinicaltrials.gov/study/NCT03114657. Both Phase III trials were stopped on 30 January 2019 following a pre-planned interim analysis (primary endpoint CDR-SB unlikely to be met); the Phase III trials showed no clinical benefit and no significant AD biomarker effect (Ostrowitzki et al. 2022, JAMA Neurol, PMC9486635). Target engagement — crenezumab lowering CSF amyloid-beta oligomers in the large majority of treated patients (median change −48% SC / −43% IV) — was demonstrated earlier in the Phase II ABBY/BLAZE studies: Yang T, et al. (2019). "Target engagement in an Alzheimer trial: Crenezumab lowers amyloid β oligomers in cerebrospinal fluid." Ann Neurol;86(2):215–224. PMID: 31168802. https://pubmed.ncbi.nlm.nih.gov/31168802/
8. Krishnamurthy N, et al. (2022). "Drug repurposing: a systematic review on root causes, barriers and facilitators." BMC Health Serv Res;22:970. PMID: 35906687. https://pubmed.ncbi.nlm.nih.gov/35906687/ (daptomycin/Cubist: "Eli Lilly has received over $333 million in royalties on the product sales to date"; "Eli Lilly out-licensed the drug to Cubist after four years on the shelf").
9. Daiichi Sankyo. Press release on the trastuzumab deruxtecan (Enhertu) global development and commercialisation collaboration with AstraZeneca, 28 March 2019 ($1.35bn upfront; up to $6.9bn total potential value comprising $5.55bn contingent — $3.8bn regulatory/other milestones + $1.75bn sales milestones; Daiichi Sankyo retaining Japan rights and manufacturing). Co-sponsorship verified via DESTINY-Breast03, ClinicalTrials.gov: NCT03529110 (AstraZeneca listed as collaborator). https://clinicaltrials.gov/study/NCT03529110
10. Recon Strategy. Analysis of biopharma deal value realised by development stage, based on the Citeline/Evaluate Deals Database (preclinical deals realise under 20% of announced "biobucks" value even nine years post-deal; Phase 1 deals realise ~32%, on an average total announced value of ~$304m). https://reconstrategy.com/2025/04/preclinical-licensing-deals-realized-value/ and https://reconstrategy.com/2025/08/phase1-licensing-deals-realized-value/
11. Goldstein I, et al. (2019). "The Serendipitous Story of Sildenafil: An Unexpected Oral Therapy for Erectile Dysfunction." Sex Med Rev;7(1):115–128. PMID: 30301707. https://pubmed.ncbi.nlm.nih.gov/30301707/ — and Ghofrani HA, Osterloh IH, Grimminger F. "Sildenafil: from angina to erectile dysfunction to pulmonary hypertension and beyond." Nat Rev Drug Discov (2006). PMID: 16883306. Open access: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7097805/
12. Krishnamurthy N, et al. (2022). "Drug repurposing: a systematic review on root causes, barriers and facilitators." BMC Health Serv Res;22:970. PMID: 35906687. https://pubmed.ncbi.nlm.nih.gov/35906687/ ("not sold here" syndrome; out-licensing under-resourcing).
13. Weber B, Zineh I, Lalonde R, Visser SAG. (2024). "How debunking biases in research and development decisions could lead to more equitable healthcare?" Clin Transl Sci;17(7):e13880. PMID: 39016187. https://pubmed.ncbi.nlm.nih.gov/39016187/ (names the sunk cost fallacy in go/no-go decisions; recommends structured decision processes, explicit criteria and deliberate bias mitigation).
14. Bieske L, Zinner M, Dahlhausen F, Trübel H. (2023). "Trends, challenges, and success factors in pharmaceutical portfolio management: Cognitive biases in decision-making and their mitigating measures." Drug Discov Today;28(10):103734. PMID: 37572999. https://pubmed.ncbi.nlm.nih.gov/37572999/ (confirmation bias, champion bias, misaligned incentives).
15. Ma X, Lu Z, Zhao Y, Sheng J. (2026). "De-risking Biopharma Asset Acquisition: Towards a Quantitative Framework for Strategic Decision-making." AAPS J;28(3). PMID: 42091774. https://pubmed.ncbi.nlm.nih.gov/42091774/ (names "sunk cost fallacies" and confirmation bias in biopharma asset-acquisition/partnering decisions).
16. Axovant Sciences / intepirdine: GSK divested the 5-HT6 antagonist for $5m in December 2014 after "preliminary efficacy" in Phase II; Axovant raised $315m and reached ~$3bn market value; MINDSET Phase 3 (~1,300 patients, added to donepezil) failed its primary endpoint September 2017 (share price down over 70% in a day); the same unmodified molecule then failed in dementia with Lewy bodies in early 2018. BioPharma Dive, STAT News and Alzforum reporting.
17. Ballard C, et al. (2020). "Drug repositioning and repurposing for Alzheimer disease." Nat Rev Neurol;16:661–673. PMID: 32939050. https://pubmed.ncbi.nlm.nih.gov/32939050/ (defines "hypothesis-driven repurposing"; phenserine as a mechanistically-grounded repositioning candidate that underperformed).