The Pharma AI ROI Paradox

by Christian Hayden and Laura Farmer

The insightful 2026 Healthcare AI ROI Scorecard from Bessemer Venture Partners and Bain contains an apparent paradox for pharma. AI pilots are proliferating: the share of pharma use cases in pilot stage rose 11 percentage points over the past year. Scaled production, however, rose just 1 point. Meanwhile, the clearest returns are early in the value chain. Preclinical discovery delivered 3.4x realized ROI, versus 2.3x for clinical development and 2.6x for commercial.

The obvious read is that pharma should chase the highest near-term returns. At Opus, we see a second possibility: the pilot-to-production gap may be partly explained by the difficulty of measuring value on a useful timeline. If a team can’t put a defensible number on a tool’s value, it’s hard to justify scaling it.

Preclinical discovery may be the exception that proves the point. It leads pharma today in both AI adoption and realized ROI. Further downstream, it gets much harder to accurately attribute value: drug development unfolds over many years and outcomes have many causes, so the ultimate financial impact of an AI-assisted decision can arrive long after it was made.

Most of the highest-stakes AI applications don’t offer the luxury of a clean, near-term payoff. An AI tool that sharpens target selection, trial design, patient identification, competitive decision-making, or portfolio prioritization may create enormous value. But the biggest payoff may be terminating a weak program before an expensive late-stage trial, or identifying a promising asset that otherwise would have been deprioritized. Given the enormous cost of late-stage failure, even a modest improvement in the probability of making the right go/no-go decision could dwarf years of incremental productivity savings.

A similar challenge exists in pharma sales and marketing. AI can deliver readily measurable efficiency gains in content development and field force operations, but assessing its impact on HCP targeting, personalized engagement, and commercial strategy is more complicated. The important question isn’t just whether AI made commercial activities faster or cheaper, but whether it improved resource allocation, influenced prescribing behavior, or generated incremental revenue that wouldn’t otherwise have materialized.

In other words, some of AI’s most consequential impacts in pharma may take years to become visible. And even then, isolating AI’s contribution to the outcome can be difficult. Companies therefore need meaningful ways to assess value along the way. Beyond direct cost savings and productivity, pharma companies should ask of an AI application:

•      Did it shorten the time to a decision?

•      Did it allow the team to evaluate more options with the same resources?

•      Did it surface risks or opportunities earlier?

•      Did it lead to better decisions?

Answering that last question is, of course, easier said than done. A tool can make a team more certain… and more wrong. The test is retrospective: did AI-informed go/no-go calls outperform historical base rates, and did earlier risk flags actually predict later failures?

Ultimately, what matters is whether AI helps companies put capital and scientific resources behind the right development programs. That’s difficult to capture in a quarterly calculation, but in an industry where one better development decision can be worth billions, it’s a measure worth getting right.

It’s a question we’re spending more time on at Opus Strategy: how do you measure AI’s contribution rigorously enough to guide investment decisions without waiting years to find out whether those investments paid off? We’d be interested to hear how others in biopharma are approaching it.

KRASCENDO-1 Raised the Bar. Can RAS(ON) Inhibitors Raise It Again?

By Christian Hayden and Laura Farmer

It may shock you to learn that KRAS, the much-venerated drug target whose success reached a crescendo this year with daraxonrasib’s swift approval, was once considered undruggable. Yes, you heard it here first, and in none of the other 10,000 articles written about KRAS.

Undruggable. Then KRAS G12C inhibitors like sotorasib and adagrasib upended that narrative, establishing direct KRAS inhibition as a viable therapeutic strategy. Now, results from ESMO 2026 suggest that the next generation can deliver substantial improvements over the G12C vanguard. As Roche’s divarasib establishes a new benchmark for KRAS G12C inhibition, an emerging class of RAS(ON) inhibitors raises a larger question: will the future of KRAS-targeted therapy be defined by better drugs within the existing mechanism, or by a fundamentally different approach?

The Phase 3 KRASCENDO-1 trial provided compelling evidence for divarasib in previously treated KRAS G12C-mutated non-small cell lung cancer. Compared with sotorasib or adagrasib, divarasib significantly improved median progression-free survival (9.8 vs. 5.7 months), median overall survival (17.7 vs. 12.0 months), and objective response rate (58.4% vs. 32.3%). Grade 3–4 treatment-related adverse events were also less frequent (24.6% vs. 34.5%), although gastrointestinal toxicity remains a concern.

The results generated widespread enthusiasm among onco KOLs, with many describing the findings as practice-changing. Some went so far as to suggest that divarasib could effectively displace existing KRAS G12C inhibitors once approved. One important caveat: approximately 91% of patients in the comparator arm received sotorasib, leaving considerable uncertainty about divarasib’s relative performance against adagrasib specifically. But even with that caveat, divarasib has demonstrated that the established KRAS(OFF) approach is still capable of delivering major clinical advances.

But is that enough?

Two Approaches to Targeting KRAS

Divarasib, like sotorasib and adagrasib, is a KRAS(OFF) inhibitor. These therapies bind the inactive, GDP-bound form of KRAS G12C, preventing its reactivation and suppressing downstream oncogenic signaling.

Meanwhile, a different approach is advancing through clinical development. RAS(ON) inhibitors target the active, GTP-bound state of RAS, potentially offering an alternative way to suppress oncogenic signaling and address certain mechanisms of resistance to existing therapies.

Revolution Medicines’ elironrasib is among the most relevant emerging competitors. Unlike divarasib, elironrasib targets KRAS G12C in its active state and has demonstrated preliminary activity in patients previously treated with KRAS(OFF) inhibitors. In an early cohort of 24 heavily pretreated patients, 92% of whom had progressed on prior KRAS(OFF) therapy, elironrasib demonstrated encouraging antitumor activity. Although these findings remain preliminary, they suggest a potential role for RAS(ON) inhibition after progression on existing targeted treatments.

The plot thickens. The emerging competition isn’t simply between active- and inactive-state inhibition. It’s also between drugs designed for specific KRAS mutations and broader approaches capable of targeting multiple RAS variants. Daraxonrasib, for instance, is a multi-selective RAS(ON) inhibitor whose development extends beyond G12C, while zoldonrasib selectively targets KRAS G12D. Meanwhile, pan-KRAS and pan-RAS approaches could expand targeted treatment to patients whose mutations currently lack effective options.

This introduces another competitive dimension: the potential trade-off between the precision of mutation-selective inhibitors and the broader reach of multi-selective therapies. Broader inhibition could expand the addressable population and potentially suppress additional resistance pathways, but their broader activity also raises important questions about selectivity, tolerability, and therapeutic index.

These agents are not all direct competitors to divarasib. Rather, they illustrate how the KRAS landscape is evolving beyond competition between individual G12C inhibitors toward a wider range of mechanisms, mutation profiles, and therapeutic strategies.

So What’s Next?

The next phase of KRAS drug development may be defined as much by treatment setting as by mechanism. For KRAS(OFF) inhibitors, the immediate opportunity is moving into earlier lines of treatment. Roche’s KRASCENDO-2 is evaluating divarasib plus pembro in first-line NSCLC, while Lilly’s SUNRAY-01 and Merck’s KANDLELIT program are investigating combinations involving olomorasib and calderasib, respectively. These studies could establish whether the benefits of KRAS inhibition can be extended beyond previously treated patients.

The first-line race is particularly consequential. Divarasib’s success in KRASCENDO-1 could establish it as the preferred option for patients who have not previously received a KRAS G12C inhibitor. But as targeted therapies move earlier in treatment, that second-line, inhibitor-naive population could shrink. The eventual commercial success of KRASCENDO-1 could therefore hinge on which agents establish themselves in first line, and whether subsequent KRAS inhibition remains effective after progression.

For RAS(ON) inhibitors, the opportunities are potentially broader. Activity following KRAS(OFF) inhibitor exposure could create a distinct role in treatment sequencing, while successful first-line combinations might eventually position these agents as direct competitors to KRAS(OFF) therapies.

However, RAS(ON) inhibitors may not have the post-progression setting to themselves. Next-generation KRAS(OFF) agents are also showing activity after prior G12C inhibition. D3 Bio’s elisrasib, for example, demonstrated a 32.3% response rate and median progression-free survival of 8.1 months in a Phase 1/2 cohort previously treated with a KRAS G12C inhibitor. While cross-trial comparisons are dicey, the findings suggest that overcoming resistance may not require abandoning the KRAS(OFF) mechanism altogether.

This creates an intriguing possibility: rather than one mechanism simply replacing the other, KRAS(OFF) and RAS(ON) inhibitors could ultimately occupy complementary positions within the treatment pathway. But that outcome is far from certain. If RAS(ON) inhibitors demonstrate superior efficacy or tolerability in earlier-line randomized trials, they could challenge KRAS(OFF) therapies directly. Divarasib’s success, meanwhile, demonstrates that the established mechanism remains a formidable competitor.

Important questions also remain around CNS activity, resistance mechanisms, and the safety of combinations with immunotherapy. These factors may prove just as consequential as the distinction between active- and inactive-state inhibition.

The Next Chapter

KRASCENDO-1 represents a major advance for patients with KRAS G12C-mutated NSCLC and a significant competitive achievement for Roche. But its implications extend well beyond the immediate competition between KRAS G12C inhibitors.

The first generation of KRAS inhibitors demonstrated that a historically difficult target could be drugged. Divarasib has now shown that substantial improvements are possible within an established mechanism. The emerging RAS(ON) pipeline will test whether targeting the active state of RAS can deliver further benefits, overcome resistance, or expand the population eligible for targeted treatment.

© 2026 Opus Strategy, LLC