A playbook I have instinctively run on the smaller scale, building industrial labs, pursued this as a student, then started decades earlier than Novartis.

Leads to exponential growth, as always.

Would recommend, always willing to repeat, no risk at all.

Experimentation & discovery-R-us.

It was a no-brainer.

Mainly didn't choose anything else.

The final section pounding the desk about how terrible ending the program was seems like it is oddly at variance with all the evidence OP had just laid out about how the program wasn't working well anymore and so wasn't actually financially a good idea. It's weird to quote a bunch of things like studies showing that 'internal R&D spending works worse than external for ROI' and then write a big moralizing sermonizing conclusion about how ending internal R&D is bad for profits and how terrible it is there's no 'patient capital' (capital which was plenty available before - what's the theory, investors stopped liking making money? insurance companies with century-long investment horizons ceased to exist? etc).

The story about picking an unpopular disease that was easy to test reminds me of why Monsanto went with glyphosate resistance as the first serious GMO target: Trivial testing.

Back when Monsanto started that kind of research, the technology to modify a plant's DNA, and checking the quality and location of the modifications were extremely crude: you'd see the modification inserted into hundreds, if not thousands of locations at once. It was definitely going to make the plant worse at growing at the beginning, and require a lot of work to use traditional breeding to improve the seedstock again. But glyphosate had a huge advantage: Testing whether your new GMO plant has your genes properly activated is trivial. plant all the modified seeds as you can, wait a few days until you have leaves, then spray the whole thing with glyphosate: If the DNA didn't make it, or it's in a place where it doesn't get expressed enough, the plant just dies. No need to use a chipper and spend a ton of money sequencing and checking the specific location of the insertion.

Today the speed and price of genomic pipelines is such that one can attempt a lot more complicated things and get results without risking so many failures, but if you make detecting failure cheap, you end up ahead anyway.

This analysis appears to propose that buying-in drug development programs is more financially efficient than developing them in-house. Presumably these bought-in programs are found among smaller biotech companies.

This overview however omits the costs incurred by all those who were not bought-in, i.e. the biotechs funded by VC, etc, who never get bought.

So in terms of the costs of innovation the overall analysis may not support either buying-in or in-house, its just that the risks are differently distributed.

A separate question, and that which appears to have been the foundation of NIBR's erstwhile success, is that in NIBR the scientists and clinicians who innovate new drug candidates remain closely involved in the later stages of drug development. This would be in theory possible with either model, i.e. it would depend more on company culture than the origin of discovery. Acqui-hires that are common in tech for example prioritize continuity of intellectual and technical know-how (as far as I understand it).