Where AI Doesn’t Change the Math — Part B
Part A argued that AI-assisted diligence widens the top of the in-licensing funnel. More opportunities can get real evaluation without proportionally more staff.
That’s a claim about screening capacity. It says nothing about what happens after screening, and that’s where the limits remain.
The Real Ceiling Is R&D Spend and Personnel, Not BD Throughput
In-licensing volume is ultimately bounded by how much a company spends on R&D and how many people are available to develop what it signs, and not by how many candidates its BD team can screen.
If a company allocates a given percentage of revenue to R&D, faster or better screening doesn’t move that number. AbbVie’s roughly 15 executed deals a year reflect what its organization can integrate and fund, not what its BD team is capable of finding. AI widens what gets evaluated. It doesn’t widen what gets funded.
Better, faster BD screening doesn’t manufacture capital or headcount out of nowhere. A smaller team with excellent AI-assisted diligence can find more good candidates than it could before. But if it can’t develop them, finding them faster just produces a longer, better-characterized list of things it still can’t do anything about.
There is a seemingly obvious solution to this. If screening quality improves enough that companies start visibly leaving good opportunities on the table (i.e., turning down assets purely for lack of development budget) that visibility could itself become an argument for shifting more R&D spend toward licensing over time (at the potential expense of internal R&D spend).
Screening quality wouldn’t just filter within a fixed budget. It could become an input to the capital allocation decision itself.
We’ll see.
What Might Actually Improve
None of this means better screening is pointless. It means its payoff shows up differently than simply “more deals.”
Today’s BD teams are making rigorous decisions within real capacity limits, which means good assets sometimes get missed simply because there wasn’t time to look closely enough.
If more of the inbound pool receives a full evaluation instead of being capacity-rationed at the door, two things plausibly follow.
First, the shortlist that reaches term sheet should be better characterized, with fewer surprises in later diligence, because more of the elimination happened on evidence rather than on which candidates happened to get analyst attention that quarter.
Second, there’s a chance this reduces herding. A meaningful share of “hot” partnering targets get bid up not because every bidder has done independent work, but because bandwidth constraints push teams toward pattern-matching on what everyone else is already chasing.
If AI-assisted diligence makes it cheap enough to independently evaluate the less-obvious candidates too, some of that crowding pressure could ease.
AI could widen what gets looked at, but it doesn’t widen what gets bought.
Whether that produces a genuinely better-selected, less herd-driven deal pool at the bottom of the funnel, or just a faster, more expensive way of arriving at the same fifteen deals a year, is an open question, and one worth watching over the next few licensing cycles.