Why your molecule stays in its lane.

Ask a business development manager whether the molecule they sell could serve a market they do not currently serve, and you will usually get a considered maybe. Ask them to go and find out, and the answer changes. Not because they lack the judgement — because they have a number this quarter.

The evidence is scattered, and that is the whole problem

The material that would answer the question is public, but it is not in one place and it is not written for a commercial reader. It sits in journals nobody in the commercial team reads. It sits in patents filed by adjacent industries, where your chemistry appears as somebody else's raw material. It sits in regulatory dockets that move a year before any announcement, which is precisely the year in which the information is worth something.

None of that is hidden. All of it is findable. It is simply spread across enough separate places that finding it is a reading job rather than a search.

Why search and a chatbot do not close the gap

This is the part people assume has already been solved, and it has not.

A search engine ranks popularity, not evidence. It returns what is linked to. The document that actually decides whether your molecule can serve an application is frequently cited by nobody, written for a different discipline, and sitting well past the first page. Search is excellent at finding what many people already wanted. It is poor at finding the one load-bearing paper that nobody in your industry has read.

A general-purpose AI answers from memory, not from the record. It responds out of what it absorbed during training — a compressed summary of the internet at some past date — rather than from a reading of the primary sources now. You get a fluent, plausible answer with no source you can check, and, worse, no account of what it failed to find. For this work that is disqualifying: a finding you cannot check is a finding you should not act on, and an answer that cannot say what it could not look at is indistinguishable from one that looked and found nothing.

And your own files are not the answer either. The material inside the company is dated, and the same ground has been walked repeatedly by people who already share the same assumptions. The places that are genuinely new are, by definition, the ones nobody in the building has looked at.

It needs a particular kind of person, and they are busy

What the work actually demands is someone who can read a patent and know what the molecule really does, and who knows which application is worth a phone call and which is noise. Deep technical training and commercial judgement, in the same head. Add the weeks of mandated reading time, and the ability to compress the result into a case management will back, and you have four things that are each common on their own and rare together.

Most companies have one or two people capable of it. Those people are already fully committed. So in practice it is not a person at all — it is a team assembled across functions and held together for months, which is expensive enough to do once. That is the constraint: not budget, and not intent.

What it costs, in P&L terms

Be precise about what is lost, because “we should do more adjacent selling” is not an argument anybody funds. Six things change when the reading actually gets done:

• Adjacent selling becomes a process rather than a project. It runs, rather than starting and stopping.
• Top-line growth comes from products you already make. No new chemistry, no new plant, no new regulatory file.
• ROCE improves on plant you already run. A line under capacity is a line with room in it.
• Growth arrives without new capital. The capex conversation does not gate it.
• New customers appear inside markets you already know — the ones who are not basic in your molecule.
• Knowledge compounds instead of leaving with people. What was learned is written down and kept.

That last one is the quiet one. Most market work ends as a slide deck, and everything learned in the process of making it is lost. The next person starts from nothing. Keeping the vocabulary and the evidence instead is what makes the second chemistry easier than the first.

What changes it

Machine reading changes the economics of the weeks, and nothing else about the problem. The AI reads at scale and writes every fact down with its source and the date it was read. A chemist then judges what any of it means, because a ranked list is not a decision and a well-evidenced market is not the same thing as an attractive one.

The honest version of the claim is narrow: PROWL makes it affordable to look properly, and cheap enough to keep looking. It does not size the market for you, and it does not tell you which opportunity is commercially attractive. It tells you what the public record actually supports, names what it could not check, and puts that in front of a person who has sold into these industries for thirty years.

A finding you cannot check is a finding you should not act on. That is why every fact carries its source, and why the leads that failed are published next to the ones that survived.