The interesting companies rarely appear where people are actively looking. They show up in accelerators, funding programmes, trade publications and registry changes, often a year before they raise capital or are ready to sell.
Tracking those sources by hand is legwork. It automates well. And that is precisely where a new problem starts.
More candidates are not better
A sourcing run covering forty sources weekly produces hundreds of hits without effort. The bottleneck has not disappeared; it has moved, from finding to sifting.
Delivering a list sorted by score at that point does not reduce the work, it relocates it. The team now reviews a hundred rows instead of forty websites, and because nobody knows why row three ranks above row seventeen, they check by hand anyway.
The benefit only appears where the shortlist is solid enough to rule candidates out.
What a score has to be able to do
A scoring system that merely counts keywords produces exactly that worthless list. It finds every company with “AI” in its self-description and does not distinguish a product from a statement of intent.
It becomes useful when the investment criteria are modelled explicitly: sector, maturity, business model, region, and above all the exclusions. In practice the exclusions are worth more than the inclusions, because they are what actually narrows the funnel.
Part of that is every result carrying a reason. Not “score 82”, but the sentence explaining why this company fits the thesis and where the uncertainty sits. Only then can an analyst check the shortlist instead of redoing it.
Dealing with uncertainty
Sourcing data is patchy. Revenue figures are missing, headcounts are out of date, ownership structures are not public.
A system that fills those gaps with plausible estimates without marking them is unusable: it produces numbers that end up in a presentation where nobody can still tell they were estimates. Estimated values belong flagged as such, with the source they were derived from.
Where the time saved should go
The temptation is to open the funnel wider at the top as soon as sifting gets cheap. That is usually the wrong direction.
The real gain is reviewing ten thoroughly instead of a hundred superficially. The time an automated shortlist frees up is most valuable exactly where experience and market access count, and that is the conversation with the company, not reading its website.
