We published an article on how we built our own AI infrastructure at Draper to automate every routine task and give our team more time to do the human work in VC.
Two years ago, our firm ran on eight Excel sheets.
Today, one auto-updated base runs everything: 400+ portfolio companies, 15,000+ contacts, 1,900+ meeting notes, searchable in plain English from a phone.
We published the full build log, the 5 decisions we made before building anything matter more than any tool:
• Decide where the data lives first. One base, one record per company. Everything else feeds it or builds on it.
• Build on infrastructure you already own. Airtable, Zapier, Claude. No new vendors. Fixing something means editing a prompt, not shipping a deploy.
• Customize the judgment. An off-the-shelf lead scorer doesn't know your thesis or your passes.
• When the system isn't sure, it stops and asks a human. No guessing where the data has to be right.
• Document every tool like a shipped product, so the next person picks it up in five minutes.
The biggest payoff is the hours it returns to the work only people can do: sitting with a founder, getting to know them deeply, and making the call.
Read the whole blog on how we built Draper’s AI stack in the comments.