Playbook · operator how-to
How to Find Startups to Invest In (Before They Raise)
A step-by-step process to find startups to invest in before they raise, using GitHub commit-velocity and contributor-influx signals on top of your usual sourcing.
Most guidance on how to find startups to invest in stops at the obvious: network harder, read Crunchbase every morning, show up to demo days. Those channels work, but they put you in the same line as every other investor, bidding on the same 40 companies the week the round is announced.
The edge is in the six to eight weeks before that announcement. In that window a startup's public GitHub activity changes in measurable ways: commit velocity spikes, new contributors arrive, new repositories appear. Those changes show up in the commit graph before they show up in a press release, and they are free to read.
This playbook is the process we run on the operating side of GitDealFlow. It is designed for angels, scouts, and pre-seed through Series A investors who want a systematic source of candidates that is not yet crowded. Total time per week is roughly 40 minutes of focused work, plus whatever you spend on the calls you decide to take.
You do not need to be an engineer to use it. You need to know which numbers move when a team gets busy, and which ones mean nothing. That is what the four-test rubric below is for.
Before you start
Prerequisites
- · A target list of 3 to 5 sectors you actually understand (not a thesis you borrowed)
- · A simple spreadsheet or CRM to hold your candidate shortlist
- · Read access to the free signals feed at /api/v1/signals.json (no auth)
Tools
- · GitDealFlow signals feed (free, no auth)
- · GitHub web UI (commit and contributor inspection)
- · A watchlist spreadsheet
Steps
- 01
Pick your sector slice
5mChoose the 3 to 5 sectors where you have an actual information edge. Fetch the signal slice for each, for example https://signals.gitdealflow.com/api/v1/signals.json?sector=ai-ml. You are looking for the current top repos ranked by composite signal score.
- 02
Screen for acceleration, not size
10mIgnore total commits and total stars. You want change: a repo whose 28-day commit velocity jumped relative to its own 90-day baseline. A steady 200 commits a week tells you nothing. Going from 80 to 240 in two weeks tells you something is happening. Filter for velocity-percentile and velocity-change, not raw volume.
- 03
Apply the four-test rubric
10mFor each shortlisted repo, check four things: (1) commit-velocity change over 14 to 28 days, (2) contributor growth (new people shipping, not one founder grinding), (3) new repository creation (platform building, not just polishing one repo), (4) a stack that matches the stated stage. Keep only candidates passing 3 of 4.
- 04
Cross-check the team, not just the code
8mOpen the org's GitHub page and the founders' profiles. You are checking whether the signal has a face: is the acceleration driven by a real team, or by one prolific developer plus bots? Look for a named founder, a coherent org, and contributor diversity. Cut anything that is clearly a hobby project or an enterprise fork.
- 05
Time-box your diligence before you reach out
5mRun the five-checks-in-five-minutes pass: commit consistency, contributor growth, technology choices, new repo creation, and the ratio of product code to maintenance activity. This is a screen to decide whether a company deserves a real meeting, not a replacement for that meeting.
- 06
Reach out with a specific, non-generic note
2mWhen you find a candidate, your first message should reference something only someone who actually looked would know: the spike in a specific repo, the new contributor pattern, the infrastructure you saw them building. That specificity is what gets a founder to reply to a cold note, and it is the thing a public GitHub signal gives you for free.
Run the play
Get the weekly breakout-startup report →Frequently asked questions
How do I find startups to invest in without a network?
Start from public signal feeds instead of warm intros. GitHub activity is public, timestamped, and free to read. A systematic weekly screen of 3 to 5 sectors will surface candidates weeks before they appear in your inbox. The network compounds later; the signal gives you a first source today.
Do I need to know how to code to use GitHub signals?
No. You need to know which numbers move when a team gets busy (commit-velocity change, contributor influx, new repos), not how to write the queries. The GitDealFlow feed pre-computes those numbers so you read ranked lists, not raw git logs.
Is GitHub data enough to make an investment decision?
No, and it should not be. Public GitHub data is a sourcing and screening layer. It tells you which companies deserve a technical deep dive and a founder call, not whether to wire the check. Use it to decide who gets a meeting, then do normal diligence.
How is this different from just reading Crunchbase?
Crunchbase tells you what already happened (a round closed, a hire announced). GitHub signals show you what is happening now, in the weeks before it becomes a Crunchbase entry. They answer the same question at different points in the timeline; use both.
How many candidates should I expect per week?
A focused 40-minute screen across 3 to 5 sectors typically yields 5 to 10 candidates worth a second look, of which 1 or 2 clear the four-test rubric and justify a founder call. That is a healthy weekly cadence for a solo angel or scout.
Related playbooks
The rubric above is the same one I run on the operating side every week. I keep my name off it on purpose, the edge is in the timing, not the messenger. Trust the math, not me.