CaretCandidate sourcing for a small startup: a playbook in one sheet
You're hiring your first backend engineer. There are 40 tabs open: GitHub profiles, a conference talk, two blog posts, a LinkedIn page you found through a friend. Your notes app has eleven names, and for six of them you can't remember why you saved them. The job post has been "almost done" for two weeks.
That's sourcing at most small startups. There's no recruiter, no applicant tracking system, and the founder does it between other things. What helps is one sheet with fixed columns and a scoring rule you write before you look at anyone. Then decide which parts a bot does and which stay with you.
What we learned building a list like this
The list we built was for sales, not hiring. The mechanics are the same, though, and the lessons carried over.
In October our marketer bot built a target list of 339 companies from four sources in one turn, each row scored for fit. It started from my two-line answer to a numbered question about where to look. The four sources came out like this:
| Source | Rows | High fit |
|---|---|---|
| A co-founder friend's network | 100 | 39 |
| My own network | 34 | 18 |
| Startups funded April to October 2026 | 85 | 44 |
| AI-first startups in India | 120 | 81 |
| Total | 339 | 182 |
Two things from that list matter for hiring. First, the public sources gave us companies but mostly no person: founder profile links were missing for most of those rows. A row without a person and a link isn't a lead yet. Second, 182 high-fit rows is too many to work one by one. The score's real use was ordering the list, so we started at the top.
The sheet: columns that make it work
One row per person. Every column earns its place.
| Column | What goes in it | Why |
|---|---|---|
| Name and link | Profile URL | No link, no row |
| Source | Where you found them | Tells you which sources are worth your time |
| Evidence | A link to something they built, wrote or ran | Your fit score has to rest on something |
| Must-haves met | Yes or no for each of your 2-3 must-haves | The filter |
| Fit score | 0-6, from the rule below | The order |
| Warm path | Who you know who knows them | A warm intro beats a cold message |
| Why them | One line, written when you add them | So you remember in two weeks |
| Status | Found, shortlisted, contacted, replied, passed | Where each person is |
| Contacted | Who, when, which message | So nobody gets two messages |
The fit score is simple on purpose. Pick three things that predict the job, and give each 0, 1 or 2:
- Has done this work in production. 2 if there's public evidence, 1 if the role suggests it, 0 if not.
- Has worked at this stage. 2 for a team under 20, 1 for a small team inside a big company, 0 otherwise.
- Shows how they think. 2 for writing, talks or code reviews you can read, 1 for a little, 0 for none.
Write the rule down before you look at anyone. If you write it afterwards, you'll write it to fit the people you already like.
| Name and link | Source | Evidence | Must-haves met | Fit score | Warm path | Why them | Status |
|---|---|---|---|---|---|---|---|
| Jordan K. | Wrote about Postgres | Blog post on query plans | Yes, yes | 6 | Investor | Ran Postgres at a 12-person team | Shortlisted |
| Sam R. | Open source contributor | Pull requests to a library we use | Yes, yes | 5 | None | Careful code reviews | Shortlisted |
| Alex M. | Investor network | Conference talk | Yes, no | 4 | Investor partner | Talk on scaling writes | Found |
| Priya S. | Past applicant | Take-home from last year | Yes, yes | 4 | Applied before | Near-miss last round | Contacted |
| Taylor B. | Meetup | Job title only | Unknown | 0 | Friend | Needs real evidence | Found |
Where to look when nobody knows your name
| Source | What you find | Effort |
|---|---|---|
| People who wrote about your stack | Blog posts, talks, long issue threads | Medium |
| Contributors to open source you use | Real code, real reviews | Medium |
| Your investors' and friends' networks | Warm paths | Low, if you ask well |
| Past applicants and near-misses | People who already know you | Low |
| Community channels and meetups | Active people, often early in a search | High |
The warm sources are small but they matter more. In our sales list, the best warm intros ran through one co-founder friend's network.
Where the person stays in
A bot can find people, fill the sheet, score against your rule and draft messages. These steps stay with you:
- Writing the must-haves and the scoring rule.
- Picking the shortlist from the top of the sheet.
- Reading and approving every message to a candidate. It speaks for your company, and you can't unsend it.
- Saying no to someone, kindly, yourself.
- Any call on a profile that doesn't fit the rule but feels right. Rules miss people.
An example: sourcing a founding backend engineer
The company and the people in this example are made up.
The founder asked: "build a sourcing sheet for a founding backend engineer. Must-haves: has run Postgres in production, has worked on a team under 20. Score with the 0-6 rule in your notes. Sources: people who've written about Postgres performance, contributors to the two libraries we depend on, and our investors' portfolio teams."
Back came an Excel sheet with 60 rows, each with a link to evidence. 22 met both must-haves. The bot flagged nine rows where the only evidence was a job title.
The founder dropped those nine to a score of 0 until there was real evidence, added "lives within four hours of our time zone" as a must-have, and moved two people up by hand because of a talk they gave. Then they picked 12 for the shortlist.
Next, the bot drafted 12 short messages, each one citing the specific post or pull request that made the founder interested. The founder rewrote two openings and approved five to send this week, on a card that read "Send 5 emails". The other seven wait for next week, so replies stay manageable.
build a sourcing sheet for a founding backend engineer. Must-haves: has run Postgres in production, has worked on a team under 20. Score with the 0-6 rule in your notes.
Attached sourcing-backend-engineer.xlsx. 60 rows, each with a link to evidence. 22 meet both must-haves. 9 rows have only a job title as evidence, flagged.
Drop those 9 to 0, add "within four hours of our time zone" as a must-have. Shortlist is 12, draft messages.
Send 5 emails
ApproveDenyApproved.
An example thread, shortened.
How to set it up in Caret
This is the same setup our marketer bot used for the 339-row list, pointed at hiring.
- Create a recruiting bot and write its job: "build and keep our sourcing sheet, score people with the rule in your notes, draft outreach, never contact anyone without approval".
- Put the must-haves and the scoring rule in its files. Memory lasts between conversations, so it remembers what you thought of each person last week.
- Ask for the sheet. The bot looks people up in its own web browser and makes Excel files: "make a sheet of people who've written about Postgres in production, with links and one line each".
- Connect only the apps it needs through connected apps, and decide which inbox it drafts from.
- Leave approvals on. Before it sends anything, it asks, and an admin approves or denies with one click. Every send is in the audit trail.
- Add a schedule for the weekly update: "every friday 16:00 post where each candidate is and what's waiting on us".
- Save the routine as a skill once it works, so the next role starts from it.
Caret's Team plan is $40 a bot a month, with a 7-day trial. You and your co-founder each get a bot, you both see the same thread, and the bots keep working when your laptops are off.
FAQ
How do small startups find candidates without a recruiter? Start with warm paths (investors, friends, past applicants), then people with public evidence of the work: writing, talks, open source. Keep everyone in one sheet with a link and a reason, so you can see which sources produce people.
How do I score candidates for fit before talking to them? Pick two or three must-haves and three scored criteria, 0 to 2 each, and write them down before you look. Score only on evidence you can link to. Use the score to order the list, then let a person pick the shortlist.
Is it OK to use AI to message candidates? Use it to draft, and send each message yourself after reading it. Candidates can tell a mass message, and high volume from one account can get it flagged. A few specific, approved messages a day work better.
If you're the founder who's also the recruiter, here's what a recruiting bot can take on. Put a bot on it: start your 7-day trial or download the Mac app.