How to Hire Your First Customer Service Agent
Dawson Chen
At some point, the founder can't keep doing support alone. Tickets pile up. Response times slip. You start spending hours a day on refunds, account issues, and questions that feel repetitive. The obvious next step is to hire someone.
Here's how most startups do it, what we learned when we tried it ourselves, and why we eventually went a different direction.
The Traditional Playbook
The default path for scaling support at a startup looks like this:
- Start with Zendesk or Intercom. Handle everything manually, no AI.
- The founder manages support directly, then eventually hands it to an ops person.
- As volume grows, you hire offshore agents or a small CS team.
- Once you're big enough (10k+ tickets/month), you bring in custom AI agents from Decagon or Sierra to automate at scale with forward deployed engineers.
This works. Plenty of companies have scaled this way. But it's slow, expensive, and every transition between stages means losing context and rebuilding processes.
Making Your First Hire
Many startups go offshore for their first support hire. Sites like onlinejobs.ph and Virtual Coworker make it easy to find full-time support agents in the Philippines for $500-$1,000/month. The talent pool is large, English proficiency is decent, and the cost lets you hire earlier than you could domestically.
When we were hiring our first support agent at our previous company, Martin(https://www.trymartin.com/), we posted a job and gave each candidate a take-home assignment: resolve 20 fake support tickets. These involved Stripe refunds, database lookups, account troubleshooting, the same kinds of things they'd handle on the job. About 10 people finished the assignment. We did one more behavioral interview with each of them and made the hire.
Onboarding Is Where It Gets Hard
The hire is the easy part. Onboarding is where things get messy.
To get our new agent up to speed, I wrote a massive knowledge base Google Doc covering every type of ticket he'd see. How to process refunds. How to look up user data. When to escalate. What tone to use. It was dozens of pages.
For the first week, I had him draft replies only. No sending. I reviewed everything. Then he started responding to a portion of tickets on his own, and I spent about two hours each week reviewing his work and giving feedback.
He'd make mistakes. I'd coach him. He'd improve. Slowly.
The Plateau
Even months into the job, mistakes kept happening. Small ones, mostly. Wrong refund amount. Slightly off tone. Missing context from a previous conversation. The kind of errors that happen when someone is working from a static document and doesn't have the full picture of your product in their head.
And there was always a chunk of tickets, maybe 20-30%, that he couldn't resolve at all. These were too technical, too product-specific, or required context that he did not have. Those bounced back to us anyway.
We were paying for a full-time hire, spending hours per week on review, and still handling a third of the tickets ourselves.
What We Did Instead
Eventually, we built an AI agent to handle the work. We trained it on our actual ticket history, connected it to our database and Stripe, and gave it the same playbook we'd written for our human agent. The AI drafted replies, suggested refunds, and flagged tickets it couldn't handle.
We went back to doing support ourselves, with the AI doing the heavy lifting. That project is what we later spun into Letterbook.
How Letterbook Changes This
Letterbook replaces the Zendesk/Intercom layer and the offshore hire in one tool. You connect your inbox, your database, and your payment system. The AI agent drafts a reply to every ticket, using your actual customer data and your knowledge base.
It also surfaces quick actions, like refunds or cancellations, for you to confirm in one click. Nothing goes out without a human approving it. But the time you spend per ticket drops dramatically.
The goal is simple: one person (or even half of a full-time employee) can own all of customer support. Instead of hiring an offshore VA and spending hours per week on QA, you review AI-drafted replies and hit send.
When to Actually Hire
This doesn't mean you'll never hire a support agent. At some point, volume or complexity will justify it. But that threshold is much higher with AI doing the first pass on every ticket.
If you're a startup doing under a few hundred tickets a day, one person with Letterbook can cover it. When you do eventually hire, that person ramps faster because the AI already has the playbook, the tone, and the context built in. They're joining a system that works, instead of building one from scratch.
