Onboarding is where good first impressions go to die. An applicant reaches out, someone replies, a step happens, and then a gap opens: the next email depends on a person remembering to send it. Most companies do not lose clients because a step failed. They lose them because nothing happened between steps, and nobody noticed the silence until the applicant did.
I am VP of Operations at an early-stage fintech company. In June and July 2026 I built our client onboarding email system to close those gaps for good. The design is simple to describe: every stage from first contact to onboarded client has a written email template, and delivery is automated. When an applicant moves to the next stage, the right email goes out. No one has to remember. Remembering is exactly the thing I removed from the process.
Who wrote what
The division of labor matters here, because it is the opposite of how most people assume this gets built.
I wrote every template myself. Every stage, every sentence. Then I directed AI coding tools to wire the delivery: which email fires at which stage, and how it gets sent without a human in the loop. I am not a coder. I did not write the automation code, and I did not need to. I described the flow, the AI built it, and I tested that each stage triggered the message I expected.
That split is deliberate. The words are the relationship. An applicant deciding whether to trust your company is reading tone, not admiring your infrastructure. So the human wrote the words and the machine got the plumbing. I would make the same split again on any system where the output lands in a person's inbox.
The mechanical part went fine. The words were the part that failed first.
The templates that sounded like a bank
My early templates were correct and dead. "We are pleased to confirm receipt of your application." "Please be advised that the next step in the process is as follows." Every sentence was accurate, and the whole thing read like it had been generated by a legal department and approved by a committee. Nobody talks like that. Nobody trusts an email that talks like that either, because it signals that a machine is processing you, even when a person hit send.
I caught it before the automation went live, and I rewrote every template before turning anything on. The test I used was blunt: read the email out loud. If I could not imagine saying the sentence to someone across a table, it got rewritten until I could. "We are pleased to confirm receipt" became a plain sentence telling the applicant we got their application and what happens next. Shorter sentences. One person talking to one person. The information stayed identical. The voice changed completely.
The order of operations was the real lesson. Fix the words first, automate second. Automation is an amplifier: it takes whatever you feed it and repeats it to every applicant, forever, at scale. If I had automated the bank-voice templates, I would have built a machine for making cold first impressions with perfect reliability. The system would have worked and hurt us every day it ran.
Since the system went live, the sequence has run itself. An applicant moves from first contact to onboarded client through a set of stages, and at every stage they hear from us in a voice I wrote and stand behind. Nothing depends on someone's memory. Nothing sits in a drafts folder waiting for a busy week to end. The gap between steps, the place where applicants used to fall through, no longer exists.
The takeaway is one you can apply this week without writing a line of code. Take the emails your company sends at each step of onboarding and read them out loud. Any sentence you would not say to a person across a table, rewrite until you would. Do that before you automate anything, because automation will faithfully scale whatever you hand it, including the tone that makes people feel processed. Write the words yourself. Let the machine do the remembering. That order, not the tooling, is what keeps the balls from dropping.