5 min read
What if the real reason customers ask for refunds has almost nothing to do with your product? For a lot of solo course creators, the refund is not a verdict on the material. It is what happens when someone gets stuck, cannot find the answer, and gives up before the lesson that would have helped them. Fix the getting-unstuck part, and refunds fall on their own.
That is the thread running through the story below. It follows Nadia, a one-person online educator. She is an illustrative composite, built from patterns common to solo course businesses rather than one real person, so treat the specific numbers as a realistic example, not a documented case. The lesson underneath, though, is very real, and you can borrow it this week.
Meet the one-woman course business
Nadia sells a self-paced course that teaches small-batch candle making, priced at 149 dollars, with a couple thousand students enrolled over three years. It is a true solo operation. She records the lessons, runs the marketing, answers every email, and processes every refund herself. For a long time the business felt healthy. Then it started feeling like a help desk with a course attached.
The problem hiding inside a good product
By early this year, Nadia was spending two to three hours a day on support email. The questions were rarely about candle making. They were about where to find the supply list, whether a downloaded worksheet worked on a phone, how to restart a module, and why a login was not sticking. The same forty questions, over and over, worded a hundred different ways.
The quieter cost was refunds. When a student hit a wall on a Saturday and did not hear back until Monday, a chunk of them did not wait. They requested their money back before they ever reached the part of the course that delivered the result they came for. Nadia estimated her refund rate hovered around eight percent, and she was almost sure most of it was frustration, not disappointment. She was losing revenue and reputation to a gap she could not staff around, because there was no one else to staff.
What she tried first, and why it fell short
Her first instinct was the obvious one: write better documentation. She built a long frequently-asked-questions page and started sending longer, more detailed replies. It helped a little and hurt a little. The page was so long that students could not find the one line that answered them, and her thorough emails simply took more of her evening to write. She was working harder to produce the same stuck customers. Doing more of a manual thing is the trap solo owners fall into most, and it rarely scales.
The setup that finally moved the needle
The shift came when Nadia stopped trying to answer faster and built something that answered for her. She set up an AI help assistant, which is a chatbot trained only on her own material so it responds in her voice with her actual answers, not generic internet advice. The tools for this are now within a solo budget, from Intercom’s Fin to purpose-built trainers like Chatbase, and even a custom version of ChatGPT fed her documents.
The important part was not the tool. It was what she fed it. She gathered everything a stuck student might need: her lesson notes, the supply lists, the worksheet instructions, and, crucially, transcripts of her own lessons. Pulling clean transcripts out of her videos was itself an AI job, the same kind of research-assistant work we described in our look at NotebookLM as the research assistant solo owners have been missing. Then she embedded the assistant on her course pages and told it, in plain instructions, to answer only from her material and to hand off to email when it was unsure.
What changed over the next two months
The results, in this illustrative example, tracked what solo owners tend to see when self-serve support finally works:
- Roughly seventy percent of incoming questions were resolved by the assistant instantly, at any hour, without Nadia touching them.
- Her daily support time fell from around two and a half hours to under forty-five minutes, spent mostly on the genuinely tricky cases the assistant escalated.
- Her refund rate drifted from about eight percent down toward five, because weekend frustration stopped turning into Monday refund requests.
On a couple thousand students at 149 dollars, shaving three points off refunds is not a rounding error. It is real money she was already losing, quietly, to slow replies. And she got her evenings back as a side effect, not the goal.
What you can take from Nadia’s setup
You do not sell candle courses, but the pattern travels. Three lessons carry over to almost any solo business:
Answer the boring questions once, permanently. If you are typing the same reply for the third time, that answer belongs in a system, not in your outbox. An AI assistant trained on your material turns your forty repeat questions into forty instant answers.
Speed is a retention tool, not just a courtesy. Much of what looks like dissatisfaction is really delay. Closing the gap between a stuck customer and an answer protects revenue you are currently losing without seeing it. The case for always-available help is one we made in our guide to AI support assistants you can set up this week.
Feed it your voice, and give it an exit. The assistant is only as good as what you train it on, and it should hand off to you when it is unsure rather than guess. That combination keeps it helpful without letting it invent answers. The same “tame the chaos with one system” move rescued the home baker who ended custom-order chaos, and it can keep your inbox sorting itself too.
Start small this week. Collect your most common questions, write the honest answers once, and put them somewhere a customer can reach at midnight without you. You are probably losing sales to silence right now. The good news is that silence is one of the cheapest problems AI can fix.
