A clear eligibility window
Pick a number of days from purchase and stick to it. Seven, fourteen, or thirty are the only sensible options for a digital product.
How-to — — by Mahmoud Zalt
Founder playbook for how to handle refund requests without killing your day: clean policy, AI triage, short replies, and a flow that protects focus.
Refunds do not cost a founder time, they cost a founder mood. A normal support ticket is a transaction, but a refund email lands as a small public verdict on the product you built, and the brain reads that verdict three or four times before it can write a calm reply. The result is a ticket that should take five minutes turning into an hour of rereading, drafting, deleting, and finally typing a stiff response that nobody enjoys. Multiply that by two or three refunds a week and you have lost an entire founder afternoon to feelings, not work. The fix is not to grow a thicker skin, it is to remove yourself from the parts of the loop that do not need a human at all and to standardise the parts that do. That is the entire game.
A small business refund policy should fit on a single screen and answer five questions before a customer ever has to write to you. It should state who qualifies, what the window is, what is excluded, how the refund is delivered, and how long it takes to land. Most founder rage starts because one of those five is missing, so the customer fills the blank with the most generous interpretation and you fill it with the strictest, and you fight in the middle. Write the policy once, link it from the footer, the receipt email, and the checkout page, and quote it verbatim in every reply. The policy is the wall, your job is to point at it kindly, not to defend it each time. A clear policy lets an AI Employee handle the first reply because the rules are no longer in your head.
Pick a number of days from purchase and stick to it. Seven, fourteen, or thirty are the only sensible options for a digital product.
List the cases you will not refund, such as partial monthly usage or one-off custom work, in two short sentences each.
Refund to the original payment method only. No store credit haggling, no exception chains, no Venmo side deals.
Promise a two-business-day reply, not a same-day miracle. Predictable beats fast and broken every week.
Footer, receipt, dashboard, and the AI Employee context. Same URL everywhere so customers and your agent quote the same words.
Yes, with one rule: let the AI Employee do the reading, the writing, and the staging, but keep the final click on the Stripe refund button human until your volume justifies more. A pre-built customer support employee can pull the order details, check the policy against the purchase date, classify the request as eligible, borderline, or out of window, draft a reply in your voice, and queue the refund as a draft action in Stripe. You then open a daily ten-minute batch and approve, decline, or edit. The mistakes you fear, refunding the wrong charge, refunding twice, refunding a fraud case, all happen at the click, so as long as the click stays with you, the model can drive the rest of the workflow safely. That setup is what most founders actually want.
| Dimension | Traditional | With Sista |
|---|---|---|
| Time per refund | Thirty to sixty minutes including emotional cost | Under two minutes to approve a pre-drafted reply |
| Reply tone consistency | Varies with the founder mood that day | Pinned to the policy voice, every time |
| Policy lookup | Reread the policy page each ticket | Policy lives in the employee memory, quoted automatically |
| Stripe action | Founder clicks refund, types reason, sends email | Refund prefilled in Stripe, reply queued, founder approves |
| Audit trail | Scattered across inbox, Stripe, and memory | Ticket, policy reference, and refund linked in one record |
The shift here is not automation for its own sake, it is moving you up the stack. You stop being the typist on refund mail and start being the editor on a small daily approval queue. The work the customer notices, the actual refund and the actual reply, still gets your fingerprints on it, but the part of the work that drained you, the rereading and the drafting from zero, disappears. Most founders feel the difference inside the first week and stop checking the inbox between batches. That is the real win.
Before we go into the reply template and the clean flow, one honest framing. The refund queue is one of the few places where speed actually helps you keep customers. A polite, quick no on a borderline case loses fewer customers than a slow yes, because the slow yes still trains the customer to feel ignored. The AI Employee is not there to refuse more refunds, it is there to make the response time predictable so the relationship survives even when the answer is not the one the customer wanted.
Length is the enemy of refund conversations. Every extra paragraph reads like a defence, every defence invites a counter, and a thread that should end in two messages stretches to seven and ends with everyone in a bad mood. The trick is to write one reply that does five small jobs in five short sentences: acknowledge, restate the request, quote the relevant policy line, give the decision, and offer one clear next step. That is the template. The AI Employee can fill it from the ticket in under a minute, and your job is to scan for the one thing only a human notices, which is whether this particular customer needs a softer or firmer tone.
A clean refund flow is not a complicated system, it is a small loop that keeps you out of the ticket until the last possible moment. Inbox stays quiet, policy stays linked, AI Employee stays in the driver seat, and you stay focused on the work that compounds. The flow below is the one I run on my own business and it has cut my refund time per week from about three hours to under twenty minutes, with no measurable change in customer happiness. The trick is to commit to the batch and to resist the urge to peek between batches, because the peeking is where the day gets killed, not the refunds themselves.
In the first weeks, yes. Keep the click on the Stripe refund button until the AI Employee has drafted at least fifty replies and you trust the classification. After that, let it auto-approve refunds inside the policy window under a small dollar cap, and route only borderline cases to you. The split saves the most time without giving up the safety of a human on the edge cases.
It can flag patterns a tired founder misses. Repeated refunds from the same card, refunds requested minutes after a heavy usage spike, accounts created the day of purchase, and language patterns common in chargeback farming. The AI Employee marks these as borderline rather than deciding, so you see them in the daily batch with the evidence inline and a one-line summary of why it flagged the case.
Refunds are a conversation, chargebacks are a dispute. If a customer skips your inbox and goes straight to their bank, the AI Employee opens a chargeback record, gathers the receipts, usage logs, and policy link, and prepares the Stripe response pack. You review and submit. The rule of thumb is to refund early if the case is borderline, because winning a chargeback costs more in time and fees than the original sale.
Reply once, publicly and politely, then move the conversation to email. The AI Employee can draft the public reply in your voice using the policy line, and queue a private email that offers a final resolution. Never debate the case in public. One short, calm public post protects your brand more than a long defence, and the email closes the loop without an audience.
Stripe handles the mechanics, the refund itself, the email receipt, and the dispute pipeline, but it does not classify the request or write the reply. That is where the AI Employee earns its keep. Connect the support employee to Stripe and your inbox, and the loop closes: ticket in, policy check, draft reply, refund queued, founder approves, Stripe executes, audit logged. One flow, two clicks, no day lost.
If you want to take this further into the wider customer support function, the next read is the practical companion to this refund flow. It covers the full support setup most solo founders need: ticket routing, first-reply drafting, knowledge base hooks, and the small set of automations that keep an inbox quiet without making customers feel ignored. Refunds are the loudest slice of support work, but the same pattern, AI for the reading and drafting, founder for the approval, applies to every other type of ticket in the inbox.
The honest takeaway is that refund requests will never become enjoyable, but they can stop being expensive. Once the policy lives in one place, the AI Employee handles the reading and the drafting, and you commit to a single daily batch, the cost of a refund collapses from an emotional event into a small administrative click. You still own the call on the borderline cases, you still feel the small sting when a customer leaves, but you no longer lose an afternoon to it. That is the realistic shape of a refund workflow for a solo founder in this category. Build the policy this week, hire the support employee, and run the daily batch for a month before you judge it. Most founders never go back to the old loop after that month, because the new one returns the most valuable thing the old one took: focus.