Chargebacks and Refund Requests: Writing Policies AI Can Apply Consistently

Refunds and Chargebacks Are Different Problems

A refund is something your organization chooses to do. A donor asks, you decide, and you return some or all of the money through your payment platform. You control the timing, the communication, and the record. A chargeback is something that happens to you. The donor contacts their card issuer and says the charge was wrong, unauthorized, or not what they agreed to. The issuer reverses the transaction, your processor debits your account, and you are given a window to accept the loss or contest it with evidence. You do not control the timing, and the issuer, not you, makes the final call.

Cardholders have strong rights in this process. For credit cards, the federal Fair Credit Billing Act gives consumers the right to dispute billing errors in writing, generally within 60 days of the statement on which the charge appeared, as the New York City Bar's consumer guide explains. Card network rules often allow longer windows than the statute, and issuers frequently accept disputes by phone or app with very little friction. From the donor's side, disputing a charge can feel easier than writing to a small nonprofit and waiting for an answer. That is the core insight behind a good policy: every refund request you handle quickly and fairly is a chargeback that never happens.

Charitable gifts add a wrinkle that retail businesses do not face. A completed gift is generally the donor's to give and the organization's to keep, and many nonprofits state that donations are non-refundable. That is a reasonable default, but it is a default, not a wall. Donors make typing mistakes, click submit twice, misunderstand what a recurring checkbox meant, buy tickets to events that get cancelled, or have their card used by someone else. A policy that says "no refunds ever" does not prevent those situations. It just pushes them toward the card issuer, where you lose control of the outcome and pay a fee besides.

So the policy has two jobs. The first is to decide, in advance and in writing, which refund requests you will honor, under what conditions, and who approves them. The second is to decide how you respond when a dispute arrives anyway: when you accept it, when you fight it, and what evidence you assemble. Both jobs benefit from AI support. Both require people to make the calls.

The Scenarios Your Policy Has to Cover

Every organization's mix is different, but a small handful of request types account for most of the refund traffic at a typical small nonprofit. Thinking each through in advance is where the policy earns its value.

Accidental duplicate gifts and mistyped amounts

These are the easiest cases and the most common. A donor clicks submit twice because the confirmation page loaded slowly, or types $1,000 when they meant $100. The fair answer is almost always a prompt refund of the error, and the policy can say so without requiring a senior approval for amounts under your threshold. The recognition rule should be specific: same donor, same or nearly same amount, within a defined period, with the donor confirming the mistake. Your records need to show which transaction was kept, which was refunded, and that only one receipt stands. If you see a cluster of tiny duplicate charges from many different "donors," that is not this scenario at all; it is likely card testing, which needs the response described in our bot fraud guide.

Recurring gift cancellations

Recurring gifts are the backbone of many small nonprofits' revenue and the most common source of avoidable disputes. A donor sets up a monthly gift, forgets about it, sees an unfamiliar descriptor on a statement a year later, and calls the bank. Your policy should separate two questions: stopping future charges, which you should always do immediately on request and without argument, and refunding past charges, which is a policy choice. Many organizations refund the most recent charge if the donor asks promptly and says they meant to cancel earlier, and decline to refund older charges that the donor was reminded about. Whatever you choose, write it down and apply it the same way to everyone.

The legal landscape here is in motion. The FTC's "click to cancel" rule, which would have required simple online cancellation for many recurring charges, was vacated in its entirety by the Eighth Circuit in July 2025, as Cooley's analysis describes. Federal law such as the Restore Online Shoppers' Confidence Act still applies to online negative-option offers, and many states have their own automatic renewal laws. Whether a given state's law reaches a recurring charitable gift depends on that statute's definitions, which is a question for counsel. The practical answer does not depend on the legal one, though: make cancellation as easy as sign-up, confirm it in writing, and send periodic reminders that the gift is ongoing. Our article on recurring donor success covers the retention side; easy cancellation and good retention are not in conflict, because donors who trust they can leave are more comfortable staying.

Event tickets and gala purchases

Event payments are partly a purchase and partly a gift, and the refund rule should reflect both. If the event is cancelled, most organizations offer a refund or the option to convert the payment into a donation, and the policy should state how the donor chooses and by when. If the donor simply cannot attend, a common approach is no refund but an offer to treat the ticket as a gift, with the receipt adjusted accordingly. The tax side matters here: when a donor pays more than $75 and receives goods or services in return, the IRS requires a written disclosure with a good-faith estimate of the value received, as the IRS explains in its guidance on quid pro quo contributions. A partial refund or a conversion to a pure gift changes that math, so the receipt has to change with it.

Donor regret and changed circumstances

Sometimes a donor simply wants the money back: a job loss, a family emergency, a disagreement with something the organization did. These are the hardest cases because there is no error to correct, and they are exactly where consistency matters most. A sound policy treats them as exceptions that always go to a named senior person, with the factors that person weighs written down: how recent the gift was, whether it has been receipted or spent, whether it was restricted, and whether a refund would set a precedent the organization can afford. An AI tool should never decide these. It can gather the facts and draft a compassionate response in either direction, but the decision belongs to a person.

Restricted gifts, matching gifts, and gifts already spent

A refund request for a restricted gift that has already been spent on its purpose, or a gift that was matched by an employer or a challenge donor, creates obligations to third parties. The policy should route these to the finance lead every time, because the answer may involve notifying the matching source, adjusting restricted net asset balances, or explaining why the gift cannot be returned. Gifts made through donor-advised funds are a separate category altogether; they are grants from the sponsoring organization, not the individual, and requests to reverse them should be referred to the sponsor rather than handled as a card refund. If your organization has a gift acceptance policy, the refund policy should cross-reference it so the two documents agree on what a completed gift is.

How Card Disputes Work, and What They Cost

When a donor disputes a charge, the issuing bank assigns it a reason code that falls into a few broad families: fraud (the cardholder says they did not authorize it), authorization or processing errors (a duplicate charge, a wrong amount), and consumer disputes (the cardholder says they cancelled, did not receive what they paid for, or the charge was not as described). The reason code matters because it determines what evidence can win. Proving that a donor received a thank-you email does nothing against a fraud claim, while proving that the same card had given several times before without complaint might. Your processor's dashboard shows the reason code, and your response should be built around it.

Disputes cost money even when you win. Most processors charge a flat fee per dispute. Stripe, for example, charges a $15 fee in the US when a dispute is opened and, since mid-2025, an additional fee if you submit evidence and lose, as one platform's support documentation describes. Other processors and donation platforms set their own fees, and some pass through the processor's fee unchanged. Check your own agreement. For a $25 gift, the fee and the staff time to respond can exceed the gift itself, which is why the policy should say explicitly when contesting is worth it.

Ratios matter more than any single dispute. Card networks monitor merchants whose disputes, measured against their transaction counts, cross certain thresholds, and processors monitor their own portfolios even more tightly. Visa's acquirer monitoring program lowered its merchant "excessive" threshold to 1.5 percent in the US on April 1, 2026, according to industry trackers of the change, and the program also sets a minimum monthly volume of disputes before it applies. A small nonprofit is unlikely to hit network thresholds through ordinary refund disputes, but card testing attacks can produce bursts of fraud reports that do, and processors may act on lower numbers than the networks. Ask your processor what thresholds it applies to your account, and treat a rising dispute count as an early warning rather than a curiosity.

Two operational points are easy to miss. First, once a formal dispute is open, refunding the donor directly usually does not close it and can result in the donor being repaid twice; follow your processor's instructions for accepting the dispute instead. Second, many processors offer early alerts or pre-dispute notifications that let you refund before a formal chargeback is filed, which can avoid the fee and keep the dispute out of your ratio. If your platform offers these, the policy should say who monitors them and how fast they respond.

Putting AI to Work on Triage and Donor Replies

Once the policy exists in decision-rule form, AI can take on the repetitive work around it. The most valuable task is triage. Refund requests arrive through the general inbox, the contact form, replies to receipts, phone messages transcribed by staff, and social media. A model given the policy text and a request can classify it by type, extract the relevant facts (donor, amount, date, platform), identify which rule applies, note what information is missing, and propose the action the rule calls for. It can also flag urgency: a request that mentions disputing the charge with the bank should jump the queue, because answering it today may prevent a chargeback tomorrow.

The second task is drafting replies. Donors asking for refunds are often embarrassed, frustrated, or both, and the tone of the response shapes whether they give again. A model can draft a reply that applies the policy, explains the outcome plainly, thanks the donor, and gives next steps, in your organization's voice. For declines, it can explain the reason and offer an alternative, such as converting a ticket to a gift. Keep a small library of approved templates for each request type, and have the model adapt them rather than writing from scratch, so the language stays consistent. A person reads and sends every reply; nothing about a donor's money should go out automatically.

The third task is consistency checking, and it is where AI's patience is most useful. Once a quarter, export a log of refund decisions with the request type, amount, outcome, and approver, and ask a model to compare each decision against the policy. Where did the same type of request get different outcomes? Which approvals were outside the stated thresholds? Which exceptions were granted, and do they share a pattern, such as favoring larger donors or people known to staff? The model surfaces the questions; the finance lead and executive director decide whether the policy or the practice needs to change. Over time this review is what keeps "consistently" from being an aspiration.

Practical setup matters more than tool choice. Many teams start with a general AI assistant and a saved prompt that contains the policy, the templates, and instructions to return a structured triage note. Teams with more volume may connect the assistant to the CRM or helpdesk, as described in our guide to AI tools for donor thank-you workflows, which uses the same pattern of AI drafting and people approving. Either way, the policy document is the single source of truth. When it changes, the prompt changes the same day.

Keeping People on the Decisions

It is tempting, once triage and drafting work well, to let the system close the loop: classify the request, apply the rule, issue the refund, send the email. Resist it. Refunds move money out of the organization, and every control your auditors care about assumes a person authorized that movement. Even where a rule is mechanical, such as refunding a confirmed duplicate, a staff member should click the button and own the decision. That keeps accountability where it belongs and keeps an AI error, such as misreading a date or matching the wrong donor, from turning into a payment.

Some decisions should never be delegated to the routine process at all. Exceptions for donor hardship, refunds of restricted or matched gifts, disputes involving major donors or board members, decisions to contest a dispute above a set amount, and anything that touches a prior fiscal year belong to named senior people. AI can prepare the file for each of these, laying out the facts, the relevant policy sections, and the options. It should not recommend an outcome on a hardship case, because those decisions carry values and relationships a model does not hold.

Governance closes the loop. The board or finance committee should approve the refund and chargeback policy, see a short annual summary of refunds and disputes, and know the thresholds at which they will hear about an individual case. Staff who handle donor money should be trained on the policy and on what they may and may not put into AI tools. If your organization has an AI acceptable use policy, add a line that names refund triage and dispute drafting as approved uses, with the data limits spelled out. If you are preparing for a financial audit, the decision log, approval records, and corrected receipts are exactly what your auditors will ask to see, and having them organized is part of good audit preparation.