Your dispute ratio came back higher this month and the only thing that changed was ad spend. The obvious read is that scaling broke something in the offer. Before you cut budget on that theory, look at what the number is measuring, because the disputes on top of that fraction and the sales underneath it almost never come from the same month. There are 6 causes that print as the same red number, and a single export that tells you which one you are holding.
Your ratio has 2 different months inside it
A chargeback ratio is a fraction with a reporting date stapled to it. Most processors put the disputes that posted during the month on top and the transactions that settled during that same month underneath. That works while volume is flat. It stops describing anything real the moment volume moves.
Disputes do not arrive with the sale. Visa's published rules give the cardholder up to 120 days from the transaction date, or from the expected delivery date, on most cardholder dispute reason codes. So the disputes posting today are reporting on orders you took 1 to 4 months ago, while the denominator underneath them is reporting on orders you took this week.
Run that through a ramp and the behavior is predictable. In the month you scale, the denominator inflates immediately while the numerator is still counting disputes from the quiet month before it. The printed ratio falls. You look cleaner while scaling than you did standing still. Then you flatten, or a winning creative dies and spend drops back. The denominator shrinks while the numerator is now reporting your peak month. The ratio spikes.
The number spikes after you stop scaling, not while you are scaling. If your worst printed month sits directly after your best spending month, that shape alone is a candidate explanation and it costs an afternoon to test.
Ask your processor which convention they use: disputes counted by the date the dispute posted, or by the date of the original sale. Both exist, and they produce different numbers from identical data. Acquirers and the card network monitoring programs generally act on the posting month version, which is why it can put you under review while every underlying cohort is healthy.
Re-date every dispute before you diagnose anything
There is a single export that separates a real deterioration from an arithmetic artifact, and everything it needs is already in your processor dashboard.
- Pull every dispute from the last 6 months with the original transaction date on the row, not the dispute posting date. Every gateway export carries it, usually labeled original transaction date, sale date or auth date.
- Group the disputes by the month of that original transaction, not by the month the dispute landed.
- Divide each month's disputes by that same month's order count. That is your cohort rate: the real dispute rate of the customers you acquired in that month.
- Chart the cohort rate against the printed rate your processor reports, same 6 months, same axis.
- Mark the last 2 months as incomplete. Their disputes have not finished arriving, so those cohorts always read low and will keep rising for months after you drew the chart.
What the 2 lines are telling you
The lines answer different questions and you need both. A flat cohort line under a spiking printed line means the denominator moved beneath a lagging numerator, and the offer is fine. A cohort line that steps up starting with the month you scaled means something you changed at scale is genuinely producing disputes. A cohort line that was already climbing before you scaled means the problem predates the budget, and scaling only made it loud enough to hear.
Keep the printed rate in front of you either way. It is what your acquirer reads and what the monitoring programs run on, so it can trigger a review whether or not the cohorts underneath it deserve one. The cohort rate tells you whether to change the offer. The printed rate tells you whether to call your acquirer first.
6 causes that print as the same red number
Once the cohort line confirms the rise is real, you are choosing between causes that look identical in a summary dashboard and route to completely different fixes.
- Denominator collapse. You flattened or cut spend, and disputes from the peak month are landing against a smaller sales base. The rate moved. The customers did not.
- A rebill wave. You acquired a trial or subscription cohort while you were scaling, and the first rebill fires on whatever cycle you sold, commonly 30 days out. Part of that cohort disputes the rebill instead of canceling, and the dispute is dated to the rebill, so it reads as fresh trouble on customers you actually bought a month ago.
- A descriptor the buyer does not recognize. New offer, new brand name on the page, an old or default descriptor on the statement. At 40 orders a day the confused buyers email support and it never reaches the dispute file. At 400 that confusion outruns the inbox, and the ones who get no answer call the bank instead.
- Fulfillment falling behind the front end. You scaled the ads faster than the warehouse can pick. Ship dates slip, tracking stops updating, and not received disputes follow on the network's clock rather than yours.
- Traffic that is not your old traffic. A new placement, a broader audience, a new geo or a new creative angle brings in buyers who read the same offer differently and expect something else in the box.
- A fraud attack that found you because you got loud. Scaling puts your checkout in front of everyone, card testers included. It arrives as small orders, mismatched billing details and fraud reason codes, and none of it is a copy problem.
The reason code split tells you which one you have
Every dispute in that export carries a reason code, and the code is the fastest discriminator you have. Split fraud from non fraud first, then read the sub buckets against the disputed orders themselves rather than against the month.
- Fraud codes climbing while non fraud stays flat (Visa 10.4, Mastercard 4837). Check AVS and CVV on those orders. A fraud code where the address and the security code both matched is usually a customer who did not recognize the charge, a descriptor problem wearing a fraud label. Real card testing looks different: failed or absent AVS, odd order sizes, repeat attempts from a narrow set of details.
- Merchandise not received (Visa 13.1, Mastercard 4855). Join those orders to their ship dates and delivery confirmations. If they shipped materially later than your median, this is an operations problem and no creative work touches it.
- Cancelled recurring transaction (Visa 13.2). Count how many of these customers reached your cancel page or emailed support before disputing. If most never tried, the cancel path is hidden. If most did try, the support queue is the leak.
- Not as described (Visa 13.3, Mastercard 4853). Trace those orders back to the ad and the landing page they came from. This bucket is a direct readout of the distance between the promise in the creative and the thing in the box.
- Credit not processed (Visa 13.6). Refund turnaround, not marketing. A customer who was told a refund was coming and never saw it files against you.
- Days to dispute, measured from the original transaction. Short dated disputes cluster around recognition and fraud. Long dated ones cluster around delivery, rebills and expectation. That distribution splits the file faster than reading codes row by row.
What each branch actually asks you to fix
Each branch routes to a different desk, and running the wrong fix costs you the month.
- Denominator artifact: nothing in the offer changes, the shape of your spend does. A hard ramp followed by a hard cut prints the worst ratio available to you, because it stacks the peak month's disputes on the quiet month's sales. Step spend instead of swinging it, and call your acquirer before the letter arrives.
- Rebill wave: fix the pre rebill notice and the cancel path, and start logging the evidence fields the networks ask for at the time of sale. You cannot go back and collect consent records after the dispute lands.
- Descriptor: change it to the name on the page and in the confirmation email, and add a contact string if your processor allows the characters. Cheapest fix on this list, and the only one where the effect starts on the next order instead of on the next cohort.
- Fulfillment: cap daily spend at the number of orders the warehouse can actually ship on time. Every day a cohort sits unshipped is a cohort you already paid to acquire and will pay for again when it disputes.
- New traffic: this is a creative and page problem, and it is the same problem an ad reviewer is looking at. The claim on the ad has to survive the customer opening the box, the same standard behind what reviewers check on a restricted offer page.
- Fraud attack: velocity limits, address checks and step up authentication on the risky slice. Do not rewrite the landing page for this one. The page is not what generated those orders.
The ad account is downstream of this and it breaks first
A dispute problem does not stay inside the processor. It reaches the ad account by 2 routes, and the slow one is the only one most operators plan for.
The slow route is cash. An acquirer that puts a reserve on you holds back a share of your settlements, so the money for next week's ads is sitting in someone else's account. If the card on your Meta account then declines at the billing threshold, the account gets restricted, and you are holding a dispute problem and an advertising outage in the same week. How billing thresholds and account spend limits actually move is worth knowing before you meet them.
The fast route never touches your processor. Meta runs its own buyer feedback signal for ecommerce advertisers, built from surveys sent to people who bought through your ads, not from any chargeback file. The 3 things driving most of your disputes, a late shipment, a statement line nobody recognizes and a box that does not match the ad, are the same 3 things those buyers report. Delivery can get more expensive on that signal alone while your processor is still counting.
Both routes argue for provisioning capacity while everything is healthy. Redundancy stood up in advance is ordinary operations. Standing up a lookalike account on the same card, pixel, domain and device after an enforcement action is circumvention, and it ends differently. If you have never moved live spend onto a second account under time pressure, the sequence for moving a campaign when an asset goes down is worth rehearsing while nothing is on fire.
All of it has a prerequisite. You cannot join a dispute back to an ad if the order does not carry the campaign it came from. With no shared key between your dispute export and your ad reporting, the reason code exercise stops at the processor and you are guessing which creative did it. Passing a click identifier through to the order and keeping a campaign structure you can read backwards is what makes the diagnosis possible.
Know where the payment relationship sits before a reserve puts you under pressure. If your ad balance is pre funded and held by a provider, an acquirer holding your settlements traps your cash in 2 places at once. An arrangement where your own card sits on the account and Meta bills you directly removes 1 of those 2 holds, and where the billing relationship actually sits is a fair thing to ask any provider before money moves.
What this costs you in the month it happens
Start with the honest part. None of this is fixed by changing where your ad account comes from. A descriptor mismatch is a descriptor mismatch on any account, a slow warehouse is slow on any account, and the offer generates the disputes. Anyone selling infrastructure as a chargeback fix is selling the wrong thing.
What the infrastructure decides is what a deliberate slowdown costs you. The right answer to a printed ratio you do not like is often to hold spend flat for 30 to 60 days while the peak cohort clears, so the numerator and the denominator come back into the same range. That is a cheap decision on a fixed cost and an expensive one on a variable cost, because a provider taking a percentage of spend has revenue tied to the exact behavior you need to pause.
A flat monthly rate that does not move when your spend does is priced as Gold at $299, Diamond at $599, Platinum at $995 and Titanium at $1995 per month, the same number in the month you scale hard and the month you deliberately do not. The trade off is real: below a certain spend a percentage is cheaper, and a flat rate only starts winning once the cut on your monthly spend exceeds the tier. What changes in a month like this one is not the total. It is that holding flat for a quarter stops costing you anything to exercise.