Every store that turns on claim automation asks the same question in the first week: how high do we set the limit? The honest answer is “lower than you think, then higher than you expect.” This is the long version of that answer.
In BuySure, automation is two controls per store. A toggle turns it on. A monetary limit says how much a qualifying, low-risk claim may be worth and still resolve without a person. Everything above the limit, and everything the AI is not sure about, goes to review. The toggle is easy. The limit is the decision.
- The six checks that happen before the AI approves anything
- A method for choosing the starting limit from your own last ninety days
- How to raise it in steps, and what to watch while you do
- Why the AI recommends but never rejects
What “qualifying” means
Before BuySure AI approves a claim, it confirms that all of the boring things are true. Boring is the point. Most fraud, and most honest mistakes, fail one of these:
- The order exists, was fulfilled, and is inside the claim window.
- The customer’s email matches the order.
- Evidence is attached, and it looks like what the claim describes: a cracked dropper in a damage claim, an empty box in a missing-item claim.
- The carrier data agrees. A stuck claim has stalled scans. A stolen claim has a delivered scan past the waiting period. A damage claim has an arrival.
- The customer’s claim history and shipping address do not show a pattern.
- Nothing about the claim is a duplicate of one already filed.
Those checks become a risk score from 0 to 100, with the reasons listed. Only a low score, on a claim under the limit, resolves automatically. A high score never does, at any value.
An automatic approval costs you a replacement. An automatic rejection costs you a customer, a chargeback and a review. That is why one of them is allowed and the other is not.
Choosing the starting limit
Start with the value below which you would not want a person to spend ten minutes. That sounds glib, but it is the real question: what is a claim worth before it deserves human attention?
For most stores the answer lands between the average order value and twice it. A skincare store with a $70 average order might start at $100. An electronics store with a $400 average might start at $150 and keep every device in review while accessories flow through.
Then look at your last ninety days of claims and ask what share falls under the number you picked. If it is under half, the limit is not doing much work; you will still be reading most claims. If it is over ninety percent, you have handed the machine nearly everything on day one, which is fine once you trust it and premature before.
A worked example
An apparel brand with a $95 average order looks at ninety days of claims: 140 claims, median value $88, a long tail up to $420 on multi-item orders. They set the limit at $120. That covers 74% of claims by count and 41% by value. The remaining 26% by count, which is 59% of the money, goes to a person. Two weeks in, they read every auto-approved claim on the timeline and would have approved all of them. They raise the limit to $160. That covers 85% by count. They stop there for a quarter.
Raising it safely
- Run a month with the limit low. Read every auto-approved claim on the timeline. Each one shows its score and its factors.
- If you would have approved them all yourself, raise the limit by a step, not a leap.
- Keep quality-issue claims routed to your team regardless of value. Those are product judgements, and the AI’s job is to bring you the evidence, not to decide whether a serum “did not work”.
- Watch the score distribution. If high-risk claims start appearing under the limit, the AI is already routing them to review. That is the system working, not failing.
Why rejections are never automatic
A wrong approval costs the store one replacement. A wrong rejection costs a customer, often a chargeback, sometimes a public review, and the trust of everyone who reads it. The asymmetry is enormous. BuySure AI recommends, routes and explains. It does not say no. A person says no, with the score and the reasons in front of them, and the timeline records who and why.
This is also why “AI decides everything” is not the goal. The goal is that people spend their judgement on the claims that need judgement, and the machine handles the ones that do not.
What the customer sees
Nothing about any of this. An auto-approved claim looks, to the customer, exactly like a claim a kind and efficient person approved quickly: a reference, a resolved status, and a replacement order or credit that already exists in Shopify. They never see a score. They see an answer.
Questions
Can I set different limits for different claim types?
The limit is per store. Routing rules decide which claim types skip automation entirely and go straight to review, which is usually the right treatment for quality issues.
Can an auto-approved claim be reversed?
The timeline records it and a person can reopen the claim, but a refund already issued in Shopify is issued. That is the reason to start low.
Does the AI learn from my overrides?
Your decisions are recorded on the claim and inform routing for your store. Rules you set always win over the recommendation.
What if I want no automation at all?
Leave the toggle off. Every claim then lands in your queue with the risk score and recommendation attached, and you decide each one. Many stores run that way for a month before turning it on.
