Chargeback defense for stores

Use a strong model, because this pipeline involves judgment calls and a weak model makes expensive mistakes with real customers and real money.

After that, give it the keys for the job. In chat, tell the agent which accounts it's working with and paste the credentials when it asks: a Google Places API key from Google Cloud Console with the places API enabled and a budget cap on it, a postcard API account with Lob, PostGrid, or Stannp, and a Vercel AI Gateway key for Jev, which the next step covers. Once a store signs you'll also connect their Stripe or Shopify account so the agent can see their disputes and orders. On Grok bot you can also log into a website once in the bot's browser and it stays logged in.

Jev is a decision model from typesafe. The agent does the gathering and the writing, and Jev is the part that decides. You give it a situation, like one dispute with its reason code, the order, and what evidence the store has, and you ask it a question with a fixed set of answers. It picks one and gives you a probability for how sure it is, from 0 to 1, and the agent only acts when Jev is sure and sends everything else to you.

You get Jev through Vercel AI Gateway. Make a Vercel account, open AI gateway from the dashboard, create an API key, and paste it into the agent when it asks. The model name is typesafe-ai/jev, and the agent calls it the same way it would call any other model. It costs almost nothing to run, when I tested all three of the results-based services in this series, 472 Jev decisions came to $0.0735 in total.

A Jev question for a dispute looks like this:

Before building anything else, have the agent make a few test disputes and run them through Jev so you know the connection works.

Every dispute comes with a reason code, which is the reason the customer gave their bank, and each reason code has its own evidence that wins. The rulebook is that list written in plain words, and the agent turns it into questions for jev. This is the one I used:

For every dispute Jev answers four questions. Fight, accept, or gather more evidence. How likely a win is, low, medium, or high. Is the evidence complete. And is the deadline at risk. The agent only prepares a fight when Jev is more than 0.8 sure and the win chance is at least medium, and everything else comes to you with Jev's answers next to it.

The line about evidence that exists at the time you submit came from my own test. My first rulebook asked for a document that can't exist until after you've submitted, and Jev kept flagging disputes as incomplete until I fixed the line. Every mistake you find gets added to the rulebook the same way.

When Jev says fight, the agent writes the response the card network actually reads. It's short and factual: what was ordered, when, how it was paid, where it was delivered, and then each piece of evidence with a one-line note on what it shows. The evidence goes in as attachments, like the tracking page, the order confirmation, and screenshots of the customer's messages, and nothing gets listed that the store doesn't actually have.

This fits businesses that take a lot of card payments and ship or deliver things, so pull furniture stores, home goods stores, boutiques, jewelers, and restaurants that take online orders. Filter for a rating of 4.0 or above, a phone listed, and a website, and put the ones with an online shop first, because online orders are where most chargebacks come from.

You don't have their real disputes yet, so the agent makes three mock disputes that fit the store's type, like a furniture order marked as not received, a jewelry order marked as fraud, and a clothing order marked as not as described. It fills in the evidence checklist for each one, runs them through Jev, and builds a page showing Jev's verdict and how sure it is next to each, plus the full response for the strongest one.

At the top of the page is their name and logo, and at the bottom is the total that would have been won back. Mark it clearly as mock data and end with one line: send us your last 3 disputes and we'll run them for real, free.

Postcard APIs like Lob, PostGrid, and Stannp mail real printed cards from an API call for a dollar or two each. The front is a screenshot of the demo page with the verdicts showing, and the back says "we set up chargeback fighting for [store name]. Scan to see which disputes you'd win." With the QR to the demo page and your phone number. Store owners also live in their Instagram DMs, so the same link can go there too.

The agent watches the demo links and messages you when one is opened. A store owner who looked at the page for a while has almost always lost a dispute recently and remembers it, so call them the same day and ask if they've had any chargebacks lately.

The two questions you'll get are how you get paid, which is a cut of what you win back and nothing if you lose, and whether they need to give you access to anything, which for the free first job is just the details of their last few disputes.

When a store says yes, ask for their last three disputes, which they can see in their Stripe, Shopify, or Square dashboard, plus the order details for each. The agent runs them exactly the same way as the demo, pulls whatever evidence the store has, and shows you the verdicts and the response for every fight before anything is submitted. If a dispute is still open and inside its deadline, that one gets fought for real straight away.

When I tested this on 30 disputes for a home goods store, run twice, Jev got 27 of 30 right both times and called the deadline risk correctly on all 30. it never told me to fight a dispute that couldn't be won and never told me to give up on one that could, and the three it got wrong were disputes it sent to gather more evidence, which cost nothing but a second look.

Once a store signs, connect their Stripe or Shopify account so the agent sees new disputes the day they come in, along with the order details, tracking, and customer messages it needs as evidence. Then put it on a daily schedule, which on Grok bot is a routine and on Hermes is a scheduled task. Every morning it checks for new disputes, asks Jev, builds the response for every fight, and shows it to you well before the deadline.

When a dispute is won, the money goes back to the store, and the agent keeps a running total of what was won. At the end of each month it sends you and the owner a statement listing every dispute, what happened, and the total won back, and you bill your cut from that.

Every new store is the same skill with a copy of the rulebook and their own account connected, so one agent can run all of them. The daily routine loops through each store in turn, and your review list shows which store each dispute belongs to. This also stacks well with guide 8, because a store that's losing money to chargebacks usually has other money going missing too.

Once every step above works on its own, this is the whole pipeline in one message. Paste it in, fill in the brackets, and the agent saves it as one skill you can rerun on a new city with a single line.