Think Build Implement Repeat
London, UK +44 7367 067226
WhatsApp FOLLOW f in X
  1. Home
  2. Blog
  3. We Keep Finding Out an Order Was Fraudulent After It Has Shipped. How Do We Catch It Earlier?
Problems We Solve

We Keep Finding Out an Order Was Fraudulent After It Has Shipped. How Do We Catch It Earlier?

Chargebacks and bad orders found after dispatch cost stock and fees. We build machine learning order scoring that holds risky orders before they leave.

Updated 3 min readBy SpiderHunts Technologies

Free estimateNo obligation

Get a free estimate

Tell us what you need. A senior engineer reads every enquiry.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →

Quick answer — TL;DR

Fraud is found after dispatch when checks rely on a person spotting something odd, or on generic payment provider rules that do not know your business. A model trained on your own past orders, including the fraudulent and problem ones, can score each new order before fulfilment and send the risky few to a review queue, leaving normal orders to go straight through.

The chargeback arrives weeks later

The order looked normal. It was paid, packed and shipped. Weeks later a chargeback notice arrives: the card was stolen. You lose the goods, the payment and a fee on top. Or a different problem: an account that places large orders on credit and never pays, a reseller exploiting a promotion, or a string of orders to addresses that turn out to be forwarding services.

Each time, someone looks back and sees the signs: a new account, an address that did not match, an unusual mix of products, a rushed delivery option. But nobody was looking at the time, because there are hundreds of orders a day and the warehouse needs to get them out.

Why fraud gets through

Payment providers such as Stripe, PayPal and Shopify Payments run their own fraud checks, and they help. But they judge the payment, not your business. They do not know that your normal customer is a trade buyer ordering the same lines each month, or that nobody genuinely orders five of a high-value item with next-day delivery to a new address.

Manual checks do not scale, and fixed rules either catch too little or flag so many good orders that staff stop paying attention. The patterns that matter are specific to your products, customers and channels, and they change as fraudsters adapt.

What late detection costs

ProblemCost
ChargebacksLost goods, lost payment and fees
Promotion abuseDiscounts taken by people they were not meant for
Bad credit accountsGoods shipped on terms that are never paid
Too many false alarmsGood customers delayed and annoyed
High chargeback ratesRisk of penalties or restrictions from payment providers

False alarms are a cost too. Holding a genuine customer's order for a check they did not expect can lose a sale as surely as fraud loses one.

How we score orders before they ship

  1. We gather order history from your e-commerce platform or ERP (Shopify, WooCommerce, Magento or a custom system) together with chargebacks, refunds, write-offs and any orders your team flagged as suspicious.
  2. We build features that describe each order in context: account age, order history, address match and type, product mix, order value against the customer's normal, delivery choice, device and time patterns where available.
  3. We train a model on your past good and bad orders, handling the fact that fraud is rare, and test it on a later period it has not seen.
  4. We connect it to the order flow through the platform's API or webhooks, so each new order is scored before it reaches fulfilment.
  5. Low-risk orders go straight through. The riskiest few are held in a review queue with the reasons shown, such as "new account, high value, address mismatch", so a person can check and release or cancel quickly.
  6. Review decisions and later outcomes feed back into the model, and we monitor it so it keeps up as patterns change.

You choose where the review threshold sits, trading how many orders a person checks against how much risk gets through. We show you what each setting would have done on your past orders.

What the team works with

A short queue of orders that genuinely deserve a second look, each with the reasons laid out. Everything else ships as normal. Staff stop scanning every order by eye and focus on the ones that matter.

Chargebacks and bad debts show up in the data as they happen, so you can see whether the pattern is changing and adjust.

The reasons shown in the queue also help with the other side of the problem. When a genuine customer is held, staff can see why and release the order quickly, and those decisions teach the model which patterns are harmless in your business.

Is this your situation?

  • You regularly discover fraud or bad orders after they have shipped.
  • Chargebacks or unpaid trade orders are a recurring cost.
  • Your payment provider's checks do not catch the patterns you see.
  • Staff check orders by eye, or not at all when it is busy.
  • Manual rules flag so many good orders that people ignore them.

FAQ

Frequently asked questions

The questions readers ask us after this guide.

Still have a question?

Ask us directly — a senior engineer will get back to you.

Ask about your project

Does this replace our payment provider's fraud checks?

No. It adds a layer that understands your business, and uses the provider's risk signals as one of its inputs where they are available.

We have very few fraud cases. Is that enough?

Rare events are hard to learn from. With few examples, we start with scoring based on how unusual an order is for your business, and add supervised learning as cases build up.

Will it delay genuine orders?

Only those above the review threshold, and you set that. The aim is to hold as few good orders as possible.

What drives the cost?

The order volume, the number of sales channels, and how the scoring connects to your order flow.

What do you need from us?

Order history, chargeback and refund records, and any orders your team already marked as suspicious.

Keep reading

More on Problems We Solve

Start here

Tell us what you are guessing today

Describe the decision, the data you keep and how it is made now. We will tell you honestly whether machine learning would help, what it would need, and if a simpler fix would do the job, we will say so.

  1. You tell us what you needTwo minutes on the form, or a message on WhatsApp.
  2. A senior engineer reviews itAnd comes back with questions, a realistic range and an honest view on fit.
  3. Free 30-minute scoping callWe talk through scope, options and a realistic estimate — with no obligation.
Free estimateNo obligation

Talk to someone who builds this

Send a short brief and we will come back with an honest view and a realistic range.

Takes under a minute. We never share your details.

  • Free consultation
  • No commitment
  • NDA on request

Prefer to talk? Book a free 30-minute call →