★★★★★
The automation was set up in a clean and practical way by the Checkout Loom team. It saved me a lot of manual work and made the whole process much easier.
JanoUnited Kingdom
Verified
AI fraud screening for online stores
AI fraud screening for online stores that holds the doubtful orders and ships the honest ones
AI fraud screening for online stores scores each order on payment checks, address and delivery signals, account history and behaviour, explains in plain words why a score is high, holds only the doubtful orders for a person to decide and lets every other order ship, because blocking too little invites chargebacks and blocking too much turns away honest customers, and the balance between them has to be set from your own order history and not from a generic rule.
- Fixed price from $890 with no fee per order
- Every score explained in plain words
- A person decides every hold
- Platforms supportedShopify, WooCommerce and BigCommerce stores, with your payment processor’s own risk data used as one signal among several.
- Typical storeStores with enough orders to see patterns, high-value or resale-prone products or a recent rise in chargebacks.
- Signals usedPayment checks, address and delivery data, account history, order velocity and behaviour on the site.
- DecisionsApprove, hold for review or cancel, with holds placed before fulfilment and every choice logged.
- OutputsA scoring service, a review queue, plain-word reasons, a decision log and a monthly fraud report.
- Delivered inSix to twenty working days depending on the plan, with scoring running quietly on live orders first.
The costly disconnect
Why fraud losses and false declines often rise together in the same store
A store that fears chargebacks tends to block too much, and a store that fears lost sales tends to block too little, and both are guessing without a measured score.
Chargebacks that arrive weeks later
A stolen-card order is shipped, paid for by the seller and then reversed weeks later with a fee, and the goods are gone. The fix is not more effort but a better rule, and that is exactly what the fraud screening build for online stores puts in place.
Honest customers who look odd
A gift shipped to a new address or a traveller paying from abroad can look suspicious, and a blunt rule cancels the sale. Many owners only notice this problem when a customer complains, so catching it earlier is one of the quiet benefits of fraud screening for online stores.
Card testing in bursts
Criminals try many stolen card numbers on low-priced items, which raises fees and can harm your standing with the processor. Left alone, this problem repeats with every order, which is why fraud screening for online stores is built to handle it by rule and not by memory.
Reviews done by a tired person
A manual check of every order does not scale, and the one rushed on a busy day is often the one that matters. It is easy to overlook until a busy week exposes it, and by then the cost to online stores is already counted in refunds, hours or lost sales in the fraud screening work.
Rules nobody revisits
A rule written after one bad week keeps blocking a whole country long after the problem has gone. The fix is not more effort but a better rule, and that is exactly what the fraud screening build for online stores puts in place.
Card network thresholds
Reports say Visa lowered its excessive merchant ratio to 1.5 percent in April 2026, which is lower than the older figure and counts disputes you win, so confirm your position with your acquirer. Many owners only notice this problem when a customer complains, so catching it earlier is one of the quiet benefits of fraud screening for online stores.
- 1.5%is the excessive merchant ratio reported for Visa from April 2026, according to payment industry blogs, to be confirmed with your acquirer.
- 2.1%of orders held for review is a realistic target in our examples, with the rest shipping without friction.
- 4agreeing signals is what a typical high-risk hold in our examples rests on, and each is shown in plain words.
Sources: payment industry write-ups on Visa’s VAMP programme from Chargeflow and Fraud.net, which can lag official notices; thresholds change, so we confirm your acquirer’s figures at the start.
What changes in your order flow once every order is scored and the doubtful ones are held
You review a short queue of explained holds, and the large majority of orders ship without delay.
How orders are checked today
- Orders are checked only when something looks odd.
- Rules are blunt and go unreviewed.
- Chargebacks are met with surprise.
- Good customers are cancelled by mistake.
- Nobody knows the hold rate.
How orders are checked afterwards
- Every order is scored before fulfilment.
- Rules and thresholds are tuned on your own history.
- Likely chargebacks are held and reviewed first.
- Borderline orders get a person’s judgement, not an automatic cancel.
- Hold rate, approval rate and losses are reported monthly.
What your review queue shows
- Fewer chargebacks.
- Fewer false declines.
- Faster fulfilment of good orders.
- A reviewable decision trail.
- A fraud rate you can quote.
What the screening does
- Gathering payment, address, history and behaviour signals for every order.
- Scoring the order and writing the reason in plain words.
- Placing a hold before fulfilment when the score and signals agree.
- Notifying the reviewer with the evidence.
- Logging every approval, hold and cancellation.
Decisions that stay with a person
- Deciding each held order.
- Contacting a customer to verify.
- Setting the hold threshold.
- Cancelling and refunding.
- Reporting card fraud to the processor.
Which online stores gain most from AI fraud screening
If you have had a chargeback you could not explain or you cancel orders on instinct, a measured score gives you something to stand on.
High-value or resale-prone products
Electronics, trainers, gift cards and luxury goods attract fraud and reward careful screening. We look at this situation in the audit and give online stores an honest view of whether the build is worth doing now or later in the fraud screening work.
Stores growing fast
Fraud grows with volume and visibility, and rules written for last year’s orders miss new patterns. If this describes your business, the fraud screening work for online stores is likely to pay back quickly, and we will tell you plainly if it will not.
Stores near card network limits
A rising dispute count needs action before the processor does it for you. Owners in this position often say the biggest change is the time returned each week, and fraud screening is built to deliver that first for online stores.
Stores with digital goods
Instant delivery leaves no time to recover goods, so the screen must run before download. This is the kind of situation where a fixed-price build is easier to justify than a monthly subscription for fraud screening for online stores.
Small teams without a fraud analyst
A score with plain reasons lets a non-specialist make confident calls. We look at this situation in the audit and give online stores an honest view of whether the build is worth doing now or later in the fraud screening work.
Stores shipping internationally
Cross-border orders have more legitimate oddities, and the scoring weighs them instead of banning them. If this describes your business, the fraud screening work for online stores is likely to pay back quickly, and we will tell you plainly if it will not.
The steps through which an order is screened before it ships
Every order passes the same short path, and only the doubtful ones stop for a person.
order_placedAn order is placed
The order, the payment result and the checks the processor returned arrive within seconds of checkout. Where the event depends on outside data, the data is checked for freshness first, because a stale figure is worse than no figure for fraud screening for online stores.
signals_gatheredSignals are gathered
Address match, delivery type, account age, order velocity and behaviour on the site are collected. Each moment like this has its own rule, so fraud screening for online stores behaves the same way every time and never has to guess what to do next.
score_computedA score is computed
The signals are weighed using rules tuned on your history, and a short reason is written. The event is logged with the rule that handled it, which lets you check later exactly what fraud screening did for online stores and why it did it.
decision_routedThe decision is routed
Low scores continue to fulfilment, middling scores wait for review and the highest are held with an alert. Because this trigger is handled separately from the others, a problem with it never delays the rest of the fraud screening work for online stores.
person_decidesA person decides
The reviewer sees the evidence, approves, contacts the customer or cancels, and the choice is logged. You decide how much of this runs without approval, and for online stores we usually begin with a review step until the first results look right in the fraud screening work.
outcome_learnedThe outcome is learned
Chargebacks and confirmed good orders feed back into the thresholds each month. Anything this trigger cannot decide safely is passed to you with the record attached, so fraud screening for online stores never acts on a doubtful reading.
One order followed from checkout to a reviewed decision
This shows how an order for a laptop is held while a normal order for socks ships at once.
- 01
Two orders arrive
A $1,380 laptop order and a $24 order for socks are placed within a minute of each other. At the end of this stage you can pause, adjust a rule or carry on, so fraud screening for online stores always moves at the pace that suits you.
- 02
Signals are gathered
The laptop order ships to a new address that differs from the billing address, uses a new email and follows three declined cards from the same device. Every step like this is recorded, so the fraud screening work for online stores can be traced from the first input to the final result without any gaps.
- 03
The score is written
The laptop order scores high with four agreeing signals, and the reason reads in plain words, while the socks order scores low. If the step cannot finish, the record says why and waits for you, so fraud screening for online stores never moves on with a half-completed task.
- 04
The routing happens
The socks order goes straight to fulfilment, and the laptop order is held before it reaches the warehouse. You can see the outcome of this step before the next one starts, which keeps the fraud screening build for online stores under your control throughout.
- 05
A person decides
The reviewer emails the cardholder’s address on file, gets no reply by the next morning and cancels and refunds the order. The step uses only the facts you supplied, so nothing in the fraud screening result for online stores depends on guesses or on outside sources you did not approve.
- 06
The loss is avoided
The decision is logged with the evidence, and the device is added to the watch list for future orders. Timing for this step is paced to the limits of your tools, so fraud screening for online stores keeps running calmly even when volume is high.
Pricing
Price your AI fraud screening build
Choose a plan, tell us your order volume and risk level, add any extras, and the total and delivery date update as you go.
What AI fraud screening costs and what raises the price
The build is a single payment, and the running cost is hosting and a little model usage.
Order volume
Higher volume needs more careful testing, and we check this factor against your real records during the audit, so the number in the quote comes from your data and not from a guess in the fraud screening build for online stores.
Product risk
High-risk categories need more signals and tuning. This is one of the factors we price in writing before work starts, so the fraud screening quote for online stores reflects your real situation and not an average.
Countries served
Each market adds address and delivery rules. If this factor is small for you, it adds little to the price, and the written quote shows exactly how much it did add for online stores in the fraud screening work.
Verification flows
Customer verification emails and replies add build. You can lower the cost by tidying this part first, and we say so plainly in the plan when the fraud screening work for online stores would benefit.
History to learn from
More past chargebacks give better tuning but take longer to prepare. Anything extra that this factor creates later is quoted before it is built, so the fixed price for fraud screening never grows by surprise for online stores.
Reviewers and roles
Several reviewers need routing and permissions, and we explain this factor in the quote in plain words, because online stores should understand what drives the price before agreeing to anything in the fraud screening work.
Hosting and model usage are billed to you directly by their providers, and the written scope estimates both.
Next step
Ready to price AI Fraud Screening for Online Stores?
AI fraud screening for online stores reads every order for the signals that separate stolen cards and takeovers from honest customers, holds only the doubtful ones for a person with the evidence beside them and lets the rest ship, so chargebacks fall without turning good customers away.
- Fixed price from $890 with no fee per order
- Every score explained in words
- A person decides every hold
Three more jobs the same screening can carry
Once every order has a score and a decision, further protection is simple.
Card-testing alerts
Bursts of small declined orders raise an alert and a temporary block on the pattern. The same records and rules that power fraud screening for online stores are reused here, which keeps the cost low and the results consistent with the rest of the build.
Repeat-customer allow list
Customers with a clean history are approved faster and held less often. You can switch this job on later without rebuilding anything, so online stores can start with the core fraud screening work and grow when the time is right.
Verification emails
Held orders can send a verification message in your voice, with the reply routed to the reviewer. Results from this job appear in the same weekly report as the main fraud screening work, giving online stores one place to judge whether it earns its keep.
Where each risk signal comes from and where it stops
Each signal has a named source and a stated limit, and the reason for a score says which signals counted.
| Store object | Fields we use | Direction | Source of truth |
|---|---|---|---|
| Payment checks | Address and security code results, 3D Secure result | Exact | Your payment processor |
| Address and delivery | Billing and shipping match, type of address | Exact | The order record and address checks |
| Account history | Age, past orders, past disputes | Exact | Your store |
| Order velocity | Orders and cards per device or email | Calculated | Your order and checkout logs |
| Behaviour on site | Time to checkout, repeated attempts | Where available | Your store events |
| Score and decision | Score, reason, reviewer, outcome | Written by us | Your delivered log |
We use data you already hold and your processor already returns, we do not buy personal data about your customers, and every signal carries its limits in the reason text.
Edge cases in order screening that we test before going live
Fraud rules fail at the edges, so we test the awkward cases on your own past orders first.
Gifts to new addresses
A gift is a normal reason for a different shipping address, so the weight of that signal depends on other signals. When the case appears, the workflow follows a written rule and records the choice it made, so you can review it afterwards if you wish in the fraud screening build for online stores.
Customers travelling or abroad
Foreign cards and unusual locations are weighed with the order history, not banned. If a rule cannot settle the case safely, it goes to you with the record attached, and fraud screening for online stores waits instead of guessing.
Freight forwarders
Known forwarding addresses add risk, and a person decides when the rest of the order is clean. Cases like this are the reason the fraud screening build for online stores includes a trial run, because a tidy example never shows them but real records do.
Subscription renewals
Renewal orders are screened differently from first orders, since the card has been used before. Your answer to this case is added to the rule table, so the same situation is handled automatically the next time it appears for online stores in the fraud screening work.
Large legitimate orders
Wholesale and event buyers are allowed through when verified, with a note for next time. The weekly report counts how often this case occurs, which tells online stores whether it deserves a better rule or is simply rare in the fraud screening work.
Digital and instant goods
Orders for digital goods are screened before the download link is released. Nothing about this case is hidden from you: it appears in the log with the date, the record and the rule that dealt with it in the fraud screening build for online stores.
Processor defaults, a fraud app and a built screening layer compared
Processor defaults are free and generic, a fraud app is broad and charges per order, and a built layer is tuned to your products, your customers and your decisions.
| Processor default | Fraud app | Checkout Loom build | |
|---|---|---|---|
| Scores every order | ✓ | ✓ | ✓ |
| Tuned to your own history | – | Partly | ✓ |
| Explains each score in words | Partly | Partly | ✓ |
| A person decides every hold | Partly | Partly | ✓ |
| No fee per order | ✓ | – | ✓ |
| Cost pattern | Included | Per order | A fixed fee from $890 |
We use your processor’s own risk checks as an input and add the layer that knows your business.
What is behind order screening and what it costs to run
The build uses the data your store and processor already return with a small scoring service, so the running cost is modest.
Store and processor data
Orders, payment checks and 3D Secure results read through the store and processor APIs. The monthly price of this layer is written next to our fixed fee before you commit, so the true cost of running fraud screening for online stores is clear from day one.
A scoring service
Rules and weights tuned on your history, versioned so every change is recorded. You can replace this layer later with another tool, because the rules live in a table you own and are not locked inside any one product in the fraud screening build for online stores.
A language model
Used to write the plain-word reason and to summarise customer replies, never to make the decision. We choose the lightest option that does the job for online stores, since a simpler layer means fewer things to maintain and fewer ways to fail in the fraud screening work.
A review queue
A simple screen or email digest where held orders wait, with the evidence beside each. If you already pay for something that covers this need, we use it, and we only add a new service when fraud screening for online stores genuinely requires one.
Fulfilment holds
A tag or status that stops fulfilment until a decision is made. Access to this layer is granted from your own account and can be removed at any time, so you stay in control of the fraud screening build for online stores.
A monthly report
Hold rate, approval rate, chargebacks avoided and false holds, in a short written summary. Its limits and prices are checked again at the start of each project, because tools change their plans, and online stores should not pay for last year’s terms in the fraud screening work.
How we keep screening accurate and fair
A wrong hold costs a customer, and a wrong approval costs a chargeback, so we measure both and report them.
Back-tested on your history
The scoring is run on past orders and compared with the chargebacks you actually had. If a check fails, nothing is sent or changed for that record, and it waits in a short list for you, because a delay costs less than a wrong action in the fraud screening build for online stores.
Quiet mode first
Scoring runs on live orders without holding anything until you are happy with the results. The safeguard is tested during the trial batch by feeding it deliberately awkward records, so online stores sees how it behaves before real customers are involved in the fraud screening work.
Two kinds of error reported
We report both false holds and missed fraud so you can see the trade-off. Everything the safeguard decides is written to the log with the reason, giving online stores a trail that is useful when someone asks what happened and why in the fraud screening work.
Alerts on spikes
A burst of high scores or declined cards raises an alert at once. During the month of corrections after launch, any gap in this safeguard is fixed in writing and the fix is added to the handover guide in the fraud screening build for online stores.
A dated decision trail
Every score, hold and decision is dated and linked to the rule version and the person. Reports on how often the safeguard fires show online stores whether the fraud screening work is getting steadier or whether a rule needs attention.
Questions answered in writing
If a hold looks wrong, the owner answers in writing and shows the signals. Because this protection is written as a rule and not as a habit, it keeps working when staff change or the business gets busier in the fraud screening build for online stores.
Customer data, fairness and access in a screening build
Screening touches personal and payment-related data, so we limit what is used and keep a record of why.
No card numbers
We never see or store full card numbers, only the check results your processor returns. We keep no copy of your keys, records or customer details in systems that belong to us, and the handover guide shows you how to confirm that in the fraud screening build for online stores.
Fair by design
We avoid signals that proxy for protected characteristics and show the reasons used for every hold. Only the fields that fraud screening needs are read, and only for as long as the task takes, which keeps the exposure for online stores as small as it can be.
Your credentials
API keys are yours, stored in a secret store and revoked from your side at any time. Every access is logged in your own accounts, so you can see who or what touched a record and when, without asking us for a report in the fraud screening build for online stores.
Minimal personal data
Only the fields needed for a decision are used, and they are not shared with other parties. If a key is ever lost, it can be replaced in minutes without rebuilding the fraud screening work for online stores, because the rules do not depend on it.
Not legal advice
We recommend you review privacy and consumer law obligations in your markets. The arrangement is reviewed at the end of the month of corrections, and any access that is no longer needed is removed in writing in the fraud screening build for online stores.
Retention you choose
Scores and signals are kept for a period you choose and then deleted. Customer details are never used to train outside models, and drafts for online stores are built only from the facts and records you supply in the fraud screening work.
What you receive with your AI fraud screening build
You finish with a working screen, a review queue and the documents to run them without us.
- A written review of your order history, chargebacks and current rules.
- A scoring service tuned on your own past orders.
- Plain-word reasons for every score.
- A review queue with the evidence beside each held order.
- Fulfilment holds that stop a flagged order before it ships.
- Alerts on card-testing bursts and other spikes.
- A monthly report of holds, approvals and losses avoided.
- A short runbook for the scoring and the review queue, opening with how to switch holds off.
- Written answers from the owner to any question about a score or a hold.
Verified client reviews and the controls behind our screening builds
Below are clients’ own words, unedited, and then the three controls that every screening we build must carry.
- 100%of holds are decided by a person.
- 2error kinds, false holds and missed fraud, are reported every month.
- 1plain-word reason accompanies every score.
Controls follow payment industry practice and are not a guarantee against fraud or a legal compliance opinion.
★★★★★
I highly recommend ordering from the Checkout Loom team. I had slow responses and poor communication with other providers, but this team was great and I am very happy with their service.
HarmetUnited States
Verified
★★★★★
I am a returning customer because my first AI automation workflow worked so well. Once again everything was built exactly as requested, and the whole process was reliable from start to finish.
WilliamUnited States
Verified
★★★★★
The determination and commitment to quality from the Checkout Loom team were clear from day one. They understood my requirements perfectly and completed everything with professionalism and care, and it is rare to find a team this dependable.
LorcanColombia
Verified
★★★★★
It was an excellent experience from start to finish, and the automation works reliably with everything configured properly. It is clear that the team has strong knowledge of automation, and I will definitely be returning for future work.
EllieUnited States
Verified
★★★★★
The team carefully investigated the problem, tested different scenarios and delivered a dependable solution. A duplicate contact issue had been causing problems for quite some time, and it is finally resolved.
RichardUnited States
Verified
★★★★★
The Checkout Loom team were easy to work with and brilliant at the work itself. I highly recommend them to anyone who needs dependable automation.
IveraAutomation client
Verified
★★★★★
Everything was configured exactly as discussed, and the team made sure the automation was stable before delivering it. It has been a huge time-saver for our whole team, and it runs without any manual work.
BellamyUnited States
Verified
★★★★★
This was excellent automation work, and it is clear that the team has strong experience with AI and automation. I highly recommend the Checkout Loom team to other businesses.
FinnUnited States
Verified
How the screening is built from order history to live holds
We prove the scoring on your past orders and run it quietly before it holds anything.
- 01
Review your history
We read your past orders, your chargebacks and your current rules and write down what has gone wrong. The time this step takes depends on how quickly your records and answers arrive, and we tell online stores early what we need to keep moving in the fraud screening work.
- 02
Build the scoring
We build the signals and weights and test them on past orders. We use plain language in every document from this step onward, so anyone at online stores can read the plan and understand it in the fraud screening work.
- 03
Run it in quiet mode
The scoring runs on live orders without holding, and we compare it with your own judgement. A short summary of the step is added to the handover guide, so the way fraud screening was built for online stores is never lost.
- 04
Turn on holds
We switch on holds with a threshold you approve and the review queue. You see the result of this step in writing before the next one begins, which keeps the fraud screening build for online stores on schedule and under your control.
- 05
Add alerts and reporting
We add the card-testing alerts and the monthly report. If this step shows a problem, we say so at once and propose the fix, and the date for fraud screening is moved in writing rather than quietly for online stores.
- 06
Hand over the screening
The runbooks and connection details come to you in writing, with written answers to later questions. Nothing in this step touches your live customers, so online stores can review it calmly and ask for changes before anything goes public in the fraud screening work.
Estimate what fraud and false declines are costing your store
Enter your monthly orders, your average order value and your fraud and false decline rates, and the estimate shows what better screening could recover.
Chargeback losses each month–
Sales lost to false declines–
Possible monthly recovery–
The recovery assumes a third of fraud losses and a third of false declines are saved, which is an assumption and not a promise, and chargeback fees are not counted.
AI fraud screening for online stores questions, answered in full
These are the questions store owners ask about scoring, false declines, chargebacks and cost before they order a screening build.
What is AI fraud screening for online stores?
It is a layer that scores every order on payment, address, history and behaviour signals, explains the score in plain words, holds the doubtful orders for a person to decide and lets the others ship.
How is fraud screening different from my payment processor’s checks?
Your processor’s checks are generic and are one of our inputs, while fraud screening tuned on your own orders knows which products, regions and customers are normal for you and can be adjusted by you.
Will it block good customers?
It is designed to hold rather than cancel, so a borderline order waits for a person who can verify it, and we report false holds each month so the threshold can be tuned.
Does it stop chargebacks completely?
No tool can, because some disputes are friendly fraud or customer error, but screening before dispatch reduces stolen-card chargebacks, and we report the numbers so you can see the effect.
Does Shopify already screen orders?
Shopify provides fraud analysis on eligible orders, and we use that result as a signal, adding your own rules, history and a review queue on top, though we do not replace Shopify Protect or your processor’s guarantees.
What about card network thresholds?
Reports say Visa lowered its excessive merchant ratio to 1.5 percent in April 2026, counting disputes even when you win them, so keeping the ratio down matters, and we recommend confirming your exact position with your acquirer.
Is it fair to customers?
We avoid signals that proxy for protected characteristics, show the reason used for every hold and keep a person in the decision, and we recommend you review privacy and consumer law in your markets.
How soon will scoring start, and what must I provide first?
You will have scoring in quiet mode between five and twenty working days after you order, depending on the plan and the speed you choose. To begin we need read access to your store and processor and an export of your recent chargebacks.
Do I pay per order or per month after the screening is built?
Only the hosting and any model usage you choose, billed to you directly by their providers.
Ship the honest orders and catch the doubtful ones before they cost you
Choose a plan in the pricing calculator and continue to checkout, or request a quote describing your orders and your recent chargebacks. Our written answer follows within a working day and names the price and the date your screening goes live.
- Fixed price from $890 with no fee per order
- Every score explained in words
- A person decides every hold