Ecommerce fraud is not one problem. Card testing, stolen cards, account takeover, fake cash-on-delivery orders, refund abuse and courier collusion each need a different control. Buying "fraud prevention" as a single tool usually disappoints: it covers one part of the journey while losses move elsewhere.
This comparison sets out five approaches for an online seller in Bangladesh and scores them on the same criteria, so you can decide what to add first rather than buying the most impressive demo.
The criteria
- Coverage: which fraud types and order stages the approach addresses.
- False positives: how often genuine customers are blocked, and how easily a decision can be reviewed.
- Operating effort: the daily work to review queues, tune rules and handle appeals.
- Evidence quality: what the record shows when you dispute a chargeback or investigate an account.
- Data requirements: what customer, device and payment data you collect, and any privacy obligation it creates.
- Cost model: subscription, per-transaction fee, percentage of value, project or internal time.
- Fit: the volume, team size and sales mix where the approach makes sense.
Option 1: Manual review and simple rules
How it works: staff review high-value orders, first-time buyers, mismatched addresses or unusual quantities before dispatch. Rules may live in a spreadsheet or in the order platform's tags.
Strengths: immediate, cheap to start and easy to explain. An experienced operations person often spots a fraudulent order from details a model cannot see, such as an implausible delivery combination.
Limits: it does not scale, is inconsistent between reviewers and competes with order processing at peak. It also leaves little evidence for a chargeback dispute beyond a screenshot.
Best fit: a new store with low volume, or a temporary control while a better process is designed. Cost model: internal hours and supervision, plus the cost of mistakes.
Option 2: Payment provider risk tools
How it works: the payment gateway or processor evaluates a transaction using its own signals and returns a risk decision. Many providers also support strong customer authentication and address verification.
Strengths: the data and the model are included with the payment relationship, implementation is usually light, and the decision arrives at the moment of payment. Good for card-not-present fraud and card testing.
Limits: coverage is concentrated on the payment event. It may not see cash-on-delivery abuse, account takeover before checkout, refund abuse or courier disputes. Rules and risk scores differ by provider, and a decline you cannot explain is a customer you cannot recover.
Best fit: card and wallet payments where the provider's coverage matches your market. Cost model: often included, sometimes priced per transaction or as a platform fee. Confirm what happens when you change providers.
Option 3: Platform-native controls
How it works: the ecommerce platform provides fraud analysis, order hold rules, address checks or app-based extensions. Shopify and WooCommerce both offer native or plugin-based options.
Strengths: lives with the orders, requires little integration and gives staff a familiar interface. Rules can be specific to products, shipping regions or customer history.
Limits: depth varies, and plugins carry maintenance risk. Data may not travel if you replatform, and coverage usually stops at the storefront rather than offline or marketplace orders.
Best fit: a single-platform seller who wants controls inside the order workflow. Cost model: subscription or app fee, plus configuration time and periodic review.
Option 4: Dedicated fraud platform
How it works: a specialist service ingests order, device, identity and payment signals, scores each order and provides a review queue, case management and reporting. Some also cover account and refund abuse.
Strengths: the deepest coverage and the best evidence trail across channels. Useful when you sell across storefronts, use several payment providers or handle high order values.
Limits: implementation requires data mapping and a clear decision policy. The monthly cost is hard to justify at low volume, and the vendor becomes part of your order path. Define a fallback for outages.
Best fit: growing multi-channel sellers with a person who can own the review queue. Cost model: subscription or per-transaction pricing plus onboarding, and internal review time.
Option 5: Custom scoring and in-house rules
How it works: you combine your own data (order history, delivery success, customer tenure, courier performance, return behaviour) into a score or a set of rules that runs inside your order system.
Strengths: fits your exact business, can include cash-on-delivery and courier signals that external tools miss, and keeps the data in your environment. Useful when fraud patterns are specific to your products or regions.
Limits: requires engineering, a labelled history of good and bad orders, and ongoing maintenance. A poorly validated model can block loyal customers or miss a new pattern. Explainability matters when a customer appeals.
Best fit: a business with enough volume and technical capacity to own a model, or a distributor with unusual order patterns. Cost model: build project, hosting, monitoring and the analyst time to review outcomes.
How the options compare
- Fastest to start: manual review, then platform-native controls.
- Best card-fraud coverage per unit of effort: payment provider risk tools.
- Best cross-channel evidence: dedicated fraud platform.
- Best fit for COD and local delivery patterns: custom scoring combined with manual review.
- Lowest operating burden: provider tools, but with the least visibility beyond payment.
- Highest control and highest maintenance: custom scoring.
Decision guidance
Start with the loss that actually hurts. If chargebacks from stolen cards dominate, fix payment risk first. If fake cash-on-delivery orders and refused parcels dominate, a card-risk tool will not help much; you need address verification, customer history and courier confirmation. If the same customer repeatedly claims non-delivery, you need case evidence rather than a better score.
Then consider volume and team. Below a few hundred orders a month, a complex platform is usually premature. Above that, manual review becomes a bottleneck and needs clear ownership. Multi-channel sellers should prioritise a shared order identity before buying another tool, because a fraud decision is only as good as the record it evaluates.
A 30-day test plan
- Collect one month of orders with outcomes: delivered, returned, disputed, charged back.
- Define three rules or one score and run them in shadow mode, without blocking orders.
- Compare the decision with what happened and count false positives.
- Review every false positive with the operations team and adjust the rule.
- Move only the highest-confidence rule into production, with a manual override.
- Review weekly for a month before adding the next control.
Metrics
- Fraud loss as a share of completed order value, by fraud type.
- False-positive rate and the value of genuine orders blocked.
- Review queue age and the share of orders reviewed before dispatch.
- Chargeback win rate and the quality of evidence submitted.
- Repeat abuse rate for accounts and delivery addresses.
If you need fraud controls wired into your order and payment flow, start with Commerce Systems. The pricing page explains how integration and support are scoped. For the monitoring side, read how payment fraud monitoring matures.
Sources
- Stripe, Radar fraud prevention documentation, accessed 17 November 2026.
- Shopify Help Center, Fraud prevention, accessed 17 November 2026.
- OWASP, Automated Threats to Web Applications, accessed 17 November 2026.
- PCI Security Standards Council, accessed 17 November 2026.
Responses (0)
No comments yet. Add the first one below.
Leave a response
Your email address will not be published. Required fields are marked *.