The head of e-commerce brings a table to the meeting: three payment providers, three fee rates. The cheapest is 0.3 percentage points below the current one, so on €1.5 million of monthly volume the saving is roughly €4,500 a month. The decision takes fifteen minutes. Six months later, finance is reconciling payouts by hand because the new provider’s settlement report lists daily net totals without order references. Engineering is patching duplicate orders that appear when a payment notification arrives twice. Customer support cannot issue a partial refund on a local bank payment method from the dashboard. The saving in the table still exists on paper, but nobody knows any more what it cost to get it.

The scenario is illustrative, but each element in it is common. This article is for CTOs and heads of e-commerce who are choosing a payment provider for an online store, a SaaS product or a B2B platform, and who will have to build or rebuild the integration. The thesis is this: the fee is the least important criterion when choosing a payment gateway, as long as the offers sit in a similar range. Total payment cost is decided by three other things: the integration model, which sets your responsibility for card data security; the quality of settlement and refunds on the provider’s side; and whether your application treats asynchronous notifications as the source of truth. The second half of the thesis is easier to dispute: do not build for several providers at launch. One provider, with an integration designed so it can be replaced, is the better starting point.

Why does comparing payment gateways on fees lead to the wrong choice?

Because the fee is the only number that is easy to compare, not the only one that matters. A table of percentages does not show fixed per-transaction charges, the cost of refunds and chargebacks, currency conversion margins, payout timing, or how many hours a month finance will spend matching payouts to orders. Nor does it show the costs that arrive once but are large: integration, testing, PCI DSS compliance and, eventually, migration to another provider.

Put the two figures from the introduction side by side. A difference of 0.3 points on €1.5 million a month is about €54,000 a year. That is real money. But if the settlement report cannot be reconciled automatically and finance spends half a full-time role on it, while engineering spends a quarter fixing notification handling, the saving shrinks quickly. More important still is the risk that never shows up as a monthly cost: goods shipped without payment, a refund issued twice, transactions declined by issuers because the integration mishandles strong customer authentication.

Fees do matter when offers differ substantially, or when volume is high enough that tenths of a percentage point pay for a team. Then it is worth negotiating and comparing pricing models, as discussed below. But the order of decisions should be the reverse of the usual one: first establish which providers meet your functional, legal and integration requirements, and only then compare their prices on your own transaction mix.

There is also a strategic argument. A payment provider is not a vendor you replace every year. Its transaction identifiers end up in your database, its card tokens in your customer profiles, its reports in your accounting system and its dashboard in your support team’s daily routine. The deeper a provider grows into your application, the more expensive it becomes to change. Choosing on fees alone is often a multi-year decision made on a criterion that may change at the first price renegotiation.

Who are payment providers and how do their business models differ?

In payments conversations the same words mean different things, so it helps to define them. A payment gateway, loosely, is the service that captures payment details from the customer and passes them on for authorisation. An acquirer is the institution that holds the relationship with the card schemes and settles card transactions on the merchant’s behalf. A payment service provider (PSP) or aggregator typically combines both roles and adds local payment methods. Internationally the field includes Stripe, Adyen, PayPal/Braintree, Checkout.com, Worldpay and many regional players; in individual European markets local providers often have better coverage of domestic methods.

The model has practical consequences. With an aggregator, the merchant signs one contract and gets many payment methods through one API, but funds land first in the provider’s account and are paid out on a schedule. With direct acquiring, the merchant contracts with an acquirer, often gets better pricing at volume and more control over transaction data, but has to assemble local methods separately and usually faces longer onboarding. Marketplace platforms are a third category: they handle splitting payments between many sellers and verifying those sellers.

The first question about any provider is its legal status. A provider that holds customer funds and passes them to the merchant is providing payment services and needs authorisation. In the EU that means a licence as a payment or e-money institution (or a bank), with passporting to other member states; you can check it in the European Banking Authority’s central register of payment institutions and in national registers. In the UK the FCA register plays the same role, and in the US money transmission is licensed largely at state level. This is not paperwork: the provider’s status determines how customer funds are protected if it runs into trouble and which supervisor you deal with in a dispute.

The second question is whether your own company is becoming a payment service provider. A platform that collects money from buyers and passes it to sellers can easily cross the line between merchant and intermediary. If you are building a marketplace or an app in which money moves between users, settle the funds-flow model with legal counsel before you choose a provider, because it determines which product you need from them. For financial institutions, there are also supervisory requirements on vendor selection itself, which we covered in our article on how DORA and PSD2 change vendor selection for financial software.

Which payment methods does your application actually need?

Start from data about your customers, not from the provider’s feature list. Payment preferences are strongly national, and the method that dominates one market may be marginal next door. Cards and wallets such as Apple Pay and Google Pay work almost everywhere, and in the US they remain the default for online purchases, alongside ACH bank transfers for B2B and recurring payments. In Europe the picture is fragmented: iDEAL in the Netherlands, Bancontact in Belgium, SEPA Direct Debit for recurring payments across the euro area, and in Poland a mobile payment system that most Western teams have never heard of. In B2B sales, payment often lives inside a B2B customer portal, where invoices and credit limits sit next to cards.

Poland is a useful illustration of how wrong a card-first assumption can be. According to the report “10 lat BLIKNĘŁO” (“Ten years of BLIK”), prepared by EY for Polski Standard Płatności and published in March 2025, BLIK had 18.5 million active users at the end of 2024, or 58 per 100 people aged 15 and over, and processed more than 2.4 billion transactions that year. Online BLIK payments reached PLN 173 billion in 2024, and as early as 2023 they accounted for about half of the value of Polish e-commerce. The same report cites a 2024 Gemius survey in which 51% of online shoppers named BLIK as their most frequently used payment method, against 22% for pay-by-link bank transfers. A store entering Poland with cards only is leaving the main checkout path unbuilt.

The lesson for integration is broader than one country: the dominant local method should be the path that works best, including refunds and, for subscription products, recurring payments. Check whether a provider supports the local methods you need natively or through a partner, how it handles a customer who does not confirm in their banking app in time, and whether recurring variants exist. The BLIK report mentions recurring BLIK payments, expanded at the end of 2024, where users authorise a regular payment without entering a code each time. For subscription apps, that kind of feature changes the economics in markets where cards are not the habit.

Beyond the dominant methods, consider buy now, pay later and instalments for higher basket values, and in B2B the invoice-and-bank-transfer flow that never touches the gateway but must still land in the same order state model. List your methods as essential, important and nice to have, with data behind each. A long list of logos does not automatically raise conversion; it does multiply the paths to test and the refund types to support.

What does a payment actually cost, and how do you compare provider offers?

Payment cost is a sum of several charges, of which the percentage fee is merely the most visible. For cards there is a structure worth understanding before negotiating. Every card transaction carries an interchange fee paid to the card issuer, a scheme fee paid to the card network, and the markup of the acquirer or provider. In the EU, Regulation (EU) 2015/751 caps interchange at 0.2% of transaction value for consumer debit cards and 0.3% for consumer credit cards. The caps do not cover commercial cards or cards issued outside the European Economic Area, so the share of those cards in your volume has a large effect on your real cost. In the US, Regulation II (the Durbin Amendment) caps debit interchange for large issuers at 21 cents plus 0.05% of the transaction, plus a one-cent fraud-prevention adjustment for eligible issuers; the Federal Reserve proposed lowering it in 2023. Credit card interchange in the US is not capped and is considerably higher than in the EU.

That produces two pricing models. With blended pricing, the provider quotes one rate for all cards and absorbs the differences. With interchange++, you pay the actual interchange, the actual scheme fee and a separately agreed provider markup. Blended is easier to budget; interchange++ is more transparent and, with a large share of low-cost consumer debit cards, usually cheaper. In the EU, Article 9 of Regulation 2015/751 requires acquirers to specify charges separately for different card categories and brands unless the merchant asks in writing for blended pricing, and Article 12 requires them to report, after each transaction, the merchant service charge and the interchange fee separately. You have a right to know what you are paying for.

Local methods are simpler in structure, but their rates often differ from card rates and vary between providers. We deliberately do not quote specific providers’ prices here: price lists change, and at meaningful volume rates are negotiated individually, so a public price page says little about what you will actually pay.

Cost componentWhat to check in the offerWhen it matters most
Percentage feeSeparately for local methods, consumer cards, commercial cards and non-EEA cardsAlways; most with high average order value
Fixed per-transaction feeWhether it applies to all methods, including refundsLow order values and micropayments
Card pricing modelBlended or interchange++; reporting per Articles 9 and 12 of Regulation 2015/751 in the EUHigh card share and volume
RefundsRefund fee, whether the original fee is returned, partial refundsHigh return rates (fashion, electronics)
ChargebacksDispute fee, evidence tooling, thresholds of scheme monitoring programmesDigital goods and high-fraud categories
Currency conversionFX margin, settlement currencies, multi-currency accountsCross-border sales
PayoutsTiming in business days, payout fees, minimum amounts, rolling reserveTight working capital
Fixed chargesMonthly fee, minimum monthly volume charge, setup feeSmall or seasonal volume
Cost on your sideReconciliation, dispute handling, integration maintenanceAlways; most often ignored

How do you compare offers fairly? Take your actual transaction mix from the past 6–12 months: count and value per payment method, share of foreign and commercial cards, refund and dispute volume. Ask each provider to price that mix, not the example on its pricing page. For a new application with no history, build three volume scenarios and see how the cost changes in each. Only then do the differences between offers become comparable, and only then can you see whether a lower percentage is eaten by a fixed fee or an FX margin.

How do settlement and reporting affect your finance team?

This is the criterion CTOs usually hear about from the CFO, and usually too late. The provider accepts payments all day but pays them into your account in batches, net of fees, refunds and chargebacks. Finance then has to answer a question: which orders make up the transfer that just arrived? If the settlement report does not say, someone answers it by hand.

So before choosing a provider, ask for sample settlement files and check a few things. Does each line carry your order reference, not just the provider’s transaction ID? Does the report show gross, fee and net separately per transaction? Are refunds and chargebacks separate lines linked to the original payment? Is there a payout identifier that ties the report to the credit on your bank statement? Is the report available through an API, or only as a download from the dashboard? These answers determine whether reconciliation can be automated.

The second topic is payout timing. The difference between next-business-day payouts and weekly payouts is working capital, which has a price at volume. Some providers also hold a rolling reserve, a portion of funds set aside against future refunds and disputes, particularly for new merchants or higher-risk categories. Ask directly and put the terms in the contract.

The third topic is integration with accounting or ERP. Your application should store payment data in a form finance can reconcile without engineering’s help: the provider’s transaction ID, payment method, amounts, currency, settlement date and payout ID. If you are building a SaaS product with subscriptions, add the logic of billing, invoicing and failed renewals, which we described in our article on the architecture decisions that set the cost of a SaaS product. It also shows why billing and the payment gateway are two separate layers that are easy to confuse.

How do you handle refunds and chargebacks without leaking money?

Refunds are part of selling, not an exception. In the EU, the Consumer Rights Directive (2011/83/EU) requires traders to reimburse consumers within 14 days of being informed of a withdrawal from a distance contract, using the same means of payment the consumer used unless the consumer agrees otherwise. The US has no general federal right of withdrawal for online purchases; refund obligations come from your own policy, state law and card network rules, while consumer card disputes are backed by federal rules such as the Fair Credit Billing Act. Either way, for the integration it means one thing: refunds must be executable through the provider’s API for every method you support, including partial refunds when the customer returns one item out of three.

When evaluating providers, check how refunds work for each method. Do local bank methods refund the same way as cards? Can you issue several partial refunds against one payment? How long after a transaction can you still refund through the API? What happens if your balance with the provider is lower than the refund amount? Does a refund generate a notification your application can process the same way as a payment? Every “no” in these answers means manual work for support or a bank transfer made outside the system.

A chargeback is a different mechanism: the customer disputes the transaction with their card issuer and the funds are pulled back from the merchant before the dispute is resolved. The merchant can submit evidence, but has to do so within the deadline and in the format the provider requires. Check whether the provider reports disputes through the API or notifications, whether evidence can be submitted electronically, and what each dispute costs regardless of outcome. For digital goods, where there is no proof of delivery, record evidence of service from the start: access logs, IP addresses, activation times.

The most common refund integration mistake is having two sources of truth. Support issues a refund in the provider’s dashboard, the application never hears about it, the order still looks paid and sales reports are overstated. The rule is simple: refunds are initiated by the application, and if one is issued in the dashboard for any reason, the application must learn about it from the provider’s notification and record it in its state model.

What do PSD2 and strong customer authentication change in the checkout?

In the European Economic Area, PSD2 introduced strong customer authentication, with the details set out in Commission Delegated Regulation (EU) 2018/389, applicable from 14 September 2019. For online card payments this means, in practice, 3D Secure 2, where the issuer can ask the customer to confirm the payment in a banking app. The UK kept an equivalent regime after Brexit. The US has no SCA mandate; there, 3D Secure is a fraud-management tool the merchant can choose to use, typically to shift liability for fraudulent transactions to the issuer.

In Europe, the exemptions matter, because they determine how many transactions pass without an extra step. The regulation allows payment providers not to apply SCA for, among others, remote payments up to EUR 30, as long as the cumulative amount since the last authentication does not exceed EUR 100 or the number of such payments does not exceed five (Article 16); subsequent payments in a series of recurring transactions of the same amount to the same payee (Article 14); payees on a trusted beneficiaries list (Article 13); and transactions assessed as low-risk by real-time transaction risk analysis (Article 18). That last exemption depends on the provider’s fraud rate: for card payments up to EUR 500 the reference fraud rate is 0.01%, up to EUR 250 it is 0.06%, and up to EUR 100 it is 0.13%.

What does this mean for provider selection? First, ask whether and how the provider requests exemptions on your behalf, and how it reports the share of transactions authenticated, exempted and declined. Second, recurring payments and merchant-initiated transactions, such as renewing a subscription on a stored card, require the first and subsequent transactions to be flagged correctly. Errors in that flagging show up as renewal declines that are hard to diagnose. Third, even when you request an exemption, the issuer can demand authentication, so the checkout must always be able to handle that step, including in a mobile app.

The rules are also moving. The EU is working on PSD3 and a Payment Services Regulation that are set to change parts of the regime, including authentication and fraud protection. Before signing a multi-year contract, ask the provider how it plans to implement new requirements and who bears the cost of changes on your side of the integration.

Redirect, iframe or your own form: how does the integration model set your PCI DSS scope?

This is the single most important technical decision for card payments and the one with the largest long-term cost impact. PCI DSS applies to anyone who accepts card payments, but the scope of obligations depends on whether card data touches your systems at all. The current version is PCI DSS 4.0.1, published in June 2024. Version 4.0 was retired on 31 December 2024, and the requirements marked as future-dated became mandatory on 31 March 2025.

In practice there are three models. A redirect to the provider’s payment page, or a provider-hosted iframe, means card data goes straight to the provider. A merchant that has fully outsourced card data processing to PCI DSS validated third parties usually qualifies for the shortest self-assessment questionnaire, SAQ A. A form embedded on your page but generated by your code, or posting data directly from the browser to the provider, usually means SAQ A-EP, because your page affects the security of the transaction even though you do not store card data. Card data passing through your servers means the full set of requirements, and it is a model most companies should avoid.

Version 4.0 added two requirements aimed specifically at payment pages: 6.4.3, on inventorying, authorising and checking the integrity of scripts on the payment page, and 11.6.1, on detecting unauthorised changes to that page. They respond to attacks in which a malicious script injected into a store skims card data. In January 2025 the PCI Security Standards Council updated SAQ A: it removed requirements 6.4.3, 11.6.1 and 12.3.1, but added an eligibility criterion under which the merchant must confirm that its site is not susceptible to attacks from scripts that could affect its e-commerce system. The new SAQ A took effect on 31 March 2025. Merchants that cannot meet the criterion move to SAQ A-EP, which according to an analysis by the auditing firm usd AG adds more than a hundred requirements.

The conclusion is less obvious than it looks: a redirect or provider iframe does not end the security conversation, it moves it from formal compliance to page hygiene. You still have to control which third-party scripts your site loads, how you protect the store’s admin accounts and whether anyone can tamper with the redirect URL. A Content Security Policy, subresource integrity and regular reviews of front-end dependencies all help. But the cost difference between SAQ A and SAQ A-EP, let alone full scope, is large enough that the integration model should follow from this analysis, not from a designer’s wish for the card form to match the rest of the page. Most providers now offer iframe-based card fields that can be styled, so you rarely have to choose between looks and liability.

Why is the webhook, not the customer’s return, the source of truth about a payment?

Because the customer’s return from the provider’s page is a browser event, not a payment confirmation. Customers close the tab after confirming in their banking app, lose signal on a train, come back in a different browser or land on the return URL before the bank has confirmed a transfer. An integration that marks orders as paid based on parameters in the return URL will, sooner or later, ship goods without payment or show the customer an error after taking their money.

The source of truth is a server-to-server notification from the provider, the webhook, or a query to the provider’s API for transaction status. That mechanism has properties you need to know. Stripe’s documentation states them plainly, and they illustrate what applies to most providers. Notifications are retried if your server does not respond with a 2xx code; in live mode Stripe attempts delivery for up to three days with exponential backoff. The same event can arrive more than once. Delivery order is not guaranteed, so a refund event can arrive before the payment event. Your server should respond with 2xx quickly and do the real processing asynchronously, because slow handling leads to timeouts and retries.

Four integration rules follow. Verify the signature of every notification with the provider’s key, against the raw request body, and reject notifications with stale timestamps to prevent replay attacks. Without verification, anyone who knows your endpoint URL can “pay” for an order. Record the IDs of processed events and ignore those you have already seen. Do not assume order: on each event, check whether the state transition makes sense and, when in doubt, fetch the current transaction state from the provider’s API. Queue the processing: the endpoint stores the event and responds, and business logic runs in a separate worker that can retry.

You also need a mechanism that catches what slips through. A periodic job comparing orders awaiting payment with their status at the provider picks up notifications that never arrived and transactions that got stuck. A daily reconciliation against the settlement report closes the loop on the finance side. These elements rarely appear in the first integration estimate and are then bolted on in a hurry when the first discrepancies surface. For a broader view of choosing communication patterns between systems, see our article on software integration patterns and approaches and the glossary entry on systems integration; for protecting the endpoints themselves, see our piece on API security.

How do you design idempotency and a payment state model?

Idempotency means that repeating an operation has the same effect as performing it once. In payments it has two sides. The first concerns requests your application sends to the provider. If a request to create a payment or a refund fails with a network error, you do not know whether the provider executed it. Without idempotency, a retry can create a second payment or a second refund. That is why providers support idempotency keys: the application generates a unique key per operation and sends it with the request, and on a retry with the same key the provider returns the result of the first request. Stripe documents keys of up to 255 characters, recommends random identifiers such as V4 UUIDs, compares the parameters of a retried request with the original and returns an error if they differ, and may prune keys once they are at least 24 hours old. Check how your provider implements this, and for every state-changing operation (payment, capture, refund) use a key tied to your own operation ID.

The second side is notification handling, covered above. Here your application provides idempotency: it stores the event ID in the same database transaction as the order state change, so a repeated event changes nothing the second time.

The foundation of both is your own payment state model, independent of the provider. A payment in your application moves through clearly defined states: created, awaiting customer, authorised, captured, failed, cancelled, partially refunded, fully refunded, disputed. Every provider event is translated into a transition in that model, and illegal transitions are rejected and logged. The model also holds the mapping between your orders and the provider’s identifiers, which makes reconciliation and any future migration easier.

Why does this matter so much? Without your own state model, payment logic spreads through the application as conditions like “if the provider status is X”. Each provider has different statuses with different semantics, so switching providers means reviewing every piece of code that looks at them. With your own model, the change is confined to one adapter that translates provider statuses into your states. Automate tests for this layer against the provider’s sandbox, including edge cases: duplicate notifications, out-of-order notifications, request timeouts and issuer declines. Our article on API testing strategy shows how to build that kind of test suite.

How do you prepare for switching providers, and should you run two from the start?

Companies switch providers for several reasons: volume has grown and direct acquiring offers better pricing, the business is entering a new market, the provider changed its terms, service quality dropped. In every case the cost of switching depends on how the integration was built. That is why the question of switching belongs before you choose the first provider.

Three things determine that cost. The first is the state model and adapter that isolate the rest of the application from the provider’s API. The second is stored card data. If you sell subscriptions on cards, the card tokens belong to the provider, and without migrating them every customer would have to re-enter their card. Before signing, ask whether the provider will export card data to another PCI DSS compliant party when the relationship ends, on what terms and how quickly, and write it into the contract. The third is transaction history: IDs, reports and disputes that must remain accessible after the switch, if only to refund transactions made before the migration.

So should you integrate two providers from day one, or adopt a payment orchestration platform that routes traffic between them? For large merchants, businesses selling in many countries, and companies for which an hour of payment downtime means major losses, the answer may be yes. For most stores and applications at launch, it is no. A second provider doubles testing, method configuration, refund handling, reconciliation and dispute work, while its benefits, better pricing or outage resilience, only materialise at volumes a new application does not yet have. It is cheaper to build an integration that is ready for a second provider than to run two from the first day.

Your commerce platform also matters. If your store runs on a SaaS platform, the list of providers may be limited to those with a ready-made plugin, and part of the payment logic sits on the platform’s side. If you are building custom or headless, you have full freedom and full responsibility for everything described here. We covered those trade-offs in our comparison of SaaS, headless and custom e-commerce development.

How do you run a payment provider selection step by step?

A selection usually takes a few weeks and is best planned so that engineering and finance work in parallel. The sequence below works because it eliminates providers on must-have criteria before the price conversation starts.

Step 1: requirements. List payment methods (essential, important, optional), markets and currencies, the business model (one-off sales, subscriptions, marketplace), expected volume in three scenarios and regulatory requirements. Settle the funds-flow model with legal counsel at this stage.

Step 2: shortlist. Choose three or four providers that meet the essential requirements, hold the right licences and support your platform or stack. Check their status in the EBA register, the relevant national register or, outside the EU, the local equivalent.

Step 3: technical evaluation. Engineering reviews the API documentation, sandbox, notification mechanism (signatures, retries, ordering), idempotency keys, refunds via API, client libraries and status page. It pays to build a prototype of the key flow, for example a local method payment plus a partial refund, in each shortlisted provider’s sandbox. Two or three days of work per provider tells you more than a sales presentation.

Step 4: financial and operational evaluation. Finance reviews sample settlement reports, payout timing, reserve terms and quotes based on your transaction mix. Support tests the dashboard: transaction search, refunds, dispute handling.

Step 5: contract. Negotiate price, but also exit terms: notice period, export of card data and transaction history, an availability commitment with consequences, and how changes to pricing and the API are communicated.

CriterionWeight (example)How to verify
Payment methods and markets20%Essential method list; sandbox test of local methods and cards
Integration model and PCI DSS scope15%Redirect and iframe card fields available; resulting SAQ for your chosen model
API and notification quality20%Prototype: signatures, retries, idempotency, partial refunds via API
Settlement and reporting15%Sample files: order reference, gross/net, payout ID, reporting API
Total cost on your mix15%Quote on 6–12 months of data or three volume scenarios
Exit terms and contract10%Card token export, notice period, transaction history access
Legal status and support5%Licence register, technical support channels, status page

The weights are examples and should reflect your situation. A store with high returns will weight refunds and reporting more heavily, a subscription app recurring payments and exit terms, a marketplace split payments. Note that total cost carries 15% here, not 60%. That follows from the thesis of this article: offers that survive the earlier steps usually differ less in price than in integration quality.

How much does a payment integration cost and how long does it take?

The honest answer depends on scope, so instead of one number it is more useful to show what the work consists of. Integrating one provider via redirect with one payment method, no API refunds and no reconciliation is relatively simple and often fits into a few weeks of work for one or two people. The production-grade integration this article describes involves more: a payment state model, notification handling with signature verification and a queue, idempotency, full and partial refunds, dispute handling, a reconciliation job, export to the finance system, recurring payments if needed, edge-case testing and monitoring.

The estimate is best built bottom-up from that list, after a short analysis phase in which the team learns the chosen provider’s API and your order system. As an illustration rather than a quote: if the integration needs two experienced developers and a tester for two to three months, and you work with a dedicated team, ARDURA’s software development services page puts a dedicated team at from about €14,500 a month, which would put the team cost for such an integration somewhere around €29,000–43,500 at that entry level, and higher with a more senior or larger team. Treat that as an order of magnitude to test against proposals, not as a price.

Four factors move the cost most. The first is the state of your order system: if orders have no clear state model, the payment integration will force you to create one. The second is the number of methods and markets. The third is recurring payments, which add stored-card management, retrying failed renewals and flagging merchant-initiated transactions. The fourth is finance’s requirements for reconciliation and reporting. Ask those four questions before anyone commits to a date.

What are the key takeaways before you sign with a payment provider?

Choosing a payment gateway looks like a procurement decision, but it is largely an architectural one. The fee matters, but it is easy to compare and easy to renegotiate. The integration model, the settlement reports and the way your application handles asynchronous notifications are much harder to change, because they grow into the system for years.

Remember six things. First, must-have criteria before price: payment methods, markets, the provider’s legal status and API quality eliminate more offers than the fee does. Second, the dominant local method in each market, whether cards and wallets in the US, iDEAL in the Netherlands or BLIK in Poland, should be your best-working path, including refunds and recurring payments. Third, compare cost on your own transaction mix with every charge included, and in the EU use your right to itemised fee information under Regulation 2015/751. Fourth, choose the integration model that minimises PCI DSS scope, and remember that even under SAQ A you must confirm your site’s resilience to script attacks. Fifth, treat the webhook as the source of truth, verify signatures, make handling idempotent and add a reconciliation job. Sixth, build your own payment state model and a provider adapter, and write exit terms, including card token export, into the contract before you sign it.

If you are planning a payment integration in a new application or rebuilding an existing one, we can help you run the technical evaluation of providers, design the state model and build the integration. See how we work in our custom software development services, or get in touch. We start with a discovery phase of 1–2 weeks that ends with a scope, an architecture and an estimate you can compare with proposals.