Skip to main content
WANFENG TECHNOLOGY
HomeServicesBenchmark SolutionsInsightsContact
☀☾Dark
▾
  1. Home
  2. /Insights
  3. /Cross-border site selection: nail six contract …
←Back to WANFENG Insights

Engineering Delivery

WANFENG INSIGHT — Traceable narrative, engineering receipts.

Nov 17, 2026

Cross-border site selection: nail six contract clauses

A multilingual storefront with payment integration is one line on a quote and almost nothing in a contract. Six variables to fix before signing: language list and URL scheme, payment channels and failure handling, order export fields, code and account ownership, renewals and SLA, section structure.

Choosing a storefront vendor for cross-border sales: six contract variables to pin down

One line on a quote — multilingual storefront, payment integration included — and almost nothing is actually agreed once it reaches a contract.

An October comparison of site-building vendors on ifeng.com ranked renewal pricing, data ownership, integrations and SLA ahead of price. That order matches how the work really goes. Page design is the only part you can change cheaply after launch. URL structure, order fields, account ownership and annual fees cost a full rework every time you touch them, and they drag SEO and finance reconciliation along with them. So the real question at selection isn't which vendor draws better pages. It's this: in three years, when you drop this vendor, can the whole site move with you?

Three options, all sold as custom

SaaS platform Template customization Full custom
Languages Plugin config; adding a language is cheap, URL rules are limited by the platform Depends on template mechanics, usually subdirectories Path rules defined by you
Payment channels Platform whitelist; local wallets wait for platform support Depends on the template ecosystem Connected one by one per acquiring entity
Order fields and export Constrained by the platform's data model Field changes quoted per request State machine and fields defined by the business
Source code and migration No source code; data exports, structure doesn't Partial delivery Full delivery
Annual cost Subscription plus plugins plus transaction fees Maintenance plus licence Maintenance plus servers

Adding a language on a SaaS platform is a config action, so it's cheap; the price is that URL rules, order fields and payment channels are all boxed in by the platform's model. Template work is the hardest to scope — how much gets changed lives in a verbal agreement and usually ends at "roughly like the demo". Full custom spends its schedule on the front end and integrations; the upside is a state machine, fields and permissions shaped by your business, the cost is that somebody has to run ops after launch.

A rough test: two languages or fewer, card payments plus one mainstream wallet, order fields close to the platform default — SaaS is enough. If you need local wallets, price tiers by customer type, quote numbers and sales-rep attribution on orders, template work or a custom build starts to pay for itself.

Languages: write the list, not "supports multiple languages"

Three things belong in the contract: a language list down to page level, the URL scheme, and who supplies the translated text.

Page-level means home, category, product, case study, blog and policy pages (privacy, returns, shipping, warranty) — which get translated and which don't. A common B2B pattern is product pages translated, blog not. That's fine. What isn't fine is stuffing an English blog into /de/ and calling it German, or pointing hreflang at a batch of empty pages. Priorities picked by feel are usually wrong: CSA Research's 2020 "Can't Read, Won't Buy" survey covered 8,709 consumers across 29 countries; 40% never buy from sites in another language, and 76% prefer product information in their own. Start with the market that produces the most orders — cheaper than launching eight languages at once.

The URL scheme has to be fixed in the contract. It's the hardest thing to change later. Subdirectories (/de/, /es/) are simple to build, pool domain authority, and are enough for most small cross-border sites; the trade-off is that a user can't tell from the URL which country they're being served. Subdomains (de.example.com) suit teams that deploy per country, at the cost of another DNS zone, certificate, monitoring setup and log pipeline. Country domains (example.de) give the strongest regional signal and local trust, cost more, and some countries restrict who may register them. Don't use ?lang=de — Google's documentation on multilingual sites advises against it, and you can't segment analytics by market.

Changing structure after launch means a full 301 map. Miss one URL out of several hundred and the rankings that URL carried are gone.

hreflang can go into the contract as acceptance criteria: language versions link to each other, an x-default is present, canonicals point at the page itself rather than at the English version. The language switcher should be an ordinary link in the header or footer, not an IP or browser-language redirect. Googlebot crawls mostly from the US without an Accept-Language header — one redirect and your German and Japanese pages may never be fetched.

Who supplies the translation changes the price by a multiple: client-ready copy, machine translation with human review, or full professional localization. MT isn't a problem in itself; generating thousands of pages nobody has reviewed is. Put the unit price for adding a language, the timeline and ownership of the translation memory in their own clause — "extensible later" is not a clause.

Payments: beyond the list, write the failure path

"Payment integration included" needs splitting into at least four items: the channel list, the acquiring entity, what happens on failure, and who owns refunds and chargebacks.

The channel list goes down to the payment method. Confirm the acquiring entity early: an entity in your own market, a Hong Kong entity or a mainland China entity each opens a different set of channels and settlement cycles. This blocks signing if it isn't settled first; everything afterwards becomes rush work.

Failure handling is what gets skipped. Three mismatches show up in production: the payment succeeds and no order is created (an async callback lost or a signature check failed), the order is created and the payment never completes (the front-end redirect breaks), and the same transaction is submitted twice, producing two orders. The engineering answer is verify callback signatures, use the channel transaction ID as the idempotency key, let the order state machine move in one direction only, log every failed callback somewhere searchable, and alert when the anomaly count crosses a threshold. The contract doesn't need those nouns. It needs who locates a mismatch and within how long, what compensation applies when that time limit is missed, who may issue a refund on the channel side, and where that action is recorded — not in someone's personal account.

Refunds and chargebacks are different things. A refund is the merchant initiating. A chargeback is the cardholder disputing with the issuing bank, and the merchant has a window to submit evidence: shipping documents, proof of delivery, the order's IP and device data, the support correspondence. Who stores that evidence, and for how long, belongs in an appendix. [TO VERIFY: chargeback response windows and evidence lists vary by acquirer — check the actual acquirer's documentation before signing.]

Order export: the field list matters more than the button

An Export CSV button in the admin does not mean finance can reconcile.

Build the field list around what finance needs to check: order number and channel transaction ID, order time and payment time (with the timezone stated — cross-border stores routinely mix UTC with local time and end up a day apart), the amount in the original currency and in the settlement currency, the FX rate source and the moment it was taken, shipping, whether tax is included or excluded, discount allocation, the status change history, the UTM and landing page a customer arrived from, and the inquiry or sales rep the order belongs to.

Agree the export method too: scheduled CSV, API pull or webhook push. Finance reconciles the acquirer's settlement report against orders in the store, and the transaction ID is the only field that reliably joins the two — without it, someone is guessing by hand. Add the retention period for historical data, the per-export cap, and the API's pagination and rate limits. A store doing a thousand orders a day with no pagination on the export endpoint will slow the database down every time someone runs it.

Data ownership and source code: can the site move?

List the domain registrant, cloud account, servers, database backups, Search Console and GA4 properties, the merchant account holder, certificates and the CDN account. Item by item, with a delivery list attached.

Source code and documentation are two sides of one thing. Hand over the code without deployment docs and a database schema and nobody can wire it up. So the contract asks for source code, deployment documentation, the third-party account list, integration notes, and 30 days of migration assistance after the engagement ends.

Renewals and SLA: the pain starts in year two

Build fees are one-time; renewals happen every year. Domain, servers, SSL, CDN, plugins or licences, maintenance — price each line, and state the notice period for increases (60 days' written notice is a common standard) plus either a fixed rate card or a cap on the increase. The ifeng.com piece listed written confirmation of renewal pricing as a checklist item, and for cross-border sites it's very real: a low year-one bundle followed by catalog pricing in year two is a familiar pattern.

SLA wording: availability measured monthly, planned maintenance windows excluded, response and recovery targets, credits as a percentage of the monthly fee, and the exclusions — content updates, image work, campaign pages, outages caused by a third-party service being withdrawn.

Site structure: SEO and AI search both depend on this layer

Structure isn't a design question, it's a selection question. The hierarchy of category pages, product pages, case studies, technical docs and the blog decides URLs and internal links. Change it after launch and you're doing bulk redirects.

Being cited by AI search starts with crawlable HTML. Product spec tables rendered entirely by front-end JS hand the crawler an empty shell. Structured data (Product, Organization, FAQPage) has to validate. And don't casually block mainstream AI crawlers' user agents in robots.txt. The acceptance criteria can be written as: core pages return server-rendered HTML, structured data validates, the sitemap is split by language, and robots.txt states which crawlers are allowed and which aren't.

Three questions during selection: can the language list land at page level, can the order export match finance's settlement report, and can the whole site move if you change vendors. Answer those three, then talk about price.

Sharing & attribution

Originally from WANFENG TECHNOLOGY, published in WANFENG Insights. Feel free to share—please link back to this page and credit WANFENG TECHNOLOGY.

You may also like

  • Technical Practices · Nov 14, 2026

    Custom Mini Program Storefront: Check Data Before Features

    Read →

  • Technical Practices · Nov 10, 2026

    AI integration: pick the embedding layer before the model

    Read →

  • Engineering Delivery · Nov 7, 2026

    Custom development contract: five clauses to check first

    Read →

Unsure what you need?

Talk 1:1 with a senior advisor. We will clarify your goals and return a plan with a ballpark quote—free of charge.

Book a free scoping call

No hard selling. Pure technical and business discussion.

WANFENG TECHNOLOGY

Custom development, technical SEO, UI/UX motion, and private LLM programs in one delivery chain. 10+ years of engineering and growth for domestic and global scenarios.

On this page

  • Home
  • Insights
  • Benchmark Solutions
  • Contact
  • RSS feed

Business inquiries

Reach us by email, WeCom, or your usual business channel.

© 2026 WANFENG TECHNOLOGY. All rights reserved.

粤ICP备2020119225号