Payment Card Tokenisation: Network Tokens vs PSP Tokens
Network tokens, PSP tokens and merchant vaults solve different problems. Here is how each shapes authorisation, chargebacks, PCI scope and vendor lock-in at checkout.
Key takeaways
- A network token is issued by the scheme and bound to a merchant, wallet or device domain
- A PSP or gateway token normally lives only inside that provider's vault, so it does not travel when you switch provider
- Tokenisation reduces stored-credential fraud, but chargeback liability still follows scheme dispute rules
- The decision with the longest half-life is who holds the Token Requestor ID, because that determines whether your token estate is portable
Three credentials, three owners
When a customer saves a card at an online checkout, the card number is stored somewhere. What differs between implementations is who stores what, in whose name, and what is left in your environment when the transaction ends.
Payment card tokenisation replaces the primary account number (PAN) with a surrogate value. That is the simple part. The hard part is that "a token" can mean three structurally different things, and those differences decide authorisation performance, chargeback handling, PCI scope, and how expensive it is to change provider later.
EMVCo, which owns the EMV chip specification, maintains the EMV Payment Tokenisation Specification. It defines an EMV payment token as a PAN substitute that carries Token Domain Restriction Controls — constraints that tie the token to a specific merchant, device or payment scenario. The framework separates the Token Requestor (the registered entity that asks the network for tokens) from the Token User (the entity that presents the token in a transaction). Those are frequently different companies, and that separation explains most portability arguments. The framework also notes that other token types — acquirer, merchant or issuer tokens — sit outside it and coexist with it. In practice, that means three layers.
Comparing the three models
| Network token | PSP / gateway token | Merchant vault token | |
|---|---|---|---|
| Issued by | Scheme token service (Visa VTS, Mastercard MDES, Amex AETS) | Your payment provider or gateway | You, or a third-party vault vendor |
| Who holds the underlying PAN | Issuer and scheme | The PSP's own vault | Your vault (or your vendor's) |
| Works outside the issuer's system | Yes — accepted across the scheme's ecosystem | No — only inside that provider's platform | No — only inside your own systems |
| Card lifecycle updates | Yes, through the scheme's update services | Depends on the provider's updater integration | Manual |
| Typical PCI scope impact | Merchant systems that never touch a reversible credential can sit outside the cardholder data environment | Similar, if the provider is in scope instead of you | You carry the scope for the vault itself |
A network token is generated by the scheme's Token Service Provider. Mastercard describes network tokens as issued by payment networks, linked to a merchant and a card, and interoperable across the payment flow — through PSPs, acquirers and other providers through to the issuer (Mastercard, Network tokenization: powering the e-commerce of today and tomorrow, white paper, April 2026). Because the token is generated inside the scheme, de-tokenisation happens at the issuer's Token Service Provider as the authorisation, refund or chargeback is routed — not in your systems, and not in your provider's systems. That is what makes a network token useful outside the four walls of whoever requested it.
A PSP or gateway token is a reference value in a provider's database. It may map to a network token behind the scenes, or it may map to a PAN that the provider holds for you. Mastercard's gateway documentation describes both models running side by side: a merchant can integrate directly with a scheme token service, or let the gateway act as the token requestor and link gateway tokens to network tokens automatically (Mastercard Gateway integration guide, network tokenisation).
A merchant vault is the oldest model: you, or a vault vendor, store the PAN in encrypted form under your own PCI DSS obligations and issue internal tokens. Most control, most operational work.
What network tokens change at authorisation
Scheme data suggests tokenised transactions authorise more often than PAN-based ones in card-not-present channels. Visa's Token Service fact sheet reports an average observed authorisation rate lift of 3.2% and a 26% reduction in fraud by payment volume for tokenised transactions compared with PAN-based transactions online (Visa Token Service Fact Sheet, "Success stories"). Visa's corporate material quotes different figures — a 30% reduction in online fraud versus PAN and a 4% authorisation uplift, with a 4.6% lift specifically for card-not-present traffic (Visa, A Deep Dive into Tokenized Transactions). The spread between those numbers is worth noting: they cover different periods, channels and measurement bases, so treat any single figure as directional rather than as a forecast.
Two mechanisms drive the effect, and both matter to product teams:
- Per-transaction cryptograms. A token presented in an e-commerce flow is accompanied by cryptogram data that the issuer validates. A stolen token copied out of its domain cannot simply be reused elsewhere, because the domain restriction controls reject it.
- Lifecycle updates. When a card is reissued after loss, fraud or expiry, the scheme can update the token binding, keeping credential-on-file and recurring payments alive without the customer re-entering a card.
The caveat: uplift figures are averages published by the schemes themselves, measured on their own networks. They vary by BIN range, geography, vertical and how much of the traffic was already authenticated.
Chargebacks and liability: what tokenisation does not change
A token is not a liability shift. It changes what a fraudster can steal; it does not change who owes money when a cardholder disputes a charge. Liability still follows scheme dispute rules, which turn on the dispute reason, the evidence you can supply, and whether the transaction was authenticated.
Mastercard and Checkout.com reported that merchants in Checkout.com's portfolio using network tokens saw a 49% reduction in fraud-related chargebacks alongside a 10.3 percentage-point increase in approval rates (Mastercard newsroom, 7 May 2026). Two things to note about that result: it is one PSP's merchant portfolio within a vendor partnership, and "fraud-related" chargebacks are only part of total disputes. Commercial disputes — non-delivery, not-as-described, subscription cancellations and friendly fraud — are unaffected by how the number was stored.
In the EEA, strong customer authentication under the PSD2 regulatory technical standards on authentication and communication (Commission Delegated Regulation (EU) 2018/389) is a separate control from tokenisation and does not switch off because a token is used. Merely storing a token does not satisfy SCA, and a tokenised transaction still needs the appropriate authentication or an applicable exemption.
Tokenisation does help dispute handling in one specific way. EMVCo's Payment Account Reference (PAR) lets acquirers, PSPs and merchants link transactions to the same underlying account across different tokens without exposing the PAN, giving dispute and support teams a consistent identifier when a cardholder holds several tokens across several channels.
Portability and lock-in: the Token Requestor question
This is where cost shows up years later. The question to ask a provider is not "do you issue network tokens?" but "who is the Token Requestor of record on my tokens?"
- If your PSP holds the Token Requestor ID, the scheme registration, provisioning history and the merchant identifier the token is bound to all belong to the provider. Switching provider means re-provisioning tokens against the underlying PANs.
- If you (or a neutral vault sitting above the processors) hold your own Token Requestor ID, the relationship is yours to carry — though tokens are also bound to the merchant identifier your provider operates, so a new provider still means a new identifier and a re-provisioning exercise.
Mastercard's merchant-boarding documentation makes the constraint concrete: a Visa Token Requestor ID configured against a token repository cannot be changed once set, and using a new identifier requires a new token repository (Mastercard Gateway integration guide, network tokenisation for merchants).
Moving stored credentials between processors is not a matter of exporting strings. Stripe, for example, publishes its requirements for PAN export as a processor-to-processor transfer with a compliance attestation and an encryption key exchange. The merchant never touches the PANs — which is exactly what keeps the merchant out of scope, and also what makes the migration a project rather than a click.
There is a second trap for recurring-billing businesses: merchant-initiated transactions must reference the identifier from the original cardholder-initiated transaction. If a migration delivers PANs and expiry dates but not those linkage identifiers, rebills can present to issuers as first-time charges and attract declines that no retry logic fixes. Write that field requirement into the migration plan before signing, not after the export request.
Costs to model
Tokenisation costs are rarely a single line item. The realistic list:
- Scheme or provider tokenisation fees. Per-token or per-transaction charges, where they apply, sit in the scheme fee schedule and in your provider contract. They differ by scheme, region and product, so read the schedule rather than assuming.
- Engineering and integration. Provisioning flows, cryptogram handling, mapping between your internal token and a network token, and re-provisioning during any provider change.
- PCI DSS assessment scope. The PCI Security Standards Council's Tokenization Guidelines information supplement is explicit that tokens which can be exchanged back to a PAN, and any system connected to the tokenisation system, remain in scope. Only systems that store or process nothing but irreversible tokens, and are properly segmented, may fall outside.
- Vault operations, if you host one. Cryptography and key management, access controls, logging, deletion on retention expiry, and the assessment burden of running a data vault that the Council itself describes as the most attractive target for attackers.
- Migration cost. Re-authorisation campaigns, involuntary churn during re-provisioning, and support load.
None of these are reasons to avoid tokenisation; they are reasons to decide deliberately which layer you own.
Jurisdiction notes
Rules that touch tokenisation differ by market, and scheme rules that are global in principle are enforced regionally:
- In the EEA, SCA applies to card-based payments under the PSD2 RTS. Tokenisation is an additional security layer, not an exemption.
- In the US, routing is a live constraint. Mastercard's gateway documentation states that U.S.-based merchants cannot process Visa Token Service tokens for debit or prepaid card transactions routed to domestic debit networks (Mastercard Gateway integration guide, network tokenisation). That affects token strategy for merchants with large debit volumes, because co-badged network tokens can be routed to competing US debit networks and the routing choice interacts directly with cost.
- In India, the central bank has pushed the strictest version of the model. Its circular of 7 September 2021 (RBI/2021-22/96) states that, with effect from 1 January 2022, no entity in the card transaction or payment chain other than card issuers and card networks may store actual card data, with only the last four digits and the issuer name retained for reconciliation. Later circulars extended the timeline before fixing the purge deadline at 1 October 2022 (RBI circular of 28 July 2022). That removes the merchant-vault option as a compliance choice.
Risks and costs
- Residual fraud. Tokenisation protects the credential; it does not authenticate the cardholder. Account takeover, stolen credentials entering at an unmanaged checkout, and social-engineering fraud remain.
- Credential lifecycle risk. Wallet tokens live at the scheme and device layer, so you cannot re-provision your way out of a customer removing a card from a wallet.
- Vendor concentration. If one PSP holds your Token Requestor ID, your stored-credential base inherits that provider's commercial behaviour. Negotiate export rights, migration clauses and audit rights while you have leverage.
- Scope creep. A vault that can de-tokenise is in PCI scope, no matter how the tokens are labelled.
- Cost stacking. Scheme tokenisation revenue, provider platform fees and the engineering effort of re-provisioning can outweigh the authorisation gain for low-value, low-frequency card-on-file books.
- Compliance drift. Rules such as recurring-payment linkage requirements are updated on published scheme timelines. A token estate built once and not monitored can quietly stop meeting them.
Bottom line
Network tokens are the strongest credential layer for e-commerce and card-on-file payments: scheme-issued, domain-restricted, updated when the card changes, and recognised by issuers in authorisation. PSP tokens are a convenience layer that usually works well inside one provider relationship, and merchant vaults trade operational burden for control.
The decision that determines the cost of the next migration is not which token type you use, but who holds the Token Requestor ID and what your contract says about getting your credential base out. For product teams, the practical sequence is: provision network tokens where the merchant identifier arrangement allows it, understand exactly which identifier your rebills carry, price tokenisation fees against a measured authorisation uplift in your own traffic, and treat portability as a contract term rather than a feature request.
Follow the shortlist
Following payments, banking and market-infrastructure change? Our global ETF and stock research watchlist is tracked daily. Premium members get the full list, the reasoning behind every entry and updates when entries change.
See premium accessInformation and education only, not investment advice.
Related
Stablecoin Regulation Compared: EU MiCA vs the US GENIUS Act
How MiCA and the GENIUS Act differ on who may issue payment stablecoins, what must back them, how redemption works and where depeg risk still sits.
FinTechTokenised Deposits vs Stablecoins: How Bank-Issued Money Differs
Tokenised bank deposits and reserve-backed stablecoins both move value on programmable rails, but they differ in legal claim, deposit insurance, balance-sheet treatment and risk.
FinTechInstant Payments for Small Business: FedNow, RTP, Zelle Compared
How FedNow, RTP and Zelle differ for US small businesses on speed, settlement finality, limits and cost, plus which emerging options fit SME cash flow.