FG Media's blog : Automatic Transaction Routing: How Smart Routing Boosts Authorization Rates

A declined card payment rarely means the customer has no money. Issuers decline for reasons that range from a fraud model's suspicion to a temporary network outage, and some of those reasons depend on which acquiring bank submitted the transaction. Automatic transaction routing takes advantage of that fact: instead of sending every sale through one merchant account, the gateway picks the best account for each transaction based on card brand, monthly volume caps, or custom logic. NMI's version of this is called the Automatic Transaction Routing Interface (ATRI), and merchants and partners who manage it reach the gateway, Merchant Central, the customer vault, and reporting tools through the NMI login hub described on nmi-portal.us.org. This article explains how routing rules work, where approval gains come from, how the same logic lowers processing costs, and which edge cases can turn a routing setup into a compliance problem.

Why a Single Merchant Account Leaves Approvals on the Table

A merchant ID (MID) is an account with one acquirer, processed through one processor. Every transaction on that MID inherits the acquirer's risk profile, its connection to the card networks, its fraud screening, and its contractual limits. When anything in that chain misbehaves, every sale is affected at once.

Three failure patterns show up repeatedly in single-MID setups.

The first is volume caps. Acquirers underwriting new, fast-growing, or higher-risk businesses often set a monthly processing limit, for example $100,000 per calendar month. Once the merchant crosses it, the acquirer may hold funds, decline further sales, or open a review. A business that sells out a product launch on day 20 of the month can find its checkout dead for the last ten days.

The second is issuer perception. Issuers score each authorization request using, among other signals, the acquirer's history and the merchant category code (MCC). A cross-border acquirer submitting a domestic cardholder's purchase often sees lower approval than a domestic acquirer would for the same card and the same amount. The merchant cannot fix this from inside one MID.

The third is outages. Processors have maintenance windows, degraded connections, and occasional full outages. With a single MID, a processor incident turns into lost revenue for its entire duration, and response codes such as 91 (issuer or switch inoperative) or 96 (system malfunction) pile up in the logs.

What Automatic Transaction Routing Does

Routing moves the decision of "which MID processes this sale" from a static configuration to a rule evaluated at authorization time. The merchant integrates once with the gateway. Behind that integration sit several MIDs, potentially with different processors, and a rule set that selects among them.

The sequence for a typical card-not-present sale looks like this. The checkout sends the payment request to the gateway. The gateway reads the attributes available before authorization: card brand from the BIN, card type (credit, debit, prepaid), issuing country, transaction amount, currency, and any merchant-defined fields such as product SKU or website. It evaluates the routing rules against those attributes and against the current state of each MID, such as month-to-date volume. It submits the authorization to the selected MID. If the response is an approval, the sale completes. If it is a decline eligible for retry, the gateway can cascade the attempt to a secondary MID.

NMI describes ATRI as splitting volume across an unlimited number of MIDs based on monthly caps, card brands, or custom logic, with automatic decline cascade routing. Because NMI connects to more than 200 processors, the MIDs in a rule set do not have to belong to the same acquirer, which is where most of the resilience comes from.

Routing by Card Brand

Card brand routing sends each network to the account that handles it best. The classic case is American Express. A merchant can accept Amex through a processor's OptBlue program, where Amex volume is priced and settled alongside Visa and Mastercard, or through a direct Amex merchant agreement. Some merchants hold both and route by amount or product line. Discover has a similar split between acquirer-handled and direct relationships.

Visa and Mastercard can also be separated when one acquirer has noticeably better approval rates or pricing for one of the two networks. Brand routing is deterministic: the BIN tells the gateway which network the card belongs to before any authorization is attempted, so the rule never guesses.

Routing by Monthly Caps

Cap-based routing keeps each MID under its contractual limit. The merchant enters each account's monthly ceiling, and the gateway tracks month-to-date processed volume. Two strategies are common. Fill-and-spill routes everything to the primary MID until it reaches a threshold, often set below the true cap to leave room for late refunds and reversals, then spills to the next MID. Proportional balancing splits volume by percentage, for example 50/30/20 across three MIDs, so each account shows steady, predictable activity instead of a sudden spike after the primary fills.

Proportional balancing tends to be gentler on underwriting. An acquirer that sees a MID go from zero to its full cap within a week may treat the pattern as suspicious, while a MID that grows gradually looks like an ordinary business.

Routing by Custom Rules

Custom logic covers everything that is not brand or cap. Typical rules route by transaction amount (sending tickets above $1,000 to the acquirer with the strongest approval record on large purchases), by issuing country (domestic cards to a domestic acquirer, foreign cards to an acquirer with local presence in that region), by currency, by storefront or product category, or by card type (debit to the account with the cheapest debit pricing). Rules can be layered, with a brand rule evaluated first and a cap rule applied within the brand group.

Routing method	Decision input	Main benefit	Typical use case	Watch-outs
Card brand	BIN-derived network (Visa, Mastercard, Amex, Discover)	Pricing and approval tuned per network	Separating Amex OptBlue from direct Amex, or splitting Visa and Mastercard across acquirers	Brand-specific contract terms and settlement timing differ
Monthly cap	Month-to-date volume versus each MID's limit	Prevents holds and freezes from cap breaches	High-growth or seasonal merchants with underwritten limits	Refunds and chargebacks still count against the original MID
Proportional split	Percentage weights	Steady volume history on every account	Maintaining multiple active acquirer relationships	Weights must be revisited as approval data comes in
Amount-based	Transaction value	Directs large tickets to the strongest approver	Electronics, travel, B2B invoices	Thresholds need testing, not guessing
Geography or currency	Issuing country, presentment currency	Domestic-style processing, fewer cross-border fees	International e-commerce	Requires acquirers licensed in the relevant regions
Decline cascade	Soft decline response from first MID	Recovers sales that would otherwise be lost	Any multi-MID setup	Must follow network retry rules and exclude hard declines

Decline Cascading: Where Most of the Approval Gain Comes From

Cascading means resubmitting a declined authorization through a different MID. Done correctly, it recovers sales lost to acquirer-specific or processor-specific problems. Done carelessly, it generates excessive retries, network fines, and a worse fraud profile with issuers.

The key distinction is between soft and hard declines. A soft decline signals a condition that may not exist on another attempt or another route: an issuer timeout, a system error, a generic refusal that may reflect the acquirer's risk score. A hard decline signals a permanent problem with the card or account: a lost or stolen card, a closed account, an invalid number. Retrying a hard decline through a second acquirer never produces a legitimate approval and can look like card testing.

Visa groups decline codes into categories, and for the category that means the issuer will never approve, it prohibits reattempts. For declines that may be approved later, Visa limits reattempts on the same card and merchant to 15 within 30 days. Mastercard sends Merchant Advice Codes (MACs) alongside declines; MAC 03 and MAC 21 tell the merchant not to try again, and repeated retries after such advice can trigger excessive-retry fees. A cascade rule set has to read these signals, not just the two-digit response code.

Response code	Meaning	Decline type	Cascade to another MID?
05	Do not honor	Soft (generic)	Often yes, once; results depend on issuer
91	Issuer or switch inoperative	Soft (technical)	Yes
96	System malfunction	Soft (technical)	Yes
51	Insufficient funds	Soft (cardholder)	No immediate cascade; a different acquirer does not add funds
61 / 65	Exceeds amount or frequency limit	Soft (cardholder)	No immediate cascade; later retry may be allowed
54	Expired card	Hard	No; update card data
14	Invalid card number	Hard	No
41 / 43	Lost card / stolen card	Hard	Never
04 / 07	Pick up card	Hard	Never

Code 05 deserves attention because it is the most common and least informative decline. Issuers return it for fraud suspicion, for internal policy, and sometimes for reasons tied to the acquirer or MCC. A single cascade on 05 to a domestic acquirer with a clean history is where merchants usually recover the most sales. A second or third cascade on the same card rarely helps and raises the retry count toward network limits.

How Routing Lowers Processing Costs

Approval gains get most of the attention, but the same rule engine directs volume toward cheaper processing. Card acceptance cost has three layers: interchange paid to the issuer, network assessment fees, and the acquirer's markup. Interchange and assessments are set by the networks and depend on card type, merchant category, and how the transaction is submitted. The markup is negotiated. Routing touches all three.

Pricing models differ between acquirers. One account might be priced interchange-plus with a small fixed markup, another on a flat or tiered schedule. On interchange-plus, a regulated debit card costs far less than a premium rewards credit card, while a flat rate charges both the same. A card-type rule that sends debit volume to the interchange-plus MID and keeps the flat-rate MID for other traffic captures that difference without any change to checkout.

Debit routing in the United States has a specific regulatory angle. Regulation II, which implements the Durbin Amendment, caps interchange on debit cards issued by large banks and requires debit cards to support at least two unaffiliated networks. Since July 2023 that requirement explicitly covers card-not-present transactions, which gives e-commerce merchants and their acquirers a choice of network on many debit sales. Acquirers that support least-cost debit routing can pass that saving through.

Cross-border fees are the next target. When a merchant's acquirer is in one country and the cardholder's issuer is in another, networks usually apply cross-border assessments and the issuer may apply a higher interchange rate. A geography rule that sends European cards to a European acquirer and US cards to a US acquirer converts many cross-border transactions into domestic ones, often cutting cost and improving approvals at the same time.

The least visible saving is negotiating leverage. A merchant whose volume sits entirely with one acquirer has little room to push on markup. A merchant who can shift 30% of its volume to another MID with one configuration change negotiates from a different position. Token portability matters here, as the next section explains.

Tokens, the Customer Vault, and Why Routing Needs Processor Independence

Stored card credentials are the hidden dependency in any routing project. If a processor issues the tokens a merchant uses for saved cards and subscriptions, those tokens work only on that processor. Routing a returning customer's payment to a different MID becomes impossible without asking the customer to re-enter the card.

A gateway-level vault solves this by storing the card data once, at the gateway, and letting any connected MID charge the token. NMI's customer vault stores card data in a PCI DSS Level 1 environment, supports token portability across processors, and includes an automatic card account updater that refreshes expired or reissued card numbers. NMI states that vault records remain the merchant's property and can be exported under PCI transfer procedures.

For routing, the practical consequence is that a subscription renewal or one-click checkout can go through whichever MID the rules select, and a cascade on a stored-card transaction works the same way as on a new card. The account updater also removes a large share of code 54 (expired card) declines before they happen, which no routing rule could fix.

Practical Scenario: Rebuilding Routing for a Growing Online Retailer

The following scenario is illustrative. The figures are hypothetical and chosen to show the mechanics, not to predict results.

An online retailer processes about $400,000 per month, with roughly 70% Visa and Mastercard credit, 20% debit, and 10% American Express. Its single acquirer has set a $300,000 monthly cap. Around the 22nd of each month, the account nears the cap, sales slow, and the finance team manually pauses promotions. Roughly 8% of authorizations decline, with code 05 as the largest share, followed by 51 and 54.

Open additional accounts. The retailer is approved for a second MID with a different domestic acquirer at a $250,000 cap, priced interchange-plus, and enrolls in direct American Express acceptance alongside its existing OptBlue coverage.
Move stored cards to the gateway vault. Saved cards and subscription profiles are consolidated in the customer vault so tokens work on either MID, and the account updater is switched on.
Set the brand rule. American Express transactions above a chosen threshold go to the direct Amex account; smaller Amex tickets stay on OptBlue through the primary MID.
Set the card-type rule. Debit transactions route to the interchange-plus MID, where regulated debit interchange passes through at cost plus markup.
Set proportional balancing for remaining credit volume at 60/40 between the two MIDs, with each MID's routing threshold at 90% of its cap to absorb refunds and late adjustments.
Configure cascading narrowly. Codes 91 and 96 cascade automatically to the other MID. Code 05 cascades once. Codes 51, 54, 14, 41, 43, 04, and 07 never cascade, and Mastercard declines carrying MAC 03 or 21 are excluded.
Run for one full month and compare. The team pulls reports by MID and by card BIN range, comparing approval rate, soft decline share, and effective cost per approved transaction against the prior month.
Adjust weights. If one acquirer approves 05-prone BIN ranges more often, the team raises its weight for those ranges instead of relying on cascades, which reduces retry counts toward network limits.

Two things change immediately in this setup. The cap problem disappears because combined capacity is $550,000 with neither account run to its ceiling. Technical declines stop costing sales because a processor incident on one side routes around itself. The rest of the gain, from 05 recoveries and cheaper debit processing, shows up only after the data from step 7 is in hand.

Edge Cases and Compliance Risks

Routing is a legitimate optimization tool. The card networks also treat certain uses of multiple MIDs as abuse, and a merchant needs to know where that line sits.

Splitting volume to evade monitoring is prohibited. Visa and Mastercard run chargeback and fraud monitoring programs with thresholds measured per MID. Deliberately spreading sales across accounts so that no single MID crosses a chargeback threshold, sometimes called MID stacking or load balancing for evasion, violates network rules and can lead to fines, account termination, and placement on the MATCH list that other acquirers check during underwriting. Routing decisions should be justified by cost, approvals, capacity, or resilience, and each acquirer should know the merchant uses others.

Refunds have to follow the original sale. A refund must be issued through the MID that processed the original authorization. The gateway normally handles this by referencing the original transaction ID, but manual refunds from a virtual terminal can go wrong if staff pick the wrong account.

Recurring billing needs consistency. Some issuers flag subscription charges that suddenly arrive from a different acquirer or with a different descriptor. For established subscribers, it often makes sense to keep renewals on the MID that processed the initial transaction and apply routing mainly to new customers or cascades.

Descriptors can drive chargebacks. Each MID has its own billing descriptor. If one says the store name and another shows a legal entity name the customer does not recognize, "I don't recognize this charge" disputes rise. Descriptors should be aligned across accounts wherever the acquirers allow it.

Reconciliation gets harder. Several MIDs mean several settlement deposits, fee statements, and batch schedules. Finance teams need gateway reports broken out by MID to match deposits to sales.

Authentication results do not always transfer. A 3D Secure authentication is tied to transaction data that includes the acquirer and merchant details. Cascading an authenticated transaction to a different MID may require re-authentication, otherwise the liability shift may not apply.

Measuring Whether Routing Works

The headline approval rate hides most of what matters. A useful review compares the same metrics per MID, per card brand, and per BIN range, over the same period. The measures that carry the most information are these:

Approval rate on first attempt, per MID, which shows the true baseline performance of each acquirer before cascades mask it.
Cascade recovery rate, meaning the share of cascaded soft declines that the secondary MID approves; a low figure means cascade rules are too broad.
Hard decline share, which routing cannot improve and which points instead to card data quality, account updater coverage, or fraud upstream.
Effective cost per approved transaction, combining interchange, assessments, and markup from each MID's statement, divided by approved sales.
Chargeback ratio per MID, tracked monthly against network thresholds, so that a routing change never pushes one account toward a monitoring program.
Month-to-date volume against cap per MID, checked weekly during high season rather than at month end.

A month of data is a minimum. Approval patterns vary by day of week, pay cycles, and issuer model updates, and short tests often reward random variation.

Managing Routing Through the NMI Portal

Merchants usually manage routing, reports, and the customer vault through the merchant portal, while ISOs and other partners use a partner portal that lists merchant accounts, residuals, and the processor matrix. NMI is a white-label gateway, so the login screen may carry the name of the ISO or payment facilitator that set up the account; the platform underneath is still NMI. The official sign-in hub is at nmi.com/logins, and merchants should bookmark the exact URL their processor sent.

Access problems follow a few patterns. The username comes from the welcome email and is not necessarily the company email address. The Forgot Password option on the login page sends a reset link that stays valid for 24 hours. When the primary merchant user is locked out, the partner can reset it from List Accounts, under Merchant Users. Partner accounts without an authenticator app enabled fall under a 90-day password policy, so enabling two-factor authentication reduces forced resets.

FAQs

What is the difference between transaction routing and load balancing?

Load balancing is one type of routing: it splits volume across MIDs by percentage or by cap to spread processing evenly. Transaction routing is the broader concept and includes brand rules, card-type rules, geography rules, amount thresholds, and decline cascading. A routing engine can apply load balancing inside a group selected by another rule.

Does smart routing work for small merchants with one merchant account?

Routing needs at least two MIDs to choose between. A merchant with one account gains nothing from routing rules themselves, although a gateway-level vault and account updater still reduce expired-card declines. Routing starts to pay off when a business nears a volume cap, sells internationally, or loses meaningful revenue during processor outages.

Is it allowed to retry a declined card on a different processor?

Yes, for soft declines and within network rules. Visa prohibits reattempts on its never-approve decline category and limits other reattempts to 15 within 30 days. Mastercard's Merchant Advice Codes 03 and 21 mean the merchant should not retry. Hard declines such as lost, stolen, or pick-up card responses should never be retried on any MID.

Can routing reduce chargebacks?

Indirectly. Routing domestic cards to domestic acquirers and aligning billing descriptors across accounts can reduce disputes from cardholders who do not recognize a charge. Routing must never be used to spread chargebacks across MIDs to stay under monitoring thresholds, which card networks treat as a rule violation.

What happens to saved cards and subscriptions if I add a new processor?

If the tokens are issued by the old processor, they usually cannot be charged through the new one. Storing card data in a gateway-level vault such as NMI's customer vault keeps tokens independent of any single processor, so subscriptions and saved cards can be charged through whichever MID the routing rules select.

How long before routing shows measurable results?

Cap relief and outage protection take effect as soon as a second MID is live. Approval and cost gains need at least one full billing month of per-MID data, because issuer behavior shifts by day of week and pay cycle. Weights and cascade rules should be adjusted after that first review, not during it.

Conclusion

Automatic transaction routing turns a merchant's payment setup from a single point of failure into a set of accounts chosen per transaction. Card brand rules put each network on its best-priced account. Cap rules keep every MID inside its underwritten limit. Custom rules send debit, large tickets, and foreign cards where they approve best and cost least. Cascading recovers soft declines that one acquirer refuses and another accepts, provided hard declines and network retry limits are respected. The foundation is a processor-independent vault, so stored cards follow the rules instead of blocking them. The limits are equally concrete: refunds follow the original MID, descriptors must match, and volume must never be split to hide chargebacks. A merchant that measures first-attempt approvals, cascade recovery, and cost per approved sale per account, and adjusts rules on real data, gets both higher authorization rates and a lower processing bill from the same configuration.

In: