The idea behind smart routing is simple: not all payment paths are equal, so don't treat them that way. Send each transaction through the acquirer that gives you the best combination of approval probability, processing cost, and speed.
The execution is where it gets interesting. Here's how it actually works inside a modern payment orchestration setup.
The routing decision in real time
When a customer clicks "pay" on your checkout page, a routing engine evaluates the transaction before sending it anywhere. The whole process takes under 100 milliseconds — the customer has no idea it's happening.
The engine looks at a set of inputs: the card brand, the BIN (the first 6–8 digits of the card number, which identify the issuing bank), the cardholder's country, the transaction amount, the currency, whether it's a first-time or recurring charge, the time of day, and any custom signals you've configured (like product category or risk score).
It then runs that combination against performance data from your acquirer pool — every acquirer you're connected to — and selects the one most likely to approve the transaction at the lowest cost.
That's the simple version. Let's look at what each axis of the decision actually involves.
Axis 1: Approval rate
Every acquirer has a different approval rate profile. The differences aren't random — they're structural. An acquirer that has a strong relationship with a specific issuing bank will have higher approval rates for cards from that issuer. An acquirer with a domestic presence in the customer's country will get approved more often than one routing the transaction cross-border.
These differences can be significant. Domestic acquiring regularly outperforms cross-border acquiring by 10–20 percentage points on approval rate. For specific issuer-acquirer pairs, the gap can be even wider.
The routing engine tracks these patterns at a granular level. Not "Acquirer A has a 94% approval rate" — that's a blended number that hides important variation. Instead: "Acquirer A approves 97% of Visa cards issued by JPMorgan Chase, 91% of Mastercards issued by Revolut, and 85% of Amex cards issued in the UK." The routing decision is made at this level of specificity.
The data that feeds these calculations comes from your own transaction history. Every authorization attempt, every approval, every decline — the routing engine records the outcome and updates its model. Over time, it builds a detailed map of which acquirers perform best for which transaction types. The more volume you process, the more precise the routing becomes.
Axis 2: Cost
Not every transaction needs the highest-performing acquirer. If a transaction is likely to be approved regardless — say, a $30 domestic Visa purchase from a well-known issuing bank — you can route it through the cheapest available acquirer without meaningfully impacting approval probability.
The cost axis weighs processing fees, which vary by acquirer, card type, and region. Interchange rates are set by the card networks and are non-negotiable, but the markup your acquirer charges on top of interchange varies significantly. Some acquirers offer lower markups on domestic transactions. Others have better rates on corporate cards or cross-border traffic.
Smart routing uses this fee data as a secondary filter. After identifying the acquirers most likely to approve the transaction, it selects the cheapest option from that shortlist. If two acquirers have nearly identical approval rates for a given transaction type, the cheaper one wins.
For merchants processing significant volume, this cost optimization adds up. Even a 5–10 basis point savings on a portion of your transactions compounds across thousands of charges per month.
Axis 3: Geography
This is where the biggest approval rate gains tend to come from, and it's the axis most often ignored by merchants running a single-acquirer setup.
When a customer in Germany pays with a card issued by a German bank, and the transaction is processed through an acquirer with a German banking license, the issuing bank sees a domestic transaction. Domestic transactions get approved at much higher rates because the risk profile is lower — the acquirer is local, the patterns are familiar, the regulatory environment is shared.
If that same transaction is processed through a U.S. acquirer, the issuing bank sees a cross-border transaction. The risk flags go up. The approval rate drops.
This is why multi-acquirer setups that include regional acquirers consistently outperform single-acquirer setups for international merchants. If you sell to customers in Europe, having at least one European acquirer in your pool means your European transactions route domestically instead of crossing the Atlantic and back.
The same logic applies within large domestic markets. A transaction from a customer in Canada processed through a Canadian acquirer will generally perform better than the same transaction routed through a U.S. acquirer, even though the geographic distance is smaller. Banking systems are national, and issuers trust local acquirers more.
How cascading fits in
Smart routing optimizes the first attempt. Cascading handles the fallback.
When a transaction is soft-declined — meaning the issuer returned a decline code that suggests the transaction could succeed under different conditions — the engine can automatically retry it through a different acquirer. The customer doesn't see a declined payment. From their perspective, it went through normally.
Not all declines are eligible for retry. Hard declines (stolen card, closed account, insufficient funds) should not be retried — and responsible cascading engines don't. The card networks also impose limits on how many retry attempts are allowed for certain decline codes, and violating those rules can result in fines.
But for eligible soft declines — things like "issuer temporarily unavailable," "do not honor" (a catch-all that often means the issuer's risk model flagged the transaction), or "declined by card scheme" — cascading through a different acquirer frequently succeeds. The new acquirer sends the authorization through a different path, and the issuer evaluates it differently.
The combination of smart routing and cascading typically recovers 20–40% of soft declines. On a merchant processing $500K/month with a 7% decline rate, that's $7,000–$14,000 in monthly revenue recovered from transactions that would have otherwise been lost.
Rules-based vs. machine learning routing
Earlier routing engines were purely rules-based. You'd configure explicit rules: "If card is Visa and issuer country is France, route to Acquirer B." Those rules worked, but they were static. They didn't adapt when acquirer performance shifted, and they couldn't optimize for combinations of factors that humans wouldn't think to test.
Newer engines use machine learning to discover routing patterns in the data. Instead of relying on predefined rules, the model learns from every transaction: this card type, from this issuer, at this time of day, for this amount, was approved by Acquirer A at 96% and by Acquirer B at 91%. Over millions of data points, the model identifies optimal routes that a rules-based system would miss.
The best implementations combine both. ML-driven routing handles the bulk of optimization, while rules provide guardrails — things like "never send transactions over $10,000 through Acquirer C" or "always prefer domestic routing for recurring charges." The rules encode business logic. The model handles the pattern recognition.
