Routing decides where a transaction goes before an issuer ever sees it. Smart routing optimizes for approval rate by picking the acquirer or path most likely to get a yes. Least-cost routing (LCR) optimizes for processing cost by picking the cheapest eligible network. Both run in the same orchestration layer, both look like "routing" on a slide, and they frequently want to send the same transaction in two different directions.
Module 2 covered the authorization funnel and Module 4 covered tokenization. Here we deal with the layer that sits above them: deciding which rail a transaction travels before authorization happens at all. Treated carelessly, that layer is where a checkout team can lose a point of approval rate chasing a basis point of savings.
What routing actually controls
Routing operates on transaction metadata available at request time: BIN, card type, issuing country, amount, currency, MCC, and whether a network token is present. A rules engine matches that metadata against an ordered set of conditions and selects the first eligible path. The decision runs in milliseconds and the cardholder never sees it.
There are two distinct moves. The first is acquirer or PSP selection: sending the transaction to one of several connected acquirers. The second, specific to debit in the US and a few other markets, is network selection: choosing which network carries a dual-network debit card.
Acquirer selection and approval rate
Different acquirers hold different relationships with issuers, different fraud profiles, and different network connectivity. The same card can be approved on one acquirer and declined on another, which is why merchants with more than one acquiring relationship can route around weakness.
Vendor-reported uplifts for adopting intelligent routing tend to cluster around two to four percentage points of approval rate, with retry and cascading logic adding more on soft declines. Treat those numbers as a ceiling, not a forecast. They come from merchants who previously had a single, poorly performing acquirer, and the gain shrinks as your baseline improves.
Least-cost routing on debit
In the US, the Durbin Amendment and the Federal Reserve's Regulation II require debit issuers to enable at least two unaffiliated networks per card, and they give merchants the right to choose which network carries the transaction. A regulated debit card might route over Visa or Mastercard, or over a PIN-debit network such as Star, Pulse, NYCE, or Accel.
LCR exploits that choice by sending each eligible debit transaction over whichever enabled network nets the lowest total cost for that specific transaction. Reported merchant savings sit in the range of roughly five to seven basis points on eligible volume, which is real money at scale but small per transaction. The important constraint: LCR only applies to dual-network debit, so it does nothing for credit, and nothing for single-network cards.
One change worth knowing. Regulation II was updated effective July 1, 2023 to make the two-unaffiliated-networks requirement apply to card-not-present debit, including e-commerce, not just in-store. That is why card-not-present LCR became viable for online merchants relatively recently, and why a routing strategy written before mid-2023 is probably leaving debit savings on the table.
Where routing backfires
The failure mode is treating cost and approval as the same optimization. They are not.
The cheapest debit network is not always the one most likely to approve. PIN-debit networks can carry higher decline rates for certain card-not-present flows, weaker tokenization support, and thinner fraud signaling back to the issuer. If your LCR rule sends a transaction to the cheapest network and that network declines it, you saved nothing and lost the sale. A six basis point saving is worthless against a one percentage point drop in approvals on the same cohort.
Routing also fights tokenization. Network tokens (Module 4) are provisioned per network, and routing a card-on-file transaction to a network where the token is not present can force a fallback to the raw PAN, which usually authorizes worse. A routing rule that ignores token presence can quietly undercut the lift you paid to get from tokenization.
Stale rules are the third trap. Issuer behavior, network fees, and acquirer performance drift. A rule that was optimal in Q1 can be wrong by Q3, and a static rule table degrades silently because nothing alerts you when a path stops performing. Routing is not a configure-once feature.
A worked example
Take a US merchant processing 1,000,000 regulated debit transactions a month, average ticket $40, current approval rate 92 percent on the Visa/Mastercard rails.
LCR to PIN-debit saves an estimated six basis points on eligible volume. On $40,000,000 of debit, that is roughly $24,000 a month in processing cost. Worth having.
Now suppose the cheaper network approves at 90.5 percent instead of 92 percent on this cohort, a 1.5 point drop. That is 15,000 transactions a month that flip from approved to declined. At a $40 ticket, even assuming half of those customers retry successfully elsewhere, you have lost roughly 7,500 sales worth about $300,000 in gross merchandise value to capture $24,000 in cost savings.
The math only works the other way if the cost-optimal network approves at parity. So the rule cannot be "route debit to cheapest." It has to be "route debit to cheapest among networks that approve within tolerance, and measure approval per network continuously." That conditional is the entire difference between LCR that helps and LCR that bleeds revenue.
How to run it without losing money
Decide the objective per segment before you write a single rule. High-margin or subscription volume should bias toward approval rate; thin-margin, high-volume retail can bias toward cost. One global routing policy almost always misprices one of those.
Measure approval rate by network and by acquirer, not just in aggregate, so a cheap path that quietly declines more shows up immediately. Gate every cost rule behind an approval-parity check, and keep raw-PAN fallback and token presence in the routing logic so you do not strand the tokenization gains from Module 4.
Re-test on a cadence. A/B a slice of traffic across paths, hold out a control, and let measured approval and cost decide, rather than a vendor's headline uplift number.
Takeaway
Smart routing raises approvals when you have more than one viable acquirer and weak baseline performance to route around. Least-cost routing cuts a few basis points on dual-network debit, and only on dual-network debit. The danger is letting the cheap path win a transaction the expensive path would have approved. Route for cost only inside an approval-parity guardrail, measure per network, and re-test on a schedule, or the savings line and the revenue line will move in opposite directions without anyone noticing.