Field Notes: How a Retailer Chose Between Card Networks and Saved 0.4%

Archi4 min read

paymentsretailinfrastructure

Contents
  1. The test that started with a spreadsheet and ended with a contract
  2. Week one: the baseline
  3. Week two: the second rail
  4. Week three: the acceptance cost question
  5. Week four: settlement speed and working capital
  6. Week five: the retry logic decision
  7. Week six: the decision

The test that started with a spreadsheet and ended with a contract#

A regional retailer with 214 stores and about $1.3 billion in annual card volume had a problem that most payment teams recognize. Its processing contract was up for renewal, and the incumbent quoted a blended rate of 2.31 percent plus 11 cents per transaction. The finance team asked for a comparison. The payments team asked for six weeks and a test environment. They got both.

The question was narrow: should the company keep routing all card-not-present traffic through one acquirer, or split traffic across two networks and route dynamically based on cost? The answer involved more variables than the original spreadsheet allowed. Here is what the testing showed, and where the numbers stopped being the point.

Week one: the baseline#

We ran 180,000 live transactions through the existing setup, tagging each by card type, issuer country, ticket size, and channel. The baseline looked clean on paper. Authorization rates sat at 94.1 percent for domestic cards and 88.7 percent for cross-border. Settlement arrived in two business days for 91 percent of volume.

The first surprise came from the long tail. Roughly 6 percent of transactions failed for reasons the processor classified as "do not honor," a bucket that hides everything from insufficient funds to issuer risk rules. When the team pulled issuer response codes, 38 percent of those failures were soft declines that a retry could recover. The incumbent config had no retry logic. That alone represented about $4.2 million in annual volume that never completed.

Week two: the second rail#

The company onboarded a second acquirer and set up a routing layer that sent 15 percent of traffic to the new rail. The goal was simple: compare authorization rates on identical transaction types. The new rail won on cross-border by 2.1 percentage points. It lost on domestic debit by 1.4 points. Neither number was stable enough to act on until week four, when the sample passed 500,000 transactions.

The routing layer itself introduced a cost nobody budgeted. The team spent 90 hours building reconciliation between the two processors. Each rail settled on a different schedule, used different descriptor formats, and reported fees with different granularity. The finance team had to write a matching script that ran nightly. That script now costs about $18,000 a year in engineering time to maintain, which eats into the savings.

Week three: the acceptance cost question#

The real decision point arrived when the acquirers submitted their final pricing. The incumbent offered 2.18 percent plus 10 cents. The challenger offered 1.94 percent plus 8 cents but required a three-year commitment and a minimum monthly volume that the company would hit only in its peak quarter.

On paper, the challenger looked obvious. The blended savings would be about 0.24 percent, or $3.1 million a year. But the challenger's contract included a clause that raised the rate by 15 basis points if monthly volume fell below the minimum for two consecutive months. The retailer's volume swings 22 percent between January and July. In a slow spring, the company could trigger the penalty and erase most of the savings.

The team modeled 36 months of volume using three years of historical data. The penalty triggered in 7 of 36 months under a conservative scenario. Under a moderate scenario, it triggered twice. The expected savings dropped from $3.1 million to about $1.9 million.

Week four: settlement speed and working capital#

The challenger settled in three business days, not two. For a retailer with $1.3 billion in annual card volume, one extra day of settlement ties up roughly $3.6 million in working capital at any given time. At a 6 percent cost of capital, that is about $216,000 a year. Small against the headline savings, but it belongs in the model.

The incumbent, meanwhile, offered to cut settlement to next-day for an extra 5 basis points. That would cost $650,000 a year. The math did not support it. The company kept two-day settlement and put the working capital difference into the comparison as a fixed cost.

Week five: the retry logic decision#

The most valuable finding had nothing to do with which acquirer won. The team built a retry engine that reattempted soft declines after a four-hour delay, with a maximum of two retries. In testing, recovery rates reached 31 percent on the first retry and 9 percent on the second. The engine added about 0.6 percent to authorization rates across both rails. That translated to roughly $7.8 million in recovered annual volume, far more than the pricing difference between the two acquirers.

The catch was operational. Retries increase decline noise for issuers. One card network flagged the company's merchant ID for a review after retry volume spiked in a single day. The team added a daily cap of 2,000 retries per merchant ID and a cooldown period. The review cleared in 48 hours, but it was a reminder that recovery tactics have limits set by the networks, not the acquirer.

Week six: the decision#

The company did not switch fully to the challenger. It split volume 60/40 in favor of the incumbent and kept the routing layer. The blended rate landed at 2.09 percent plus 9 cents, saving about $2.6 million a year before the routing maintenance cost. The retry engine stayed, because it produced more value than either pricing negotiation.

The lesson from six weeks of testing is that acceptance cost is only one line in a longer ledger. Settlement timing, penalty clauses, reconciliation overhead, and network rules all move the number. The team that wins the negotiation is usually the one that built the retry logic first and asked for pricing second.

More articles