Ramp's share of card transactions synced to the ERP with no coding edits fell from 82% to 68% for mid-market customers. Why?

Introduction

One of Ramp's core promises to finance teams is a faster month-end close. A large part of that comes from accounting automation: Ramp suggests or applies GL accounts, departments, classes, and vendor mappings to card transactions, reimbursements, and bills, then syncs them to the customer's ERP (such as NetSuite, QuickBooks, Sage Intacct, or Xero). In this scenario, the share of card transactions that sync to the ERP with Ramp's suggested coding and no human edits has dropped from 82% to 68% over four weeks among mid-market customers. Controllers are spending more time fixing coding, and customer success has noticed close-related complaints.

This is a hard root cause question because the metric depends on machine learning predictions, customer-configured accounting structures, ERP integrations, and human behavior all at once.

Step 1

Clarifying Questions (3 minutes)

• Looking at the metric definition, is "no human edits" measured at sync time or at period close? If at close, edits made during review late in the month could be lagging. I'll assume it is measured at sync.

• Considering the scope, is the drop limited to mid-market customers, or do startups and enterprise show it too? I'll assume enterprise is flat, startups are slightly down, and mid-market is sharply down.

• Thinking about recent changes, have we shipped any updates to the coding model, the suggestion UI, or ERP connectors in the past six weeks? I'll ask for the full change log, including model retraining dates.

• Considering customer behavior, did a large cohort of mid-market customers onboard recently or change ERP setups, such as at fiscal year start? I'll check onboarding cohorts and chart-of-accounts changes.