Skip to main content

Stacking Detection in MCA: Why Speed Matters More Than Accuracy

Starter Stack AI2026-02-205 min read
MCARisk ManagementDocument AI

The Stacking Problem Nobody Talks About

Every MCA funder knows stacking is a portfolio killer. When a merchant takes multiple cash advances simultaneously, default rates spike. The industry estimates stacking-related losses at 15–25% of total defaults for high-volume funders, according to data from deBanked's annual industry survey.

But here's what most funders miss: the stacking detection accuracy rate is irrelevant if the detection happens after you've already funded the deal.

The Timing Gap

Consider the typical MCA workflow:

  1. Application received
  2. Bank statements uploaded
  3. Manual review begins (stacking check happens here)
  4. Approval decision
  5. Funding

For a high-volume funder processing 200+ applications per day, that manual review step takes 1–2 days per batch. By the time a stacking flag surfaces, 30–50% of those deals may already be funded.

The math is simple: A 95%-accurate stacking model that runs in 10 minutes prevents more losses than a 99%-accurate model that runs in 48 hours.

What Changes With Real-Time Detection

When we deploy Document Intelligence with stacking detection for MCA funders, the key metric isn't accuracy — it's time-to-flag:

  • Bank statement cross-referencing happens within minutes of upload
  • Historical pattern matching across the funder's entire portfolio runs in parallel
  • Risk scores per merchant are available before the underwriter opens the file

One client went from detecting stacking in 1–2 days to flagging it in 10 minutes across 1,200 weekly applications. The estimated exposure avoided: $1M+ in the first quarter alone.

The Build vs. Buy Decision

Some funders try to build stacking detection in-house using rule-based filters on bank statements. This works at low volume but breaks down when:

  • Merchants use multiple bank accounts
  • Advance amounts are split across transactions
  • Payment patterns overlap with legitimate business expenses

AI-powered detection handles these edge cases because it learns from patterns across thousands of statements, not just the rules you've written.

What to Look For in a Stacking Detection System

If you're evaluating options, prioritize these capabilities:

  • Pre-funding detection — flags must arrive before the funding decision, not after
  • Cross-application matching — the system should compare the current application against every other active application, not just historical data
  • Bank statement parsing at scale — must handle 500+ statements per day without manual intervention
  • Configurable risk thresholds — different funders have different appetite for stacking risk

The Bottom Line

Stacking detection is a speed problem disguised as an accuracy problem. If your detection model is 99% accurate but arrives 48 hours late, it's a reporting tool — not a risk prevention tool. Invest in speed first, then optimize accuracy.