Smart-routing: channel prioritization

Smart-routing: channel prioritization

1) Why Smart-routing

Conversion growth: + 5-15 pp to Auth Rate due to the choice of the "correct" PSP/method for the bank, BIN, device and risk.
Cost reduction: − 10-30 bps to effective rate when taking into account commissions, FX, 3DS, chargeback expenses.
Stability: traffic auto-crossing in case of PSP/ACS/bank failures; management of degradation without downtime.
Compliance/RG: built-in filters of sanctions, geo, age/self-exclusion, velocity-limits and KYT.

2) Signals and features for prioritization

Payment and banking

BIN/IIN, issuing bank, debit/credit/prepaid; local/cross-border.
Historical metrics per bank/method: Auth Rate, Soft Decline share, p95 latency, fault tolerance.

Payer profile

Geo/IP/ASN, device/OS/browser/emulation, behavioral patterns, anti-fraud risk rate, RG status.

Transaction parameters

Amount/currency, FX, re-payment (MIT/COF), card/wallet novelty, freshness of 3DS session.

Operating rooms

Commissions (MDR + scheme + 3DS + PSP fee), reserve/hold, settlement speed, channel availability (health).

3) Scoring model of prioritization

• Score(route) = w₁·P(approve) − w₂·Fee − w₃·LatencyPenalty + w₄·Health − w₅·RiskPenalty

• P (approve) - probability of approval by bank/method history for similar profiles.
• Fee - full effective cost/transaction (including 3DS/FX).
• LatencyPenalty - penalty for p95> threshold.
• Health - integral indicator of availability/errors in the last N minutes.
• RiskPenalty - high risk/velocity/anomaly surcharge.

Notes

wᵢ weights are calibrated by A/B tests and by region.
Use "cooling" periods and smoothing so as not to "jump" between routes.

4) Decision flow

1. Pre-check compliance: sanctions/PEP, age, self-exclusion, geo-blocks → instant refusal if it does not pass.
2. Risk scoring: device + behavior + payment history → choice of 3DS/SCA policy.
3. Route candidates: PSP-A/PSP-B/local APM/open banking/alternative wallet.
4. Ranking by Score (route); attempt # 1.
5. Smart Retry with soft-decline: 3DS mode change (challenge↔frictionless), MID, time backoff; attempt # 2/# 3.
6. Fallback on APM/open banking for behavioral risk or network failures.
7. Logging: Route Journal (PSP, reason, latency, 3DS, fee) for training/reporting.

5) 3DS/SCA dynamics

Frictionless by default for low risk/small checks and heated COF/MIT.
Step-up challenge for high risk/BIN groups/sum anomalies.
Fallback strategies: degradation of ACS → increase the challenge rate or switch to open banking/APM; for mass timeouts, use circuit-breaker.

6) Channels and prioritization heuristics (by region)

EC/UK: cards + open banking (SEPA Instant/FPS). On the maps - strict 3DS orchestration, TRA/whitelisting, high weight issuer historians.
USA: cards + ACH/RTP, instant payouts priority (OCT/RTP). Keep an eye on the chargeback tooling at the PSP.
LATHAM: Pix (BR), SPEI (MX), PSE (CO), vouchers. Preference for APM/open banking for new customers and high-risk BIN.
Turkey/CA/MEA: local transfers/wallets, tougher sanctions/TAC, velocity limits.
Asia/India: UPI/e-wallets, routing by issuing banks and time of day; thin cutting of limits.

7) Payout-routing

Priorities: speed (instant/near-instant) ↔ cost ↔ availability of the channel.
KYT/AML/RG output: retentions, velocity, source of funds, bundles of devices/cards, queues during bank degradation.
Failover: queued payouts, moving to alternate tracks (RTP/FPS/SEPA Instant/Pix).

8) Strategy pseudocode

on PaymentRequest(PI):
if!Compliance.pass(PI): return REJECT risk = Risk.score(PI)
policy3ds = ThreeDS.select(risk, PI)
candidates = Catalog.filter(PI.geo, PI.method, PI.bin, health_ok=True)

ranked = rank_by_score(candidates, PI, risk, policy3ds)
for route in ranked:
res = PSP.call(route, PI, policy3ds, idempotency_key=PI.key)
log(RouteJournal, route, res)
if res.approved: return APPROVE if is_soft_decline(res): policy3ds = tweak_3ds(policy3ds); continue if is_retryable(res): backoff(); continue return DECLINE

9) KPIs and objectives

Auth Rate (cards): EU 85-92 %/US 80-88 %/LATAM 70-85% (after tuning - upper corridor).
Routing Efficiency (conversion ÷ cost): + 5-10% to baseline in 1-2 quarters.
p95 latency (auth): < 3 с; webhooks: < 60 с.
Circuit-break RTO: <2 min; RPO: 0 (due to idempotency).
Share of Instant Payouts: ≥ 70% of "light" checks.
Chargeback rate: < 0. 5% by count (depends on product/GEO).

10) A/B design and training

Split by GEO/BIN/method; minimum 10k transactions/branch, 2-4 weeks.
Метрики: Auth Rate, challenge rate, post-3DS approvals, average fee, latency, decline mix, payout time.
Guardrails: no retray control groups; anti-drift (random "noise" in route selection for unbiased evaluation).

11) Degradation playbooks

Issuer spike (reason 05/51/62): local circuit-breaker on BIN/bank → reroute on alt-PSP/APM; raise challenge rate.
ACS timeouts/soft declines: include "challenge-all" for affected banks, increase open banking share.
PSP-outage: global breaker, transfer to backup MIDs; half-open canary for 1-5% of recovery traffic.
Payment delays: prioritization of VIP/small amounts, queued payouts, alternative rails.

12) Data, log and observability

Route Journal: PSP/MID, BIN/issuer, reason, latency, fee, 3DS mode, retry chain, итог.
Dashboards: Auth Rate (by banks/methods), decline mix, p95 latency, 3DS success, payout SLA, health PSP.
Alerts: thresholds on spikes/latency, signal from external status pages, automatic breaker.

13) Implementation checklists

Architecture/Data

• Single Payment Intent and strict idempotence on all hops.
• Route directory with weights and SLA thresholds health-pings.
• Signed webhooks, raw reason codes, Route Journal storage.

Rules/Safety

• Compliance filters (sanctions/age/self-exclusion) prior to routing.
• 3DS Policy: TRA/whitelisting/step-up; ACS degradation fallback.
• Circuit-breakers local/global; canary-return.

Operations/KPIs

• A/B frame, guardrails, regular scale recalibration.
• Auth Rate Targets, latency, costs, payouts; QBR с PSP.
• Runbooks incidents and RACI escalations.

14) Economy: How to count the "cheap" route

Effective rate = (MDR + scheme + 3DS + PSP fees + FX + chargeback cost − rebates) / Approved Volume.
Add a risk premium: high-risk channels make chargeback and reserves/holdback more expensive.

15) The bottom line

Smart-routing is data + rules + stability. Use a complete picture of signals (bank/BIN, risk, cost, channel health), dynamic 3DS policy and clear degradation playbooks. Regular A/B calibration and transparent KPIs turn routing from "black magic" into a controllable lever of P&L, RG and compliance.