Routing AI Workflow

Designed an AI-assisted workflow that helps operators build, review, and manage payment routing rules through a visual, agent-supported canvas

Designed a payment routing admin panel that helps operators manage rules,
monitor traffic, and reduce manual errors
in high-volume payment flows.

Project overview

Checkout is a global payment platform processing transactions across dozens of providers, MIDs, and regions. Routing decisions — which acquirer takes a transaction, when to cascade after a decline, how traffic splits across A/B tests — directly affect approval rates and revenue. As routing logic grew more complex, the existing table-based admin panel could no longer keep up.

Problem

Once acquiring moved to the automation layer, operators lost control over routing. They could see outcomes but not the logic behind them — no way to override, no way to A/B a provider, no way to enforce a compliance rule without engineering. The existing rule-editor was a flat table for simple pre-auth conditions, with no view of how
traffic actually flowed.

Goals

Reduce the time it takes to build and update routing rules.

Lower the risk of misconfigured routing logic.

Make routing setups easier to scale, manage across providers

Cut rule creation time by 30%.

Reduce routing configuration errors by 25%.

Reduce time to detect misrouted or underperforming traffic segments

Improve approval rates through smarter provider segmentation.

Constraints

Design System

Built on a new component system, requiring new patterns for canvas, nodes, and connections instead of reusing existing table components.

Routing logic

Conditions, splits, cascade order, and fallback behavior had to stay semantically identical to the existing system.

Financial Risk Tolerance

Payment routing directly affects approval rates and revenue. Any pattern that allowed silent failures was excluded by design.

Key insights

No bulk visibility

Operators could only expand
rules one at a time, never see
the full picture at once.

Operators could only expand rules one at a time, never see the full picture at once.

Suggestions, not decisions

Operators are open to AI suggestions — but only with full control before changes are applied.

Context got lost

Building a rule meant jumping across disconnected steps,
re-entering information each time.

Key desicion

Three interactions were redesigned around the canvas itself — how operators build a routing flow from scratch, resolve errors with AI assistance, and step in manually when needed.

Operators can build a routing flow node by node on the canvas, without leaving context to configure each step separately.

Operators can review an AI-suggested fix, see exactly what it changes, and approve it before it applies.

Operators can trace an error back to its source node and resolve it directly, without waiting on automation.

Final interface

The final design turns routing into a visual workflow — traffic paths, conditions, AI suggestions, and merchant performance live on one canvas instead of scattered across nested tables.

Impact

The canvas turned routing logic into a visible path — operators can follow the flow from condition to provider

Works