Skip to content

Nodal Router

Take your people out of the reviews that don’t need them.

A task router for high-consequence workflows where a person approves every AI decision. It learns from your own review history which cases are safe to approve automatically, and sends the rest to your team with the evidence attached.

See it on your review queue (20 minutes)

Nothing changes until you’ve watched it work on your own decisions.

Your queue, with room to move Concept diagram
AI decisions
Nodal
Router
✓Approved automatically,
with evidence
Only categories you switch on
Needs reviewYour team decides

New, unusual or below the bar? Always to your team.

The bottleneck

AI got faster.
Your review queue didn’t.

Your AI drafts the message or recommends the action in seconds. Then it waits for a person. That review exists for a reason: a wrong answer costs you a customer, a patient or a compliance finding. So the queue grows, your best people spend their day approving things that were already right, and the return on your AI stalls.

  • Reviewers approve most items unchanged.
  • Hiring more reviewers is the only way to scale.
  • Nobody can say which cases are safe to stop checking.

A controlled rollout

Watch first. Prove it.
Then let the routine work move.

Router watches

Watch.

The router runs beside your team in shadow mode. It approves nothing. You see what it would have done on every case.

Shadow mode · your team decides

Evidence builds

Prove.

For each type of case it shows how often it matched your reviewers, and what happened afterwards: did the customer stay, did the payment clear, was the visit kept.

First category · only after your approval

Your team stays

Switch on, one category at a time.

You choose which categories go automatic. Everything new, unusual or below the bar still goes to a person.

More categories · the same control

Conceptual rollout. Teal marks categories you have approved for automatic routing.

The reason behind each decision

Evidence, not a second opinion from another AI.

It does not use an LLM to judge an LLM.

Checks are deterministic and built from your own past decisions, so the same case gets the same answer every time.

It tests against outcomes, not only against what a reviewer clicked.

A rubber-stamped approval is weak evidence. A good result is strong evidence.

Every automatic approval has an audit trail.

You can show a regulator, a client or your own board why a case did not need a person.

Example evidence Illustrative · not customer data
✓ Approved automatically

Appointment reminder

Matched 412 similar past cases

Checks passed
✓ Approved message template
✓ Appointment details match
✓ No new or unusual request
Outcome of past cases
Visits kept after the reminder
Routing basis
Category enabled by your team

Example only. Case counts and outcomes here illustrate the audit trail; they are not performance claims.

Is this for you?

Built for queues where getting it wrong matters.

A good fit

  • A person approves AI output case by case
  • A wrong answer is costly
  • Hundreds of cases a week or more
  • You record what was suggested and what was sent

Not a fit

  • Review is required by law for that step
  • Mistakes are cheap, so you already let the AI run
  • A handful of approvals a week

For your technical team

Fits where your work happens.

Runs in your environment.

Your data stays with you.

Learning happens through context and checks, with no model retraining.

Start with your own decisions

See what your queue
would look like.

Send us a sample of past decisions. We will show you which categories your reviewers already approve unchanged, and what it would take to prove they are safe.

Book 20 minutes