The PXI Platform

One agent runs the entire loop.

Numr PXI finds where your revenue journeys stall, predicts why each customer stopped, sends the message that moves them forward, and proves the recovery against a holdout. Syna Drive runs the loop. You approve every step.

See the results See how it's built See the agent at work

Watch the 2-minute demo

The Results

Start with what it recovered.

Every PXI deployment rolls up to one view: how many customers entered the journey, how many stalled, how many were moved, and what that was worth. Measured against a holdout, not modeled.

Journey volume

Customers who entered the journey.

Dropped

Customers who stalled before completing.

% nudged

Stalled customers PXI reached.

Incremental revenue

Recovery proven against a holdout.

app.numr.ai/journeys

Numr PXI journey dashboard: the Impact Across Journey Sankey with drop-off points and recovered revenue, framed by journey volume, lift, and incremental revenue metrics

Illustrative data

One journey, every drop-off point, and the recovered revenue at each. This is the screen your CFO will ask for.

The Sankey shows the whole flow, entry to completion.

The left rail prices each drop-off per product line.

Every dollar here survived a control group.

How the number gets made

A number like that has machinery behind it.

The next four sections open it up on the real product, stage by stage. Then the record of who ran it, and who approved it.

01 Journeys & Pods

Every drop-off has an address.

PXI maps your journey as a flow of customers and shows exactly where they stop, what each stall costs, and what recovering it is worth.

The unit of work is the Pod: a group of customers stalled at the same point, with a precise definition, priced with a value per customer and a share of the journey.

For example:

Stall: started identity verification, still not finished

Success: verified before your cutoff

Value: $180 / customer

A vague activation problem becomes a priced, bounded population you can act on. Every figure shown is illustrative.

One stall point, defined, priced, and tracked. (Illustrative data.)

Numr PXI Not Verified Pod detail: pod definition, journey volume / lift / revenue, and the impact-across-journey Sankey

02 Discovery & Modeling

Ask the stalled. Predict the rest.

PXI finds out why customers stall by asking them, then trains a model so it knows why the next customer stalled without asking.

1 Discovery

Discovery listens everywhere you already do, from Numr CXM surveys to live Syna Drive conversations, then sorts what customers say into themes you approve.

Numr PXI Theme Discovery: the domain and theme taxonomy with approval controls, and the customer-voice data sources rail

Then

2 Modeling

Approved themes go to modeling, which predicts which theme every future customer will hit, using signals you already have.

Numr PXI Theme Prediction Modeling: draft versus published version comparison, F1 metrics, and per-theme results

03 Experimentation

Personal messages, built from parts you approved.

Every customer gets a message matched to their predicted reason for stalling, and every message is tested against a control group before it earns a wider audience.

Numr PXI experiment EXP_1012 detail: allocation and results from test, treatment versus control with lift and confidence interval

app.numr.ai/interventions

Numr PXI Message Wizard: the approved five-part message skeleton, the composed email, and the customer context panel

04 Production

Where expected lift meets observed lift.

In production, every live nudge is watched: expected lift versus observed lift, incremental dollars, and an automatic pause the moment performance drops.

Numr PXI customer drill-down: Quick View with predicted stall reason and model confidence, and the full message log by channel, template, version and delivery status

app.numr.ai/production

Numr PXI Production Monitor: live nudges with expected versus observed lift, exposures, and active or paused status

The agent, on the record

The top of this page made a claim. Here is the record.

One agent running the loop, one human approving every step. Every action is logged, and at every gate the decision is yours.

Between approvals, Syna Drive works alone: it surfaces new stall themes, drafts the next model version, runs the experiment, and pauses whatever degrades. Nothing reaches a customer, and no model reaches production, without a named human decision on the record.

Audit Trail

TimestampActionActor

2026-03-10 09:45 Approved A. Rao

2026-03-10 10:02 Trained System

2026-03-10 10:31 Started System

2026-03-14 14:02 Closed System

2026-03-14 14:15 Promoted A. Rao

2026-03-15 10:42 Auto-Paused System

Illustrative data

System does the work. At every gate, a name: yours.

The agent carries the work. You keep the judgment.

See the full control panel and audit trail in the 2-minute demo.

Trust & Deployment

Built to clear your risk review.

Compliance, risk, and IT will have questions. These are the answers.

Human approval, always.

No message sends, no model ships, without a named person's sign-off. That's a feature, not a speed bump.

Everything on the record.

Full audit trail with before and after states.

Model governance built in.

Versioned models, published validation metrics, every version retested against control, every change on the audit trail.

Pilot without integration.

One weekly customer file. That's the whole ask.

Production without rip-and-replace.

PXI connects into the systems you already run: Salesforce, Braze, Qualtrics, Agentforce, ServiceNow. Our team builds each integration, so your stack stays where it is.

Integrations

Built to fit the stack you already run.

PXI connects into your core and digital banking systems by API. No rip and replace, and one less thing for IT and risk to sign off on.

Core banking

Digital banking

A representative set, not a fixed list. If your core or digital banking platform exposes an API, PXI integrates with it.

Start here

You've seen the machinery. Point it at one stall.

The pilot is deliberately small: one stall point in one journey, one weekly customer file, one measured revenue number in 90 days, proven against a holdout. If the number is real, you'll know what scaling it is worth. If it isn't, you'll have spent one file a week finding out.