ORIA Lab · Weak signals

See disengagement before it turns into a loss.

300 fictional customers, 18 months of history. The detector combines trend, change detection and peer comparison: tune its sensitivity and compare it with a classic threshold rule.

100% local, nothing is sent ≈ 0.15 MB downloaded on first run

How it works

  1. Trend

    The slope of activity over recent months, computed by least squares and expressed in % per month.

  2. Change detection

    A cumulative-sum test (CUSUM) spots a lasting slide below the customer’s usual level, even a slow one.

  3. Peer comparison

    A robust score (median, median absolute deviation) positions the customer’s trajectory against its segment.

  4. Score and threshold

    These indicators, with payment delays and complaints, form a risk score; the alert threshold is the slider you set.

What it means for your business

In production, signals are built with your teams from your own data (transactions, usage, delays, complaints), thresholds are calibrated on your past cases, and every alert is explained and checked against real cases before rollout. Models can be richer; the decision stays human.

Weak-signal detection & decision support
The data is synthetic, generated for this demo, and every calculation runs in your browser. Nothing is sent.

Frequently asked questions

Are the customers and figures real?
No. The 300 customers and their histories are generated by a script, with disengagement patterns injected on purpose. The results shown apply to this data, not to your business.
What is a weak signal?
A slow or subtle change — gradually declining activity, lengthening delays, repeated complaints — that crosses no classic alert threshold but often precedes an exit or an incident.
How do you choose the right sensitivity?
It is a trade-off between early alerts and false alarms, which depends on what an intervention costs your teams. In a pilot, we calibrate it with your business teams on real cases.

Other demos

100% local · free

Document → data

A scanned invoice becomes structured, cross-checked data in seconds, without leaving your browser.

Launch the demo

100% local · free

Questions on documents

Ask a question about internal procedures: the answer always cites its source document, or is not given at all.

Launch the demo

Next step

45 minutes to identify your first win.

A structured conversation with an AI architect — no commitment, no jargon.

Book my free assessment