Supplier drift, caught before the line stops.

Your supplier is not late yet. They are already inconsistent.

A promised lead time is a contract. An actual lead time is a distribution, and distributions drift. Lead Time watches the spread as well as the mean, per supplier, and tells you which one changed and on what day.

No spam. One email when it is ready to try.

DetectedLead-time variance · supplier=Kestrel Machiningchangepoint 2026-08-09 · mean lead time still normal here
2.4597.43412.410changepointJun 29Aug 27

Sixteen real series, one supplier drifting, detection computed live in your browser.

The problem

Purchase orders are tracked one at a time. Drift is not one at a time.

Every ERP tracks whether a given PO landed on the promised date. None of them tracks whether the shape of a supplier’s delivery distribution has moved, because that question does not belong to any single order. So the signal that a supplier is in trouble sits distributed across two hundred order lines where nobody is looking, and procurement finds out the way procurement always finds out: a line stops, and somebody starts making phone calls.

4 days
Head start in the demo. Lead-time variance on the failing supplier is flagged four days before mean lead time crosses anything.
Detected changepoint indices in this product’s own scenario
16 → 4
Series in, alerts out. Four signals on one supplier, silence on the other three and on the all-supplier on-time rate in the QBR deck.
Computed in your browser from the same scenario
66 vs 4
Alerts a 15%-deviation trailing-mean rule sends across those 16 series, against four from the detector. 33 of the 66 fire before anything is wrong.
Both rules run over identical data in the demo

The insight

The variance moves before the mean does.

This is the whole product. A supplier under strain does not go from eighteen days to twenty-six days on a Tuesday. They lose a shift, or a machine, or a sub-tier vendor, and the first thing that happens is that some orders still land in seventeen days and some now take twenty-four. The average barely moves, so every average-based scorecard says the supplier is fine. What has actually changed is the second moment, and it changed first. Watch only the mean and you have thrown away the early half of the signal, which is also the only half you could still have done something about. So run detection on the dispersion series as a first-class series, not as an error bar on the mean.

Method

Rolling dispersion of realised lead times is computed as its own series per supplier and part family, then CUSUM on day-over-day percentage changes and Bayesian Online Changepoint Detection (Adams & MacKay 2007) run on the dispersion, the mean, defect rate, and on-time rate independently, with Benjamini-Hochberg FDR control across every supplier by measure combination in the run.

How it works

Four steps, no data science team

01
Point it at receipts you already have

PO date, promised date, receipt date, quantity, and inspection result. A nightly export from SAP, Oracle, NetSuite, or a CSV drop. No supplier portal, no supplier cooperation required.

02
It builds a distribution per supplier, not a scorecard

Realised lead time, its rolling dispersion, incoming defect rate, and on-time rate, held per supplier and part family rather than rolled into one monthly percentage.

03
Detection on the spread and the mean, separately

Both go through CUSUM and BOCD as independent series, which is what lets the alert say "the variance moved on the 7th and the mean followed on the 12th" instead of averaging the two into one late signal.

04
A message you can take to the supplier call

Which supplier, which measure, which day, what the distribution looked like before and after, and what safety stock the new distribution implies. Timed so the call happens before the expedite freight quote does.

Who it is for

The procurement manager who gets the call when a line stops

Manufacturers and distributors with fifty to a few thousand active suppliers, an ERP with clean receipt history, and at least one line stoppage in recent memory that traced back to a supplier nobody had flagged. Procurement owns it, supply chain planning uses it.

Pricing

Pilot
$0
Up to 25 suppliers, 12 months of receipts
  • Full detection engine
  • Weekly digest
  • One ERP export
Most common
Standard
$1,200/mo
Up to 500 suppliers
  • Every supplier and part family
  • Variance and mean tracked separately
  • Safety-stock implications per alert
  • Slack, email, and webhook
  • Five years of receipt history
Enterprise
$4,500/mo
Unlimited suppliers, multi-site
  • Self-hosted or VPC deployment
  • Multi-plant and multi-ERP
  • Sub-tier supplier rollups
  • SSO and audit log

Competition

What exists, and what it does not do

WhoWhat they doThe gap
SAP Ariba / Coupa supplier managementSupplier scorecards and performance ratings inside the procurement suite.Scorecards are monthly averages. A monthly average is exactly the statistic that hides a variance change, and by the time it moves the drift is two months old.
Everstream / ResilincExternal supply chain risk monitoring from news, weather, and financial signals.Watches the world for events that might affect a supplier. This watches the supplier’s own delivery data for the event that already did. Both are useful; only one of them notices a lost second shift.
Built-in ERP OTIF reportingOn-time-in-full percentages straight out of SAP, Oracle, or NetSuite.A binary per order rolled into a percentage per month. It has no concept of a distribution, so it cannot report a spread that doubled while the percentage held.
A planner with a pivot tableWhat most procurement teams actually run.Genuinely good at the supplier they are already suspicious about. Cannot review five hundred suppliers every morning, which is the only cadence at which drift gets caught early.
How this fails

The honest failure mode: receipt data is dirtier than the pitch assumes. If receipt dates are stamped when paperwork is processed rather than when the truck arrived, the dispersion series is measuring the receiving dock’s Monday backlog and not the supplier at all. That is not a modelling problem, it is a data problem inside the customer, and it can only be found after the sale. The second failure mode is organisational: knowing a supplier is drifting is worth nothing if the company is single-sourced on that part and has no alternative to switch to, which is the situation for most of the parts that matter.

Market

Priced against one avoided line stoppage, not against software budgets

A single unplanned line stoppage at a mid-size manufacturer costs more than a year of the Standard tier. There are roughly fifty thousand mid-market manufacturers and distributors in North America and Europe with enough supplier count for this to matter. Three thousand at the Standard tier is $43M ARR.

Get early access

No spam. One email when it is ready to try.

Or just go look at the demo first →