Your AI data scientist predicts which orders will arrive late, which customers will leave, and which machines will break down.

Built from your history, tested on months it never saw, and run every morning. Your team approves every change.

Scope your first prediction See how it works

Builds the model

You ask the question. It builds the model and shows its work.

Ask the way you’d ask a data scientist. Before anything is saved, it tests the model on months it never saw and shows how far off its dates were next to your current rule’s.

Intelligence/Chat

How many days will each order take to arrive?

No saved prediction covers this yet, so I’ll build one. Should I count from the order to the doorstep, or only the days with the carrier?

Order to the doorstep

Worked for 6 minutes: joined orders, shipments, and carrier scans, then compared 4 models

I learned from January to April, then predicted May and June orders I hadn’t seen. I left out live carrier tracking, since you won’t have it when you promise the date. Its dates came about 2 days closer to the real delivery than your current rule.

How far off each delivery date was

Average days off the real delivery date, early or late, on the same 2.5M May and June orders

Your current ruleNew model

3.2 days

1.3 daysabout 2 days closer

Save this as a prediction?

Predicts days to the doorstep for every new order

Revise detailsSave

It asks before it assumes

One quick question settles what you mean, like order to doorstep.

Tested on orders it never saw

It leaves out anything you won’t know yet when you promise the date.

Nothing saved without you

You see how it did first. Saving it is your call.

Keeps it current

It keeps testing a better version. You decide when it changes.

Every Monday, the model in use and a new version predict the same recent orders. If the new one comes closer, it waits for your approval. Until then, nothing changes.

Predictions/Order arrival

How many days will each order take to arrive?

Checked every Monday for a better version

Model in use

1.1days off on average

Down from 1.3 in July, after 2 approved changes. Your current rule: 3.2.

13 Mondays, a new version tested each time

Model in use10 did worse, kept2 approved1 turned down

Sep 28

Kept the current model

A new version that added each warehouse’s backlog did worse.

Sep 21

Turned down a version that scored better

Your AI data scientist flagged that it used live carrier tracking, which you won’t have when you promise the date. Declined by Ana Ruiz.

Sep 14

New model took over

Predicted the same orders about 2 hours closer, by learning each carrier’s slow days. Approved by Ana Ruiz.

Same test for both

Both predict the same recent orders, so the better score is a fair one.

A person approves every change

A better score only proposes it. Your team makes the call.

Better on paper can still lose

It used tracking you won’t have yet. Your AI data scientist caught it.

Quiet, until a problem
is on its way.

Tell it what matters in your own words. Each morning it messages the people who own the problem, while there’s still time to act.

Alerts in plain words

Type “Tell me when…” and it says back how it read the alert, so you can check it.

Nothing triggered, no message

A quiet morning stays quiet. Your channel only hears about what changed.

Sent where your team works

Slack, Teams, or email, with a link to the full list.

Your alert

“Tell me when a busy lane is about to run 2 or more days past its promise”

How Pedestal reads it

Busy lane: Over 100 orders a day
Late: 2+ days past the promise, on average

Checked every morning · sent 3 times this month

fulfillment-ops

PedestalApp6:04 AM

1,240 orders on 3 busy lanes are likely to arrive late

Reno → Billings520 orders4 days late
Memphis → Spokane410 orders3 days late
Columbus → Boise310 orders2 days late

Promised in 6 days, predicted to take 8 to 10. Most ship economy ground.

Open the listSee why

Puts it to work

Every morning, a short list of what needs your team.

The worst go to the top, each with its reason. Late orders are one question. If your data records how something turned out, it can learn to predict it.

ExampleLate orders, with sample data

Which new orders will arrive later than promised?

In testing, its dates were 1.3 days off on average, against 3.2 for your current rule.

This morning’s 41,280 new orders

5+ days late: 640
1–4 days late: 5,230
On track: 35,410

More examplesSame idea, other questions, with sample data

Which refrigerated containers will fail in the next 14 days?

In testing, it caught 31 of 37 failures at least a week before they happened.

Which orders will turn into a “Where is my order?” call?

In testing, 6 in 10 calls came from the 3% of orders it flagged.

Which warehouses will run past capacity next week?

In testing, it came within 5% of the real volume on 8 weeks it never saw.

“Late” means what your team says it means.

Which scan means delivered. Which customers only receive on Mondays. Your team settles these once, and every prediction uses the same answers.

Your definitions, not column names

It uses what your team agreed, not what a column happens to be called.

Gaps get flagged

A missing or unclear definition comes back as a proposed fix for your team.

Agents can act on it

Agents can act on these predictions, like upgrading shipping, within your limits.

See how Knowledge keeps those definitions, and how Agents act on its predictions.

Bring a decision your team still makes on a rule of thumb.

Tell us the decision and the rule you use today. We’ll scope your AI data scientist’s first assignment with you.

Scope your first prediction

Your AI data scientist predicts which which orders will arrive late, which customers will leave, and which machines will break down.which orders will arrive late.customers will leave.machines will break down.