Data Readiness for AI

Data Readiness for AI

Step 1 of 9

ready0 of 4

needs → what exists

Order historyneed 24 mo

Carrier scan eventsneed all lanes

Scan timestamps filledneed ≥ 95%

Deliveries labelled late / on timeneed 5,000

readiness

historyscansqualitylabels

????

Use case: predict a day ahead which deliveries will be late. Before any model, list what it needs from data.

The Idea

Before choosing a model, list what the use case needs from data: which records, how far back, how complete, labelled with what. Then check each need against what exists. The usual gaps are access (a silo with no owner or agreement), quality (missing or inconsistent fields) and labels (nobody recorded the outcome). A data strategy closes them with named owners, sharing rules and quality fixed at the source.

Real-World Example

A restaurant planning a new dish checks the pantry before the menu. The spices are there, the fish is at the other branch, the oven thermometer is broken, and nobody wrote down which plates came back uneaten.

The Tradeoff

Data work is slow and unglamorous, so teams skip it and blame the model later. Budget it first; for many use cases it is most of the effort.

Your turn

Put the steps in the right order.

1. 1Check each need against the data that exists↑↓
2. 2Write down what the use case needs from data↑↓
3. 3Fix the gaps at the source, with named owners↑↓
4. 4Sort each gap into access, quality or labels↑↓

Check order

Mini quiz

1 / 3

The data a use case needs exists, but in another division with no sharing agreement. It is:

• AReady, because the data exists
• BA gap until an owner agrees access and terms
• CIrrelevant to the model
• DSolved by choosing a larger model