A consumer lender is building a model to predict which new personal loans will default in the first 12 months. The team trained it on four years of loan history. On the held-out test set it catches 93% of defaults with very few false alarms. The scorecard the bank uses today catches 61%. During the review before the go/no-go meeting, a credit analyst points out that the strongest input in the model is a field called "hardship plan offered". Collections staff fill in that field after a customer has already missed two payments. The sponsor says the model beats the success criteria in the scope statement by a wide margin and wants it live in the loan approval flow next month. What should the project professional do? A. Approve the go/no-go, since the model clearly exceeds the agreed success criteria, and add the hardship field to the production monitoring plan. B. Hold the go/no-go, have the team remove every input that is not known at the moment of the loan decision, and re-evaluate on held-out data. C. Ask the data scientists to tune the model so it relies less on the hardship field, then rerun the test set and compare the new score. D. Deploy the model as a limited pilot in one region and compare its predictions with real defaults before scaling it across the business. Reply with the letter and one line on why. A one-letter answer is fine too. I will post the answer and the ECO task behind it in the comments on Monday.

A telecom operator's finance team wants an AI model to flag invoices that break the company payment policy. The policy fits on two pages. Invoices above 10,000 euros need two approvals. No payment goes to a vendor that is not on the approved list. An invoice number that appears twice is rejected. The sponsor has the budget approved and a vendor demo booked for next week. The data scientist says a model can reach 97% accuracy on last year's invoices. Two questions. Q1. You are at the Phase I gate. Go or No-Go? One sentence of reasoning, not just the word. Q2. Whatever you decided, what is the one question the team should have answered before the vendor demo was booked? And if the answer is No-Go, does that send the project back to an earlier phase, or does it end the project here? Say why. Answer in the comments with one sentence of reasoning. That sentence matters more than the verdict. I reply to every answer, so try it before you read the other replies. Exam on the calendar? Tell us the date as well.