What a sponsor receives at the end of a Sprint

The situation

Six projects, five teams, one budget.

A building-materials distributor with 14 branches and one central warehouse. In 18 months, six AI projects started in five teams. Two came from vendors. The board asked for one list: what gets funded this year, and on what evidence.

This page comes from 11 interviews in week one and a half-day workshop with the leadership team in week two.

The decisions

Fund two. Fix two. Stop two.

Every line names who owns it, why, and what happens next.

1. 01

   Supplier invoice matching

   Finance · pilot in two branches

   Fund

   Why. Eight weeks of pilot data against a baseline. Matching time fell from 11 minutes an invoice to 4. A wrong match is caught before payment, so a mistake is cheap and easy to undo.

   Next. Roll out to all 14 branches. A person still approves every payment over $10,000.

2. 02

   Quote drafting for sales reps

   Sales · three versions in three branches

   Fund, as one tool

   Why. Three branches built the same thing three times. Reps who use any version send a quote in a third of the time. The risk is a wrong price, and a rep already reads every quote before it goes out.

   Next. Keep one version and retire two. Count pricing errors per 100 quotes from the first day.

3. 03

   Delivery route planning

   Logistics · vendor proposal

   Fix

   Why. The vendor's 12% fuel saving comes from other companies' routes. Nobody has tested it on this company's deliveries.

   Next. Before anything is signed, the vendor runs its planner on last year's deliveries. The test below decides.

4. 04

   Safety report triage

   Operations · built in-house

   Fix

   Why. The tool sorts near-miss reports so the serious ones get read first. A missed serious report is the most expensive error on this list, and nobody has measured how often the tool misses one.

   Next. The safety team labels 300 past reports. The tool is tested against them, then used as a second reader. It never decides which reports go unread.

5. 05

   Demand forecasting platform

   Supply chain · vendor proposal

   Stop for now

   Why. Nobody knows how accurate today's forecast is. Without that number, no pilot could show the platform is better.

   Next. Measure today's forecast error for one quarter. Come back with that number.

6. 06

   Automated credit limits for trade customers

   Finance · idea, not yet built

   Stop

   Why. The gain is small. A wrong limit damages a customer relationship and is hard to take back. Nobody is assigned to review the decisions.

   Next. Do not pilot. Credit decisions stay with the credit team.

The evidence

What turns a "fix" into a "fund".

Each test is agreed before it runs. The result decides, not the demo.

Route planning

Fund if the vendor's plan covers the same 12 weeks of deliveries in at least 5% fewer miles than the routes actually driven. Stop if it does not.

Safety report triage

Fund as a second reader if the tool misses no more than 2 of the 60 serious reports in the labeled set. Stop if it misses more.

The plan

The next 90 days.

Days 1 to 30

Invoice matching goes to four more branches. Sales picks the quote tool to keep. The safety team labels 300 reports. The route vendor receives last year's delivery data.

Days 31 to 60

The route test and the safety test come back. The leadership team makes the two fund-or-stop calls.

Days 61 to 90

Invoice matching runs in all 14 branches. Every branch uses the same quote tool. The first quarter of forecast error is measured.

Day 90

The sponsor reads this list again, with a result next to each line.

Behind this page, a real readout carries the architecture and evaluation plan for the funded use cases. Every decision comes from the same eight questions: the AI Use-Case Triage Canvas.