AI and Data Situation Assessment

1. The investment

A demand forecasting model covering 1,800 active products across four plants. Approved 14 months ago at $1.2M, including the vendor build, cloud costs and a two-person data team. The business case: planners build the weekly production plan from the forecast, which cuts expedited freight and finished-goods overstock.

2. Where it stands

Built, and live for five months. In back-testing, forecast error sits inside the target the business case set. The weekly production plan at all four plants is still built in the planners’ own spreadsheets. The forecast is opened, then changed. Across the 12 planning weeks reviewed, 71% of forecast lines were changed before the plan was published, and none of the changes carried a reason.

3. What we reviewed

Interviews with the CFO as sponsor, the VP Supply Chain, the four plant planners, two plant managers and the data science lead. Twelve weeks of published plans compared line by line with the forecast. The business case and the steering committee minutes.

4. The actual constraint

Planners override the forecast and nobody owns the override.

There is no rule for when a planner may change the number, no record of why, and no one who reviews the changes. Plant managers are measured on on-time delivery, so a planner who follows the forecast and misses carries the blame, and a planner who overrides and misses does not. The data team owns the model and has no authority over planning. Data quality and model accuracy were checked; neither is what stops the forecast being used.

5. Who holds the decision

The VP Supply Chain. Production planning reports to this role, and the planning rules are set here. The CFO sponsors the investment and does not own planning, so a decision from the CFO alone would not change what planners do on Monday.

6. Options

1. Assign override authority. Keep the model. The VP sets an override rule and reviews overrides weekly. Cost: planner time and a logging change in the planning file. Low risk and reversible.
2. Pause and re-scope. Limit the forecast to the two highest-volume product families at one plant for a quarter, then decide. Slower payback, and the override problem returns when scope widens unless option 1 is adopted with it.
3. Retire. Stop run costs of about $180,000 a year and write off the build. The expediting and overstock the business case targeted stay where they are.

7. Recommended mandate

Option 1, owned by the VP Supply Chain:

• Named owner of the planning forecast: the VP Supply Chain, with one plant planning lead as deputy.
• Override rule: a change of more than 15% on a forecast line needs a reason code and plant manager sign-off, logged in the published plan.
• Measurement: plant managers measured on forecast adherence alongside on-time delivery.
• 60-day operating review: every two weeks, the VP reviews override rate and reasons, the share of each plan built on the forecast, expedited freight and finished-goods days on hand, against the baseline in this brief. At day 60, expand, adjust or retire.

Decision needed: the VP Supply Chain, at the next monthly operating review.