Demand Planning Checklist

Extract shipment history at SKU/customer/week grain from NetSuite, Epicor, or whichever ERP is system-of-record. Use shipments rather than booked orders so the baseline reflects actual demand fulfilled. Pull a minimum of 24 months so seasonality models have two cycles to fit on.

Compare the shipment extract to the order book and current backlog. Stockout-suppressed demand and orders pushed by allocation will undercount true demand if you forecast on shipments alone — flag periods where on-hand was zero and lost-sales events are likely.

Strip out one-time spikes the forecast shouldn't repeat: a single large project order, a competitor outage that bled volume to you for a quarter, a recall return. Document each adjustment with the SKU, period, original value, and adjusted value so the cleansing is auditable at the consensus meeting.

Mark promo periods (price-off, end-cap, distributor program) so the statistical engine can decompose lift from base. Tag known seasonality drivers — holiday programs, summer cooling load, ag spring-prep — at the SKU-family level rather than on every SKU individually.

New SKUs, supersessions, and ECN-driven part-number changes break the history trail if not mapped. Pull the item-master change log and BOM rev list; build a like-for-like map so a successor SKU inherits its predecessor's history.

Run the engine (NetSuite Demand Planning, Epicor Smart IP&O, or a Minitab/Excel model) at the SKU-location grain. Most engines auto-fit Croston for intermittent items, Holt-Winters for seasonal, simple exponential smoothing for stable — review the model selected for A items rather than accepting the default.

Compute MAPE and bias on last cycle's forecast vs. actuals at the family and A-SKU level. Persistent bias > 10% in one direction means the model is over- or under-forecasting systematically — root-cause before publishing the next forecast on the same logic.

Pull leading indicators relevant to your end-markets — ISM PMI, housing starts, ag commodity prices, automotive SAAR, distributor POS where available. Note any signal strong enough to warrant overriding the statistical baseline at the family level.

NPI SKUs have no history and the engine will forecast zero or near-zero. Decide if the planner needs to seed a launch curve from a like-item analog or from the NPI plan committed by the product team.

Apply a like-item analog or the product team's committed launch curve to each NPI SKU through month 6 post-launch. Document the analog used so accuracy can be measured once real history accrues.

Send the regional sales managers their baseline at the customer-family grain along with last 6 months actuals. Ask for overrides on customers/programs where they have specific intel — RFQ wins, lost programs, distributor stocking changes — not for blanket percentage adjustments.

Get the next 6 months of promo plans — SKUs included, depth of discount, channel, expected lift. Lift estimates from marketing tend to be optimistic; cross-check against historical lift on comparable promos before loading.

Lock the override deltas at the family level — statistical baseline + sales adjustment + marketing lift = consensus demand. Track the override magnitude as a metric; chronic large overrides mean the statistical model needs work.

Load the consensus demand against rough-cut capacity at each constrained work center. Identify any month where load exceeds available hours on the bottleneck CCR — that's a feasibility issue you'll surface at the consensus meeting, not after the schedule has been published.

Build the gap deck: which families, which months, which work centers, magnitude of the gap, and the candidate levers (overtime, second shift, outsourcing the spillover, demand-shaping by lead-time quote). Bring options, not just the problem.

For raw material with lead times beyond the planning horizon (castings, custom electronics, alloy bar), the buyer needs the consensus number now to place blanket-PO releases. Walk the long-lead BOM list with purchasing line by line.

Recompute safety stock at the SKU-location level using the demand variability and lead-time variability from the last cycle. A items get tighter service-level targets (98%+); C items get lower (90-92%) to keep working capital in line.

Sales, operations, finance, and the GM walk the consensus deck: prior-cycle accuracy, current consensus, capacity feasibility, financial impact vs. plan. The meeting decides on any unresolved overrides and signs off on the number that drives the master schedule.

Publish the signed-off consensus to the ERP/MRP so MPS regeneration runs against the new number. Lock the prior cycle's snapshot for accuracy measurement next month — without the snapshot you can't compute MAPE on what was actually committed.

Send the published forecast to the production scheduler, master scheduler, buyers, and finance. Include the change vs. prior cycle so downstream consumers see what's moved rather than reverse-engineering the delta.