Events, sparse demand, and totals

A promotion plan starts with demand expected without the event and adds an explicit incremental effect. Northstar baseline forty with a twenty-five-percent uplift gives fifty. Keep the ten-unit addition separate. Record dates, price, channel, displays, distribution, and whether the affected customers already appear in the baseline or firm orders. A promoted week of seventy versus normal forty does not prove thirty incremental units caused by promotion. Distribution, weather, holidays, or stock availability may also change. When events always occur together, the data may not identify their separate effects. Describe what the comparison shows without claiming more than the evidence supports. Suppose a promoted region rises from forty to seventy while an untreated region rises from thirty to forty-five. Changes are thirty and fifteen, leaving a difference in changes of fifteen. A causal interpretation requires comparable untreated trends and no other unequal shocks. One before and one after observation cannot establish those assumptions. Promotions can pull demand from later weeks or shift customers between Northstar SKUs. Inspect the event window and follow-up weeks at both product and company levels. A large uplift on one SKU can coexist with a small company increase. More units also need not mean higher margin; commercial evaluation requires its own price and cost inputs. Use conditional cases when evidence is limited. Baseline forty plus uplift assumptions zero, ten, twenty gives forty, fifty, sixty. These are not automatically probability intervals. State what would make the high case plausible, such as verified extra stores. Do not average scenarios into an expected value unless defensible probabilities are supplied. Afterward, preserve the original forecast and record whether the offer actually ran as planned. Stockouts can hide demand during the event. Separate execution differences, measurement limits, and forecasting error. Replacing an original uplift with the observed difference destroys the record needed to learn whether the adjustment process helped. For weekly demand [zero, zero, eight, zero, zero, zero, four, zero], total twelve over eight observed weeks gives one point five per week. Averaging only positive events gives six per event, a different unit. Removing zeros changes exposure and inflates the weekly rate. First verify that zeros are genuine observations rather than missing exports or stockout artifacts. A rate is not an event schedule. Demand of one point five per week does not mean exactly one or two arrive each week. Keep event sizes and gaps visible and inspect totals over replenishment lead time. This course uses transparent rate and event scenarios; a tiny history does not justify an elaborate probability claim. Four leaves A-North, A-South, B-North, B-South can add by product or region. Leaves eighteen, twelve, twelve, nine total fifty one, with product totals thirty and twenty-one and region totals thirty and twenty-one. Use the same time bucket and unit throughout. Adding a monthly revenue forecast to weekly bottle demand is not reconciliation. If an independent company forecast is fifty five while leaves sum to fifty one, choose a documented rule. Bottom-up keeps leaves. Proportional adjustment multiplies leaves by fifty five divided by fifty one and preserves their shares. It creates coherent totals but is not automatically most accurate. Explain why the total or leaf detail deserves priority in this decision. A firm new order assigned to A-North should not automatically be spread across every leaf. Hold justified fixed information and allocate only the unresolved remainder. Structural zeros, such as a product not sold in a region, need special handling. Mathematical convenience is not a reason to invent demand where the product is unavailable. Check performance at both total and operating levels. Opposite regional errors can cancel at company level while creating two service problems. Keep fractional expected forecasts during evaluation. If whole-unit allocation is needed, define a rounding rule that preserves the chosen total, and distinguish that operational allocation from a forecast of individual events. For the lab, draw a box round the full event window. Do not stop at the day of the offer. Show which units moved from a later week and which moved from a second item. Then turn to the sparse series and count all the weeks, not just those with a sale. Finally add each small plan to the top. If the sums do not match, say which view you trust and why. A rule that makes sums match is a useful check. It is not proof that each part is right. Keep that limit clear in the memo you hand to your team.

1. Show a baseline bridge

   Original flat diagram with separate labeled input and output boxes, using Northstar bottle icons.

   • Show a baseline bridge
   • Fifteen is an adjusted comparison; causal interpretation needs additional assumptions.
   • Check units and timing
2. Inspect controls and spillovers

   Original calculation card with aligned quantities and an arrow showing one change.

   • Inspect controls and spillovers
   • Fifteen is an adjusted comparison; causal interpretation needs additional assumptions.
   • Check units and timing
3. Keep uncertainty explicit

   Original comparison chart with clearly labeled cases and a small assumption note.

   • Keep uncertainty explicit
   • Fifteen is an adjusted comparison; causal interpretation needs additional assumptions.
   • Keep assumptions visible
4. Keep zero exposure periods

   Original flat diagram with separate labeled input and output boxes, using Northstar bottle icons.

   • Keep zero exposure periods
   • Preserve zero periods and reconcile product and regional forecasts to one common total.
   • Check units and timing
5. Map the aggregation paths

   Original calculation card with aligned quantities and an arrow showing one change.

   • Map the aggregation paths
   • Preserve zero periods and reconcile product and regional forecasts to one common total.
   • Check units and timing
6. Preserve known information

   Original comparison chart with clearly labeled cases and a small assumption note.

   • Preserve known information
   • Preserve zero periods and reconcile product and regional forecasts to one common total.
   • Keep assumptions visible