Toys sold 158% above a normal September day during Amazon’s two-day Prime Big Deal Days on 6–7 October, and discounts across U.S. online retail peaked at 20% off list price [1]. During that same week, PepsiCo reported that the price help it funded all year lifted volume in its North American food arm. That move cost the unit 280 basis points of core margin [2], and its drinks arm sold less volume. A standard plan views both events as sales, yet only one acts like true demand. Your 158% spike is a loan from later weeks, and the margin line is the interest.

The big picture:

The October event is now big enough to bend a month. Shoppers spent $9.86 billion in 48 hours, up 8.5% on last year’s $9.09 billion [1]. Adobe reads it as a nudge to stores to run “earlier pre-holiday promotions” [1]. That is a shift in timing, not raw growth. A plan built in monthly buckets has no room for it.

Gartner frames the same failure ahead of next week’s Denver summit. Its note says firms chase better forecasts, only to see their factories overreact when high-level monthly buckets collide with granular weekly master schedules, so a good number can still land as a costly switch [3]. The same note says sloppy use of a forecast creates phantom demand spikes. Those false signals drive costly changeovers, rush runs and stock build-ups [3]. The sales were real, and the demand was borrowed from a week still to come.

PepsiCo’s quarter is the same lever pulled the other way. By the firm’s own account, its price help cut the food unit’s core margin by 280 basis points, partly offset by savings and higher volume [2]. Group core margin was 16.9%, down 35 basis points, with $178 million of tariff refunds worth four points [2]. It then cut its full-year core earnings per share (EPS) growth goal to +1% to +2%, from a floor of +4% to +6% [2].

What strikes me here is that these are two kinds of cost, and the plan has room for one. A promo lends units from nearby weeks and takes them back in the trough. A standing price cut buys volume at a margin cost that no forecast review opens. The firm’s reported plan to raise soft-drink prices, after a quarter in which its North American volume fell 2%, is the same call seen from the other side [4].

By the numbers

  • 158% — Toys above baseline: toy sales during the 6–7 October event against a normal September 2026 day, with electronics up 85% and video games up 90% [1].
  • 20% — Peak discount: the deepest cut Adobe saw, hitting computers, TVs, toys and clothes [1].
  • 280 bps — Food margin cost: the drop in PepsiCo’s North American food core margin, which the firm links to its price help [2].
  • +1% to +2% — EPS outlook: full-year core guidance, cut from a floor of +4% to +6% [2].

What I’d watch:

The question I would want answered before a fourth-quarter plan is set is whether anyone has taken the event out. A two-day window that runs 158% above a September day is not a trend. A plan that resets on it books stock it will have to give back.

  • Where the plan resets: a December forecast built on October’s weekly sales carries the loan into next quarter’s stock.
  • Whether margin is in the room: PepsiCo said North American core margin would “remain under pressure” in the fourth quarter [2] and shut a Knoxville bottling plant, with 112 jobs and pay to 1 January 2027 [4]. Price and plant both trail the volume that spending did not buy.
  • Whether the promo dates become an input: stores moving holiday events earlier [1] shifts the windows, so a model trained on last year’s dates learns the wrong ones.
  • Whether better sensing helps at all: Kroger reports 98% item-level read rates in a bakery test across roughly 1,500 stores [5]. Cleaner shelf data does not tell a discount apart from a price level. That split is a choice about the model, not a sensor.

The cheap test already exists in forecasting practice. Score each layer against a dumb test and keep the ones that earn their place (Forecast Value Added: 52% Lose to Last Period). A promo rate that only works in event weeks should fail, and the fail shows up before the money is spent.

The catch

The honest limit is that these two sets of numbers cannot be subtracted from each other, and neither source shows what would have sold without the discount. Adobe’s lift is a two-day panel reading against a September average, while PepsiCo’s figures cover the twelve weeks to 5 September, a month before the event [1][2]. Nothing here shows that a promo caused the margin problem, or that the price push failed. The firm says the spending lifted volume, and some promo volume is new buyers rather than borrowed ones.

Where I’ve landed is narrower than the headline, and the error is a paperwork one. A promo and a price level are two tools. They arrive in the same field and get read in the same meeting. Neither line shows what the spike gives back or what the level already costs. A plan can close that gap without new software.

At a glance

  • The Big Shift: A two-day October event pushed U.S. online sales in some groups 85% to 158% above a normal day, while PepsiCo’s year-round price help cost its food unit 280 basis points of margin and left its drinks volume smaller. This piece is about the two ways a price move lands in a demand plan.
  • Why It Matters: A forecast that books an event spike as demand commits stock to volume that has to come back out of a later week. One that ignores a standing price cut never sees the margin it is already spending.
  • What I’d watch: Whether the fourth-quarter plan is set off the event weeks, and whether margin enters the same review as volume.
  • Forecast use: how much of a promo’s volume the plan books now rather than borrowing from the next month.
  • Promo dates drift: stores pulling holiday events earlier moves the dates a model learns from.
  • Level versus event: a standing price cut and a two-day discount need two rates, not one.
  • The Catch: The sources cover different periods and measure different things, so the pair shows a planning gap rather than proving a cause. Part of any promo lift is real new demand, and PepsiCo ties its margin drop to several causes at once.

Related reading

Sources

[1] Adobe Analytics’ final Prime Big Deal Days figures, reported by Digital Commerce 360 on 2026-10-08 — https://www.digitalcommerce360.com/2026/10/08/amazon-prime-day-big-deal-days-sales-2026 [2] PepsiCo third-quarter 2026 earnings release and prepared management remarks, 2026-10-08 — https://investors.pepsico.com/docs/pepsico-5v9wci20/media/Files/investors/q3-2026-earnings-release.pdf and https://investor.pepsico.com/docs/pepsico-5v9wci20/media/Files/investors/q3-2026-prepared-management-remarks.pdf [3] Gartner Supply Chain Planning Summit 2026, Denver, 2–3 November, sessions page — https://www.gartner.com/en/conferences/na/supply-chain-planning-us/sessions [4] Supply Chain Dive, “PepsiCo shutters Tennessee bottling plant amid soda slump”, 2026-10-26 — https://www.supplychaindive.com/news/pepsico-shutters-tennessee-bottling-plant-amid-soda-slump/833586 [5] Supply Chain Dive, “Kroger taps RFID, computer vision data to boost inventory management”, 2026-10-27 — https://www.supplychaindive.com/news/kroger-taps-rfid-computer-vision-data-to-boost-inventory-management/872819/