Latest / MBA Training AI / Price elasticity in FMCG: weekly SKU-level models

Transcript

Price elasticity in FMCG: weekly SKU-level models · MBA Training AI

• 0:00
  Host: Welcome to the Leaders Insights podcast.
• 0:07
  Host: Today's episode, Building AI Elasticity Models for FMCG Assortment and Price Optimization.
• 0:13
  Host: Let's start with the thing everyone in consumer goods is whispering about this week.
• 0:18
  Host: Price elasticity models have existed since the 1980s.
• 0:21
  Host: Why is this suddenly a hot story in September 2026?
• 0:25
  Expert: Because the old models were built to answer one question a quarter.
• 0:28
  Expert: And the business now needs answers on thousands of products every week.
• 0:33
  Expert: Elasticity.
• 0:34
  Expert: How much sales move when you change the price?
• 0:36
  Expert: Was always calculated at a level so broad it was basically useless.
• 0:41
  Expert: You'd get detergent is price sensitive and then have to guess which pack size, which
• 0:46
  Expert: retailer, which region.
• 0:48
  Host: But the math hasn't changed.
• 0:49
  Host: A price coefficient is a price coefficient?
• 0:52
  Host: What's actually new?
• 0:54
  Expert: The math hasn't.
• 0:55
  Expert: The scale has.
• 0:56
  Expert: The old regressions choke on sparse data.
• 0:59
  Expert: Most individual products barely sell enough units in a given store week to estimate anything
• 1:04
  Expert: reliable.
• 1:05
  Expert: What's new is pooling.
• 1:07
  Expert: You borrow signal across similar products and stores.
• 1:10
  Expert: So a slow moving SQ.
• 1:12
  Expert: Stock keeping unit, one specific product and size, C, inherits patterns from its neighbors.
• 1:18
  Expert: That's the difference between an estimate you'd stake a negotiation on and noise.
• 1:23
  Host: The article anchors this on a specific number.
• 1:27
  Expert: The playbook sites teams moving from elasticity refreshes every 13 weeks down to weekly across
• 1:34
  Expert: roughly 40,000 SQ's per category.
• 1:37
  Expert: That cadence is the whole point.
• 1:39
  Expert: A quarterly number is a historical artifact by the time you use it.
• 1:43
  Expert: A weekly one survives contact with an actual trade negotiation.
• 1:47
  Host: You mentioned vendors are loud about this.
• 1:49
  Host: Whose figures should I distrust?
• 1:51
  Expert: Distrust is strong, but scrutinize.
• 1:53
  Expert: Yes.
• 1:54
  Expert: OpenAI and Google AI have both published throughput claims about running these models cheaply at
• 2:00
  Expert: scale.
• 2:01
  Expert: Very impressive per inference costs.
• 2:04
  Expert: Worth remembering they sell the compute those models run on, so the economics are conveniently
• 2:08
  Expert: rosy.
• 2:09
  Expert: Cross check against O'Reilly radar's engineering write-ups, which put real production costs
• 2:14
  Expert: higher once you include the data plumbing nobody demos.
• 2:18
  Host: The data plumbing.
• 2:19
  Host: That's the part that always gets skipped in the sales deck.
• 2:22
  Expert: It's 80% of the job.
• 2:24
  Expert: Your point of sale data, the scanner records from checkout, arrives late in different formats
• 2:29
  Expert: per retailer with promotions logged in consistently.
• 2:33
  Expert: If you can't cleanly separate a price cut from a display promotion sitting on top of
• 2:37
  Expert: it, your model attributes the sales lift to the wrong lever.
• 2:42
  Expert: Garbage estimate, confident decimal places.
• 2:45
  Host: Give me a concrete failure.
• 2:47
  Host: Where does this go wrong in the real world?
• 2:49
  Expert: Classic one.
• 2:50
  Expert: A brand sees demand jump when they discount.
• 2:53
  Expert: Model says very elastic, cut price permanently.
• 2:57
  Expert: But the jump was pantry loading.
• 2:59
  Expert: Shoppers stockpiling because it's cheap, then buying nothing for six weeks.
• 3:03
  Expert: The model saw a spike, missed the crater.
• 3:06
  Expert: MIT Technology Review flagged exactly this pattern earlier in 2026.
• 3:12
  Expert: Models that don't account for forward buying overstate elasticity and quietly destroy
• 3:17
  Expert: margin.
• 3:18
  Host: So how do you catch that?
• 3:19
  Expert: You model the post-promotion dip explicitly, and you validate against holdout stores.
• 3:25
  Expert: Locations where you deliberately don't run the price change, so you have a real comparison.
• 3:30
  Expert: If your model can't predict what happened in stores, it never saw.
• 3:33
  Expert: It's memorizing, not learning.
• 3:36
  Host: Let's talk power.
• 3:37
  Host: This connects to trade negotiation.
• 3:39
  Host: How does a better elasticity number change what happens in the room with a retailer?
• 3:43
  Expert: Enormously, because retailers have had the better data for years.
• 3:47
  Expert: The retailer knows your true elasticity from their own scanners.
• 3:51
  Expert: When a brand walks in with a course quarterly estimate, they're negotiating half-blind.
• 3:57
  Expert: A weekly, skew-level model means you can say, this promotion depth loses us both money.
• 4:03
  Expert: Here's the depth that grows the category.
• 4:06
  Expert: And back it with the same granularity they have.
• 4:08
  Host: Isn't there a risk you over-trust the model and stop using judgment?
• 4:12
  Expert: Yes.
• 4:13
  Expert: Discipline teams treat outputs as a decision input, not a verdict.
• 4:18
  Expert: The model tells you elasticity with a confidence range.
• 4:21
  Expert: If that range is wide, that's the model admitting it doesn't know.
• 4:24
  Expert: And that's when a human should override.
• 4:27
  Host: Give me the one thing a listener can do Monday morning.
• 4:30
  Expert: Pull your last four promotions and check whether anyone measured the post-promotion dip.
• 4:35
  Expert: If nobody has that number, every elasticity figure in your building is inflated, and
• 4:40
  Expert: you're leaving margin on the table before you touch a single model.