Newsletter · · Ashutosh Agarwal

Instacart's Cart Becomes Grocery's New Ad Machine - Food: Brands, Private Label & Grocery - Week of August 27, 2026

Food: Brands, Private Label & Grocery for the week of August 27, 2026. Podcast synthesis on Instacart's nearly 1 billion dollars of ad and other revenue, a McCormick campaign that reached a 93 percent new audience at 20 percent cheaper CPMs, the Caper smart cart tripling year over year on NVIDIA Jetson silicon, and an analyst calling return on ad spend a costly misnomer.

Food: Brands, Private Label & Grocery

Week of August 27, 2026: Instacart's Cart Becomes Grocery's New Ad Machine

Most weeks, this letter is about what's happening to your grocery bill: the cocoa spike hiding in a Hershey bar, the private-label box quietly replacing a name brand, the weight-loss drug rearranging the middle of the store.

This week, the loudest voices in the podcasts weren't talking about the food at all. They were talking about the cart, the physical shopping cart, with a screen bolted to the handle and a chip from NVIDIA doing the thinking, and the fast-growing advertising business now running through both the cart and the app. The two conversations that landed were unusually rich, and they're about the part of grocery that actually makes money right now.

TL;DR

  • Grocery's real profit engine spoke, and it's ads. Instacart says it did nearly $1 billion in "ads and other revenue" last year, and its executives spent the week explaining exactly how that machine works, both on the open web and, increasingly, inside the store on a smart cart. (operators)
  • The "picks and shovels" pitch, with receipts. A McCormick spice campaign reached an audience that was 93% new (only 7% overlap with the brand's own data) and cut ad costs 20%+. A Hershey Reese's shoppable-TV ad over March Madness turned ~1 in 3 buyers into new-to-brand customers. (operator + analyst)
  • The smart cart is scaling. Instacart's Caper cart (a $350M acquisition) is now in 100+ cities, thousands of carts, tripling year over year, feeding on millions of sensor inputs a day. A single "did you forget the yogurt?" nudge at checkout lifted sales ~0.9%. (operator)
  • The debate was whether retail-media measurement is real. One analyst made the case that the industry's favorite metric (ROAS) is "a costly misnomer," and that AI shopping is evolution, not revolution for grocery ad networks. (analyst)
  • Walmart's ugly quarter was the macro backdrop, its slowest U.S. comparable-sales growth in six years.

What's new

1. Instacart makes the "grocery data is different" case, and shows the numbers behind it

The headline this week came from Cannes, of all places. On The CPG Guys, "Instacart's Adam Silverblatt & Media Ads & Commerce's Andrew Lipsman - Live from Cannes Lions 2026" (Aug 22), two people who think about this for a living laid out why the advertising attached to your grocery order is worth so much, and, unusually, backed it with named campaigns and real percentages.

The two voices are worth separating, because one sells the product and one doesn't:

  • Adam Silverblatt runs Instacart's off-platform ad sales. He's an operator, talking his own book.
  • Andrew Lipsman is an independent analyst (he coined the phrase "retail media"). He's a pundit, not selling anything, which makes his framing the more useful check.

Why grocery data is special (Silverblatt, operator). Groceries are roughly a $1.3 trillion slice of the U.S. economy, and, his words, "you have to eat to survive." That makes grocery buying "habituated ... frequent," and, critically, tied to a real purchase by a real person at a real delivery address. He contrasted that with the "messy historical loyalty card signals" the industry leaned on for decades, data built on "propensity to behaviors, not actual purchase behaviors," and usually stale. His tell: "How many of you have loyalty cards from a retailer where the address is no longer an address from where you live? 100% of the people raise their hands."

Then the receipts:

McCormick (spices). Before spending a dollar, the brand ran an "overlap analysis," a Venn diagram of its own taco-seasoning buyers against Instacart's. The overlap was just 7%, meaning 93% of the Instacart audience couldn't have been reached with the brand's other data. The Cinco de Mayo campaign (run on The Trade Desk) beat its new-customer goals and cut CPMs, the cost to reach a thousand people, by "over 20%."

Hershey's Reese's (March Madness). An "immersive" shoppable-TV ad on Roku, the candy "coming in your face," in Silverblatt's telling, let viewers scan a QR code and get the product delivered "in under 60 minutes ... without ever leaving the couch." The kicker: "nearly a third of the consumers who purchase were new to their brand," and over 80% of the brand-lift/consideration movement came from people who hadn't bought the brand before.

Deep Indian Foods (an up-and-comer). Running a multi-month Trade Desk campaign, they flipped on Instacart's live optimization signal mid-flight and doubled their return on ad spend while cutting costs, instead of waiting weeks for a post-mortem.

The analyst's caution (Lipsman, pundit). Here's where it gets useful, because Lipsman spent the segment poking at the industry's comfort metric. Return on ad spend (ROAS) is, he said, "a very costly misnomer": a finance chief hears "return" and thinks it's a real financial return, but a reported $4 ROAS plainly doesn't mean the business grew 400%, "that's all chief customer officers would live for." His preferred lens is incrementality, did the ad actually create a sale that wouldn't have happened? By that measure, he argued, the picture flips: onsite ads "always look great" on ROAS but muted on true incrementality, while offsite ads "don't look good" on ROAS yet are "sometimes the highest performer." He also took a swing at the old marketing-mix models brands use to allocate budget, calling them "not fit for purpose in the age of retail media" because they steer money toward wherever it went last year: TV, social, newspaper inserts.

Net for a book: this is the clearest, most-sourced articulation in weeks of why grocery retail media commands a premium, and why the skeptics ("I'm just paying a data tax for reach I could buy cheaper") may be measuring it wrong. It also lands a specific, positive read-through for Hershey and McCormick as advertisers getting real lift, and for the broader argument that CPG ad dollars keep migrating toward retailer data.

2. The smart cart grows up, and the pitch quietly shifted from "skip the line" to "sell more stuff"

The second rich conversation was The AI Why with Liam Lawson, "What It Takes to Build a Smart Shopping Cart Used by Millions | David McIntosh, Chief Connected Stores Officer, Instacart" (Aug 20). McIntosh runs Instacart's in-store business and is an operator, but he was talking to a tech-product host, not a stock analyst, so the discussion was refreshingly concrete about mechanics.

The plain-English version: Instacart bought a company called Caper, a shopping cart with a screen, cameras, sensors, and a scale built in, for $350 million a couple of years ago. You shop straight into your bags, a screen shows a running total, and you skip the checkout line. What's changed is scale and, more importantly, purpose.

The numbers he gave:

  • 100+ cities, thousands of carts, thousands of connected stores, "tripled year over year," now capturing "millions of sensor inputs every single day."
  • Instacart "did nearly a billion dollars of ads and other revenue last year," and that ad business is now the reason retailers want the carts.
  • The company draws on "more than 1.6 billion lifetime orders" of online history to decide what to recommend to you in the aisle.
  • A concrete lift: a "did you forget?" reminder at checkout, "did you forget the yogurt?", drove a "0.9%, nearly 1%" sales increase, a figure he said Instacart disclosed on an earnings call the month prior.

The tell worth underlining. McIntosh was candid that the original sales pitch, efficiency and skip-the-line, is not what made this take off: "the pull in the market has come from being able to drive sales lift for our retailer partners and being able to drive retail media revenue. And that's the flip that we really saw snowball adoption." In other words, the cart isn't a checkout gadget. It's an in-store advertising and merchandising surface that a shopper stares at for, on average, 30-plus minutes per trip. Once a customer uses it three times, he said, "it's extraordinarily sticky."

A few other threads with read-through value:

  • Why a cart and not ceiling cameras (the Amazon approach): capex and modularity. Rewiring a store with cameras is "extraordinarily expensive"; retailers "just couldn't bear all of that capex." The cart bolts onto a behavior a billion people already know.
  • The AI is genuinely hard: each cart runs an NVIDIA Jetson chip fusing three camera views with a weights-and-measures-certified scale (used like an "X-ray" of the basket), because most retailers "do not have accurate planograms," they don't actually know what's on which shelf, so the cart has to build the map itself.
  • Affordability angle: the cart supports EBT-SNAP running totals (Instacart pioneered SNAP online), and its "Cart Assistant" now does budget meal-planning ("family of five ... budget is $100 a week"), available inside ChatGPT and on retailer sites. Sprouts Farmers Market was named as rolling it out on Sprouts.com and, in future, on the in-store cart.

One caveat worth flagging: McIntosh gave retail-media economics, not contribution profit per delivered order or micro-fulfillment-center math, so the delivery unit economics remain unquantified from this interview.

3. Backdrop: Walmart's ugly quarter

You couldn't open a markets podcast this week without hearing about Walmart's print. On Insight On Business, "The Business News Headlines 20 August 2026" (Aug 20), the summary was blunt: Walmart posted its slowest U.S. comparable-sales growth in six years, 2.6% in the quarter versus 4.1% the quarter before, and the stock fell about 8%.

This belongs in the backdrop bin on purpose. Every voice on it this week was a pundit reacting to the tape, not a grocery operator or a sector analyst adding anything about format share, the Kroger/Albertsons aftermath, or fuel rewards. But the largest grocer in America decelerating hard is a real demand signal for at-home food, so it's worth keeping on the dashboard.

The debate

The argument that aired this week sits one level down the value chain, is retail media as good as its boosters claim?, and it's genuinely two-sided even inside a single, pro-industry conversation:

The bull case (as argued): Grocery purchase data is uniquely rich, deterministic, and fast, and it reaches audiences other data can't (that 93% "new reach" figure). Move the ads up the funnel and onto TV and the open web, measure them properly, and you get real incremental sales. The McCormick, Hershey, and Deep Indian results are the proof points.

The skeptic's case (from the analyst, not a hater): The industry is largely measuring the wrong thing. ROAS flatters everyone ("you buy your own brand keywords and solve the problem," as the hosts joked), and until brands adopt true incrementality with independent, third-party measurement, a lot of retail-media spend is brands "double paying to acquire shoppers they already have." Standardization, Lipsman argued, isn't coming ("a cat-herding exercise"); the best you can hope for is "good enough" outside measurement.

There was also a mini-debate on AI and the future of shopping. The fashionable fear is that "agentic" AI, a bot that does your shopping for you, guts retail-media networks by moving the decision upstream into ChatGPT. Lipsman pushed back hard: it's "evolution, not revolution," just "advanced search," and "the human never gets out of the loop." Instacart's own move is telling: rather than fight the LLMs, it put its "Cart Assistant" inside ChatGPT and on retailers' own sites. If he's right, the AI shopping wave is a feature these networks absorb, not a wrecking ball. That's a live question worth revisiting as the data comes in.

The names in play

  • Instacart (CART), the whole week, really. The through-line: its economic center of gravity is ads and in-store technology, not delivery fees. Nearly $1B in ad and other revenue last year, a smart-cart fleet tripling annually, and an off-platform ad business pushing onto the Trade Desk, Roku, Meta, and TikTok. Whatever you think of the delivery unit economics, the ad-and-data story is the one management wants you focused on.
  • Hershey (HSY) and McCormick (MKC), surfaced as advertisers, but with concrete, positive results (Reese's ~1-in-3 new-to-brand; McCormick 93% incremental reach, 20%+ cheaper CPMs). A small data point, but it says these brands are spending into measurable lift, not flying blind.
  • Sprouts Farmers Market (SFM), named as an early adopter of Instacart's AI Cart Assistant, online and eventually in-store. A tell on which grocers are leaning into the connected-store pitch.
  • Roku (ROKU), the shoppable-TV partner making "scan the ad, get it in 60 minutes" actually happen. If "performance TV" is the next ad wave, Roku is standing in the current.
  • NVIDIA (NVDA), a Jetson chip on every Caper cart. A reminder that the grocery-tech buildout runs on the same silicon everyone else is buying.
  • Walmart (WMT), backdrop this week: slowest U.S. comp growth in six years, stock down ~8%.

Read-throughs

  • CPG retail-media budgets: the clearest signal of the week. The pitch to brands is sharpening from "buy our ads" to "here's provable incremental reach you can't get elsewhere." Expect continued dollar migration from linear TV and generic programmatic toward retailer-data-backed campaigns, and more pressure on the old marketing-mix models that under-credit it.
  • Confectioners and packaged brands as advertisers: Hershey and McCormick getting real new-to-brand lift is a small positive for the "brands can still buy growth efficiently" argument, a modest counter to the structural-decline narrative.
  • Grocers adopting connected-store tech: the capex framing matters. Instacart's cart-based model is deliberately modular and cheap to adopt versus rewiring a store, a lower-friction path for grocers (Sprouts named) to add an in-store ad surface and sales-lift engine without a big build.
  • Physical-AI supply chain: NVIDIA Jetson on every cart, "millions of sensor inputs a day." The grocery-digitization theme is another demand tributary for edge-AI hardware.

What changed from prior weeks

  • Retail media finally got a real, substantive voice. For several weeks this letter flagged retail-media networks, take rates, advertiser mix, the actual mechanics, as the thing worth hearing on. This week it was the headline, from two credible voices, with named campaigns and hard percentages. That's the meaningful shift.
  • Walmart: continuation, not news. Second straight week the largest U.S. grocer's weakness dominated, but again voiced by macro pundits, not grocery analysts.

The one-line version: the money showed up where it actually is, in grocery's advertising and data layer, not on the shelf.