Cuisinart's AI Ad Beat the Video Its Own Team Made

Cuisinart ran two ad campaigns at the same time, for the same product, with the same money behind them. One 15-second video came from its own production process. The other came out of Amazon’s AI creative tool. The AI version drew 18% more views of the product page and cost 14% less per view.

The number that should interest a marketing lead more is the clock. Conair’s normal video production runs three to six months. This took about four weeks.

Kelsey Smithuysen, who directs Amazon advertising at Conair, the company that owns Cuisinart, told AdExchanger the brand had wanted to use generative creative for a while. The problem was that it kept failing the brand’s own standards. That objection has a shelf life now:

“Another year from now, we can definitely see ourselves using the videos just as they are coming out of the AI tool if they continue to progress in the way they’ve been progressing.”

That’s a schedule for retiring a review step, stated by the person who owns the review.

The distance a brand team just watched close is the story. Twelve months ago the output couldn’t clear an internal bar. Now it beats the thing the bar was protecting, on the metric the campaign was bought against.

What shipped, and when

Amazon put Creative Agent into beta on September 17, 2025. It’s a conversational tool inside Creative Studio that researches the product and audience, sketches concepts as storyboards, then produces finished video and display ads with animation, voiceover, and music. It runs on models available through Amazon Bedrock, including Amazon Nova and Anthropic’s Claude. The finished ads can run on Sponsored Brands, Sponsored Display, and Brand Stores.

None of that is this week’s news, and neither is the test. Marketing Dive reported the Cuisinart results on July 6, with the performance numbers attached. AdExchanger returned to the story on August 5 with Smithuysen’s forward-looking read. The tool itself has been in beta for roughly 11 months. What’s worth your attention is none of those dates. It’s that a brand ran a controlled test against its own work and published the side that lost.

The honest version of “the AI won” also has a fraction in it. The tool produced about 80% of the finished video. The remaining 20% was manual correction by Conair’s agency, Global Overview, fixing logos, colors, and brand compliance.

Amazon’s pitch has been consistent about where the savings sit. The company says sophisticated creative traditionally took “tens of thousands of dollars and weeks of time,” and that its tools produce ads “in just hours, at no additional cost.” Its earlier AI Video Generator made the same argument in smaller form, turning a single product image into an eight-second spot.

Jay Richman, a vice president at Amazon Ads, framed the strategy plainly when Creative Agent launched:

“This is about more than speed—it’s about giving every advertiser and agency access to the kind of strategic, high-quality creative support that once only large brands could afford.”

That’s a vendor describing its own product, so discount it accordingly. The part worth taking seriously is who agrees. Destaney Wishon, CEO of the Amazon agency BTR Media, put the effect on her book of business in one sentence:

“The ability to quickly iterate and produce professional-quality ads dramatically changes what we can offer to clients on mid-market level budgets.”

The mechanism: the brand standard was doing two jobs

For most teams, “it doesn’t meet our brand standards” has been carrying two arguments at once. One is real: the color’s wrong, the product’s mishandled, the tone misses. The other is a proxy for something less comfortable, which is that the output looked cheap and cheap-looking work reflects on the person who approved it.

An A/B test at equal spend separates those. Once the AI version outperforms on the metric the campaign was bought against, the second argument has nothing left to stand on. The first one survives, and you can see exactly what it shrinks to: Global Overview’s 20%, fixing logos and colors and brand compliance. That’s a checklist a person runs at the end, not a production cycle a team runs from the start. The review didn’t disappear. It got cheaper and moved later.

That’s why Smithuysen’s timeline matters more than the test result. She’s naming the month even that 20% stops being worth doing.

What this means if you’re not Cuisinart

Most readers here aren’t buying Amazon ad inventory for a kitchen-appliance brand. The transferable part is the test design, and it’s cheap.

Two campaigns. Same product, same audience, same spend, same window. One creative from your normal process, one from whatever generative tool your ad platform already includes. Then look at the number the campaign was bought against, not at which one you’d rather show in a deck.

The uncomfortable part is that this test answers a budget question nobody asked out loud. If the machine version wins on a mid-funnel product video, the exposed line item is the internal production hours that were never itemized in the first place. Those hours are somebody’s job, and the test makes them visible. We’ve written before about how AI adoption tends to stick when it removes a cost everyone already resented, and about the checking work that gets created downstream when it doesn’t.

Two cautions on the evidence. This is one month, one product, and a brand that wanted the result to work. And the metric that moved is product-page views, which sits well short of a sale. An 18% lift in people clicking through to look at a food processor is a real signal about which creative earns attention. It is not a revenue number, and Conair hasn’t published one.

Justin Swenson, Conair’s senior vice president of e-commerce, is careful about this in a way the coverage mostly isn’t:

“there is definitely something pretty compelling here. Now it’s on us to figure out how to take this and scale it.”

One test is not a capability. He’s describing a promising result and an unsolved operational problem, which is the correct posture and a rarer one than it should be.

Run the test anyway. Not because the AI will win, but because you’ll find out what your brand standard has been protecting. If it’s the product, the checklist survives contact with the data. If it’s the org chart, better to learn that from a one-month test you designed than from a budget meeting you didn’t.