Mid-size advertisers now make their own AI ad variants. Sort your retainer before your clients do

On October 2, Digiday reported that three advertisers well below the biggest budgets are now making large volumes of AI ad variants with their own teams. Two days earlier, Runway announced a product that makes, publishes and remakes paid ads in one loop. Both point at the same part of an agency retainer. To see how that work moves to agents while you keep the review step, see how to run a one-person agency with AI agents.

What the three advertisers are doing

The three are Wyndham Hotels & Resorts, Opella and BetMGM. Digiday noted that Wyndham’s $565 million marketing and loyalty budget last year is dwarfed by the budgets of first movers such as Unilever and L’Oréal.

• Wyndham. The company has worked with the AI firm Adora for a year. Its five-person in-house marketing team makes versions of assets from real images of its hotels for Instagram. Wyndham’s Mike Shiwdin told Digiday the team makes 15 times as many assets as before and has cut concepting, production and approval time by 75%. He also reported a six times improvement in site visit rate over the company’s benchmark. These are company-reported figures.
• Opella. The consumer health company has an 85-person team using generative AI tools. Julien Leterrier, its global head of martech, said it makes 20 times as much creative as before and has made more than 20,000 pieces of content. Most are variants or core assets for different languages and markets.
• BetMGM. The company spends about $160 million a year on media and has a 120-person in-house creative department. It uses AI tools for some imagery and video. It has held back from AI assets that carry betting odds information.

What they still send to agencies

All three still work with creative and media agencies. Wyndham’s agency of record is Havas. Digiday reported that agencies get the briefs for larger-scale, more traditional creative that benefit from an outside perspective, such as BetMGM’s recent campaign starring Jon Hamm.

The limits on AI follow the same line. Opella has firm rules on showing medical professionals, and adds AI disclosures where markets require them, such as the EU. Leterrier put it in three words: “No fake doctors.” BetMGM’s Victor Gonzalez said, “You can’t hold AI accountable.”

Two more details matter. Pencil’s chief executive told Digiday that its fastest-growing client group is companies with less than $1 billion in annual income. And Kantar’s early-stage creative tests rose 17% between December 2025 and July 2026, which Kantar’s Kristi Rogers tied directly to more brands adopting AI creative production.

The tool is now built to run that loop

Runway Ads takes a brand kit, past ads and product images. It makes video and image ads, publishes approved variants to Meta, Google and TikTok, reads results back and makes the next round from what earned spend. Every variant goes through an automated brand check before an approval queue. Human approval is on by default. Teams can switch to automated publishing by campaign or variant type.

Runway names agencies as one use case: an agency running paid for a roster of clients could use the same test-and-regenerate loop on every account. On its own Meta and TikTok ads since July, Runway reports, in company-reported figures, weekly ad volume rising from 77 to roughly 900, return on ad spend doubling, conversion up about 34% and cost per subscriber down 41% while spend rose. For now it is a pilot with select enterprise partners.

A decision rule: sort every line of your retainer

Variants moved in house. Accountable claims and outside ideas did not. Take each line of each retainer and ask three questions in order.

1. Does it carry a claim someone must answer for? Prices, health, odds, legal terms, real people. If yes, it is an approval line. Sell the rules, the review and the sign-off.
2. Does it need a point of view from outside the client? A campaign idea, a positioning shift, a creative route to test before production. If yes, it is a judgment line. Larger, more traditional creative that needs an outside view is the work all three advertisers still give to agencies.
3. Could a small in-house team make it from a brand kit and past ads? Resizes, language versions, hook variants, fresh cuts of a winning ad. If yes, and the first two answers were no, it is a factory line.

Factory lines are what a client can now move in house or onto a tool. For each, decide: run it yourself on agents at a lower cost to you, price it as a system with the rules and testing attached, or plan to hand it over with a clean playbook.

Take a hypothetical retainer. Monthly social variants and platform resizes are factory lines. Translated versions are factory lines unless the market has disclosure rules, which makes them approval lines. A supplement brand’s claim review is an approval line. The quarterly campaign concept is a judgment line. If most of the fee sits in the factory pile, that is the fee most at risk.

What to do this week

• Sort every retainer line into factory, approval or judgment, and add up the monthly fee in each pile.
• For each factory line, write down which of the three choices you will make.
• For each approval line, write the rule in one sentence and name the person who signs off.
• Ask your two largest clients what they now make in house, before they bring it up.

The guide on running a one-person agency with AI agents covers how to move repeatable work to agents one piece at a time and keep the review step. If you would rather build that approval line with a coach working alongside your team, that is the kind of process our Build engagement builds with you and teaches your team to run.