Puglisi Consulting for Artificial Intelligence

Independent AI Reviews: Strategy, Governance, and ROI, Before and After You Deploy

Digital Marketing Strategy using AI: Cheap Ad Creative Leaves Brand Fit as the Scarce Work #AIg

by

Wall of digital advertising images, from the INFORMS release on AI-generated ads
Image: INFORMS, AI-Generated Ads Outperform Human Designers in Live Campaign news release (August 4, 2026).

Generative tools have made ad images cheap to produce, and that moves the hard work to whoever decides which images may run under the brand’s name. Basil C. Puglisi reads the latest research and trade coverage as one signal: production capacity stopped being the constraint, and brand judgment took its place.

For a marketing team, the question for this quarter is who rates the generated variations against which written rules, and who holds the authority to stop one before money is spent.

Cheap creative leaves brand judgment as the scarce work

Marketing Dive reported on November 5, 2026 that Clorox’s Fresh Step and the AI-powered agency RECE went from brief to three 15-second spots in 30 days at one-tenth the cost of a traditional production. Marketing Dive doesn’t say who supplied the cost comparison or how it was calculated, so treat the figures as reported, not audited.

They still show the trend. When production shrinks to about a month at a fraction of the cost, the time that once let reviewers catch off-brand work shrinks with it. Puglisi’s September 16, 2026 post on HAIA-SMART Creative, a rating tool for social media visuals, addresses a related question. It makes this observation: “Image generators made visuals cheap, and the cost of that abundance was sameness.” Abundance without a rating step produces volume, and the rating step needs an owner.

What the Marketing Science study tested

Remi Daviet of INSEAD and Yohei Nishimura of the Wisconsin School of Business studied this problem in a paper in Marketing Science titled “Leveraging Generative AI to Create Visual Content in Digital Advertising.” INFORMS, the journal’s publisher, said in an August 4, 2026 release that the authors used an active-learning engine in a live campaign to find high-performing AI images while keeping them on-brand. INFORMS also reported that the system outperformed human-made visuals and an AI model optimized only for aesthetics.

The release gives the numbers. In a live Instagram campaign for an outdoor-activities company, the AI portfolio recorded a mean click-through rate of 0.98%, against 0.65% for a professional human designer and 0.78% for the aesthetics-only AI benchmark. The release adds that the designs “were vetted for brand standards” before they ran, and that a later peak-season test without retraining came out 3.38% to 3.24%, a much narrower margin.

A caution belongs here. The release is the publisher’s own summary of one campaign, for one company, on one platform, and it doesn’t say how brand fit was scored. Check the paper itself before building a plan on those numbers. The release’s own conclusion is the useful part: the bottleneck “is shifting from production cost to the intelligent exploration of creative possibilities under brand constraints.”

Puglisi’s reading doesn’t depend on the click figures. Any system built to keep images on-brand needs a definition of “on-brand,” and a person who owns the brand has to supply it. That input is the part a general-purpose model can’t invent for a company it has never worked for.

Brand rules live in people and old files

Gene De Libero, a contributor to MarTech, wrote on November 4, 2026 that brand rules and approval rights mainly exist in brand PDFs, old email threads and the heads of the longest-tenured people. He argues that slow production once gave those people time to flag off-brand work, and that “AI removed the brake.” This is a contributor’s opinion piece, not a study, so it describes a practitioner’s concern without showing how often the problem occurs.

Scott Brinker raised the same issue in his chiefmartec newsletter on October 9, 2026, calling it the alignment gap. His point for teams was that they have to decide “who that human is, what authority they have, and the time and tools we give them to make decisions.” Those are management decisions, and a vendor demo won’t make them. De Libero’s piece builds on Brinker’s, so read them as two commentators on one argument rather than two separate findings.

An automated brand checker is machine checking machine

Puglisi separates two arrangements that teams often treat as one. In Responsible AI, a machine checks the machine, and any human who’s present can be bypassed. In AI Governance under Puglisi’s Checkpoint-Based Governance (CBG), the work stops at a point where “a named human holds Tier 0 authority to accept, modify, or reject.”

An automated brand-fit filter, whether a model trained on the brand owner’s ratings or a vendor’s brand checker, is the first arrangement. It can screen volume well once it is calibrated. It cannot hold accountability for what runs, because no named person answers for its decision.

Puglisi would place the human checkpoint in two places in this kind of workflow: the ratings that teach or calibrate the filter, and the owner’s final vetting before spend. A team has to choose both, because neither happens by default.

Five steps for a client marketing team

  1. Name the brand owner and write the red lines first. One person owns the ratings that train or calibrate any brand-fit filter. Start with a short list of what creative must show and must never show, because a checker can only learn rules that exist on paper.
  2. Budget that person’s rating hours. Put rating time on the plan as a line item with a calendar slot. Treated as a favor, it’s the first thing cut when a launch date moves.
  3. Test agreement before a machine screens live creative. Hold back a sample the owner has rated, run the automated checker or a general vision model on it, and compare. Keep the owner’s accept, modify or reject decision at final vetting before spend.
  4. Re-rate when conditions change. A revised brand rule or a different image generator can shift what’s acceptable, so pull a fresh sample whenever either changes.
  5. Judge creative as a portfolio. Compare the average ad and the weakest ad alongside the best performer, and keep targeting and budget fixed while testing so that creatives are compared under similar campaign conditions.

Three measures for the dashboard

Puglisi would put three numbers in front of the brand owner each month:

  • Agreement rate between the automated checker and the owner’s ratings on a monthly sample. A falling rate is the signal to re-rate.
  • Modify and reject share at the owner’s final vetting. A share near zero deserves a spot check, because it can mean the owner has stopped looking.
  • Share of test spend going to owner-vetted creative. The target is all of it.

These measures are Puglisi’s recommendations, not findings from the study. They haven’t been validated across companies, and the underlying evidence is thin: a publisher’s summary of one study, a trade report of one brand’s campaign, and two commentators making a shared argument. Fresh Step’s one-tenth cost figure tells a team how cheap production can get, and says nothing about how well any brand-fit checker performs.

What to watch next

The open question is the price of the rating hours. If brand-fit checkers become standard, the scarce resource is likely to be the brand owner’s time, so measure how many hours calibration takes and how often the answers drift. Ask any vendor selling a brand checker for its agreement rate against a customer’s own ratings, and watch whether the figure is published or only promised.

Sources

Questions readers ask

What is Checkpoint-Based Governance in an ad creative workflow?

Checkpoint-Based Governance (CBG) is Basil C. Puglisi’s framework for placing a named human decision at defined points in an AI workflow. In ad creative, that person accepts, modifies or rejects images at set stages, such as final vetting before spend, and the decision is recorded.

Why does cheap AI ad creative make brand judgment the scarce work?

When AI shortens production time, slow production no longer gives experienced people time to flag off-brand work. MarTech contributor Gene De Libero describes this as AI removing the brake. Review capacity and clear approval rights then limit how much creative a brand can safely run.

Can an automated brand checker replace the brand owner?

An automated checker can screen large volumes once calibrated, but it is machine checking machine, which Puglisi files under Responsible AI. It cannot answer for what runs. Teams should test its agreement against the owner’s own ratings on a held-out sample and keep the owner’s final vetting before spend.

What should a marketing team measure when it uses a brand-fit filter?

Puglisi recommends three monthly measures: the agreement rate between the checker and the owner’s ratings on a sample, the owner’s modify and reject share at final vetting, and the share of test spend going to owner-vetted creative. These are recommendations rather than study findings, and teams should adjust them to their own brand.

How reliable is the evidence behind this advice?

The evidence is limited. It rests on INFORMS’s summary of one Marketing Science study run with one company on one platform, a Marketing Dive report of one brand’s campaign, and two commentators, De Libero and Brinker, making a shared argument. Check the paper before relying on its figures.

#AIg

Reader Interactions

Leave a Reply

Your email address will not be published. Required fields are marked *