Colgate-Palmolive: scale the creative test, not the content pile

The Colgate-Palmolive and Hill’s pilot offers a useful pattern for testing AI-assisted creative production against a brand standard, without treating faster content generation as proof of incremental sales.

What is actually reported about the pilot?

A December 2025 Think with Google interview describes a Hill’s Pet Nutrition video-ad pilot involving Colgate-Palmolive marketers, internal digital specialists, BCG and Google technologists. The executive reports two consumer-ready ads produced four to six times faster than the usual process, lower cost per concept and the same or better consideration lift. These are company-reported results in a vendor-published account. COL25

The article does not provide the raw timing, cost base, research sample or causal sales evidence needed to reproduce those claims. It supports examining a bounded production-and-review pattern. It does not establish a general speed guarantee, a revenue effect or a software customer result for Upskill Pro.

The management question is which creative work deserves scale

A marketing lead faces a choice between making more assets, improving the range of concepts and learning which message works. AI can expand the first two while leaving the third unresolved. The useful constraint is not the number of videos a system can generate; it is the team’s ability to approve brand-appropriate work and learn from a comparable test.

Our Decision Loop reconstruction is a proposed approach for a brand team, not the company’s reported internal method. It starts with one audience, one communication objective and one acceptance standard. The brand owner, creative reviewer and measurement owner must agree on those before treating fast production as a reason to expand.

Apply the Loop to a bounded creative decision

Signal: identify a real content bottleneck and the audience response the team wants to understand. Structure: distinguish production delay, weak concepts and insufficient distribution evidence. Options: compare the existing process, an AI-assisted concept route and a narrower creative brief that removes unnecessary variants. The model should not choose the campaign’s business objective.

Pilot: the brand owner authorizes a limited set of concepts for human review against the same factual, brand and rights criteria. Keep the measurement setup comparable to the existing route. Measure: examine approved usable assets, total effort, time to approval and the selected audience outcome. Our proposed metrics are a test design for readers, not additional results from the Hill’s account.

Use an approval-and-learning brief instead of an asset count

The working output should state the audience, communication hypothesis, concept alternatives, evidence for the claim, human review decision and intended test. Record why a concept was rejected. That rejection can reveal a factual risk, poor brand fit or an execution problem that asset-volume metrics conceal.

Define the next decision before production begins. A pilot can justify revising the message, improving the production process or continuing the established route. Approval to create a draft is not approval to distribute it. When rights, factual claims or audience suitability are uncertain, the relevant owner resolves the question before publication.

What is the strongest reason not to copy this result?

The company’s baseline production process, creative standards, specialist support and media context differ from a smaller team’s. A reported improvement on two ads is a limited reference class, not proof of a repeatable portfolio effect. A team can produce faster while spending more reviewer time or learning less about demand.

The wrong response would be to dismiss the pilot because its evidence is incomplete. Its value is showing a focused management experiment with a named business objective and a small group close to the work. Borrow the bounded question and the evaluation discipline. Build your own baseline before borrowing the outcome.

What should a marketing manager do next week?

Select one upcoming brief with an existing comparison route. The marketing owner and creative reviewer agree the approval criteria; the measurement owner defines the audience outcome and comparison conditions. Run AI-assisted concept preparation on authorized material and record all review and revision effort, not only generation time.

The marketing owner is accountable for this pilot and assigns the operator who prepares the concepts. Before starting, the measurement owner records a primary decision metric, the minimum meaningful improvement the team would act on and the quality floor that cannot be traded away. Count total review effort and all produced variants. Treat insufficient audience evidence as inconclusive rather than use a small production-speed gain to justify wider distribution. These are locally agreed criteria, not a universal percentage benchmark.

Predeclare the concepts, comparison conditions and decision metric before testing. Track all variants, including rejected or unpromoted ones, and use the same approval rule for both routes. Selecting the best result from a larger AI-generated pool without adjusting the comparison can create a winner-selection effect rather than a more effective creative process.

A concept chosen using exploratory audience data needs confirmation in a separate held-out audience or a new test before that result is treated as confirmed. Alternatively, use a pre-agreed multiple-comparison method appropriate to the test. Do not use the same audience observations both to select the winner and to claim a confirmed effectiveness gain.

Review after the first comparable creative test. Do not scale if usable approvals fall, factual or brand defects increase, or an apparent audience gain cannot be separated from a different distribution setup. Continue when the evidence supports the specific next decision. The first positive value event is an approved concept that improves a real test, not an expanding folder of unreviewed content.

Questions managers ask

Did the Hill’s pilot prove an increase in sales?

The public account reports production speed and consideration lift, not a causal sales result. It does not supply enough underlying data to calculate your expected revenue effect.

How should the four-to-six-times speed claim be read?

As an executive-reported comparison for two ads in a vendor-published pilot account. The baseline timings and replication data are not provided.

What should a marketing team measure first?

Usable approved creative, total preparation and review effort, time to approval and a comparable audience outcome. Separate draft creation, publication approval and campaign effectiveness.

Sources and scope

COL25: Inside Colgate-Palmolive’s content creation gen-AI pilot — Think with Google; 2025-12. Primary vendor interview with named customer executive; public article.

BCG26: AI at Work: Why Strategy Matters More Than Tools — BCG; 2026-06-03. Primary consulting survey; public article and press release.

Research checked 2026-10-27. Next review 2026-11-27. Suggested pilots are editorial proposals, not observed Upskill Pro customer results.