AI is already being used across marketing workflows, from brainstorming campaign concepts to producing video, adapting creative for different markets, and drafting real-time social responses. The documented examples show a consistent pattern: AI can accelerate production, but people still choose ideas, review outputs, protect likenesses and user data, and decide what is ready to publish.
The cases below illustrate distinct approaches rather than a complete or independently audited set of 24 deployments. Their reported results are campaign-specific, not forecasts for other teams.
As an Amazon Associate I earn from qualifying purchases.
Where AI fits in marketing and creative production
The examples span several tasks. Some use AI to expand or visualize ideas; others generate finished assets, adapt them for an audience, or respond to live events. These workflows differ in their inputs, risks, and measures of success, so they are more useful as patterns to evaluate than as a ranking of tools.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →- Concepting: explore trends, develop campaign directions, or guide a brainstorm.
- Asset production: generate images, video scenes, or short-form creative.
- Localization and personalization: adapt content to markets or build material around user-provided information.
- Real-time engagement: draft timely responses or create experiences that react to live context.
Four campaign examples—and what each demonstrates
Lysol: synthetic video production with consumer testing
In a January 2026 case study, Lysol VP Benoit Veryser described an eight-week pilot to create 15- and 30-second broadcast-ready spots for Laundry Sanitizer. The team used Gemini to examine consumer trends and competitor positioning and generate concepts; people selected three directions. It then used Veo and Imagen to produce synthetic creative, iterated on imperfect outputs, and used consumer testing to optimize the final assets. The case also describes a “digital twin” approach using consenting, compensated actors.
#1 Best Overall
Think with Google reported that Lysol’s cost per asset fell 80% compared with its traditional process. Veryser also said the best AI asset performed nearly identically to Lysol’s top traditional asset in short-term sales-lift studies. These are results reported for this pilot by a vendor-published case study, not a general cost or performance guarantee. Veryser drew a useful distinction: “experimenting with generative AI tools isn’t the same as shipping AI-built creative.” Read the Lysol case study at Think with Google.
Snapple, State Farm, and Lyft: three forms of campaign experience
A June 2026 Think with Google article describes three different uses. Snapple and agency Deutsch used Veo 3 to turn brand trivia into stylized short films. Human creatives guided and approved frames, and the article says the team acknowledged AI use in social assets.
State Farm and Highdive created “Stan the State Farm Stanchion Pad,” a real-time social commentator for NBA Finals moments. Gemini tracked highlights and drafted responses; a team reviewed those responses before generated video was made.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Lyft and SpecialGuest’s Lyft Local experience used live route data with AI-generated localized narrative, audio, and visuals to make a ride a guided discovery experience. The article also mentions partner offers within that experience. These are examples of creative patterns, not a measured comparison between vendors or campaigns. Think with Google’s editorial team summarizes the human role this way: “Brilliant AI-powered content requires human imagination to give it the spark and judgment to ensure it resonates.” Read the campaign examples at Think with Google.
Dept and Pit Viper: rapid concepting and video scenes
Google’s 2025 case study says agency Dept used Gemini to develop and refine campaign concepts for Pit Viper, then Veo 2 to generate video scenes. Dept creative director Paul Bjork said AI helped generate ideas and guide the brainstorm, but the work still required curation: “The magic comes from the choices we make and how we use the tools, not the tool itself.”
The case reports a 400% decrease in production timeline. After launch, it reports a 3.8% lift in ad recall in four days, a 6.5% lift among people aged 18–24, and reach of 10.3 million users. These are figures reported for this campaign in Google’s case study; they should not be treated as expected outcomes elsewhere. Read the Dept and Pit Viper case study at Think with Google.
WPP T&Pm: pre-visualization, localization, and personalization
Microsoft’s June 3, 2025 customer story describes agency T&Pm using Azure OpenAI and Sora to pre-visualize campaign ideas and create video variants for markets, languages, recipes, and platforms. One example adapts a rice-brand concept for audiences in the United States, China, and India. Another uses a pet photo and story to create a personalized animated video for a dog-food campaign.
The story presents these workflows as evolving and does not report a quantified performance result. It also emphasizes privacy, security, and responsible-AI guardrails for user-supplied content. That matters in personalization: a workflow that accepts a person’s photo or story needs clear rules for consent, use, access, and handling—not just a prompt that produces a relevant-looking asset. T&Pm’s Oli Egan said, “If we can establish real personal relevance to consumers in the content we create, it’s more effective.” Read Microsoft’s T&Pm customer story.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the reported results can—and cannot—tell you
The headline numbers measure different things: asset cost, production timeline, ad recall, and audience reach. They come from separate campaigns and publisher-reported case studies, not a controlled head-to-head comparison. A percentage reduction in production time is not interchangeable with a percentage change in cost, and neither alone establishes whether a campaign was more effective overall.
- Cost: Lysol’s reported 80% lower cost per asset compares its pilot with its traditional process; the case study does not make that result universal.
- Speed: Dept’s reported 400% decrease in production timeline is a campaign-specific figure. The source does not establish that other teams can use the same method to achieve an equivalent reduction.
- Response and reach: Dept’s reported ad-recall lifts and 10.3 million users reached describe that campaign. Reach is not itself evidence of incremental sales or a causal effect of AI.
- Unquantified workflows: The T&Pm and Snapple, State Farm, and Lyft examples illustrate applications but do not provide directly comparable performance figures in the cited articles.
Use these cases to identify possible workflows and questions for a pilot, not to build a business forecast from someone else’s result. The examples do not establish that one platform or process is best across campaign types.
How to build a workflow that keeps people accountable
In the documented cases, AI did not independently decide what to publish. People selected concepts, guided or curated generations, reviewed responses and frames, tested assets, or set guardrails. For a marketing team, the practical design question is not simply which model to use; it is where human judgment is required and who owns each approval.
- Choose a bounded task and outcome. Decide whether the pilot is for concept exploration, video production, localization, personalization, or real-time engagement. Set a relevant measure—such as time to an approved asset, cost per approved asset, consumer-test results, or a campaign outcome—before comparing it with the existing process.
- Define the human decision points. Assign people to select concepts, assess brand fit, review generated material, check factual claims, and approve publication. For live responses, specify who reviews a draft and what can be published without escalation.
- Set rules for likenesses and supplied material. If a workflow uses an actor’s likeness, document consent and compensation; if it accepts personal photos or stories, define permitted uses and privacy and security safeguards. Lysol’s case describes consenting, compensated actors, while Microsoft’s T&Pm story emphasizes guardrails for user-supplied content.
- Test the output, not just the generation speed. A fast draft is not a successful campaign asset. Review quality and brand fit, and use consumer or campaign testing appropriate to the intended outcome. Lysol’s reported process included consumer testing before final optimization.
- Compare like with like before scaling. Track the same task and unit in the AI-assisted and existing workflows. Separate production efficiency from audience or business results, and record review effort and rework so apparent speed gains do not conceal added approval burden.
Choosing the right use case
A useful first choice depends on the job to be done, the sensitivity of the inputs, and how much review the output needs. The cases suggest a practical way to distinguish them:
- For concept exploration: use AI to expand options or visualize directions, with creative leads selecting and refining the ideas.
- For synthetic video or imagery: plan for iteration, curation, and approval rather than assuming first-generation output is production-ready.
- For localization: define which elements may change by market and have appropriate reviewers check whether the adapted creative remains accurate and suitable.
- For personal content: establish safeguards for user-provided photos and stories before inviting submissions.
- For live social activity: decide how drafts are checked, what context may be used, and how the team handles an incorrect or unsuitable response.
The available examples support these as distinct use cases, but they do not establish an industry-wide average return or prove that all 24 deployments implied by the broader title have been verified. The best fit for a particular team depends on its task, input risks, review capacity, and chosen measure of success. The AI Weekly collection is presented as 24 deployments, but the full set is not audited in the cited material.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
