As a digital agency based in Ho Chi Minh City, Vietnam, we have compiled a set of practical generative AI use cases, primarily centered around Google Workspace, designed specifically for non-engineers.
We are eager to exchange ideas with other companies on how AI is being utilized in digital advertising. Additionally, our experience as a trilingual team (Japanese, English, Vietnamese) focusing on breaking down language barriers may serve as a useful reference.
To create real value, these efficiency gains must be paired with solid organizational design. In the near future, autonomous agents will likely replace many manual tasks and eliminate communication friction**.** However, we believe that mastering today's AI tools is the essential foundation for that automated future. This article documents our current practices to help build that foundation.
We build advertising reports using Google Sheets + ETL, and add AI-generated analysis comments when needed.
.png)
Our typical client communication structure:
Although tools like Databox (dashboards) and Meltwater (social listening) include AI summaries, we often return to Google Sheets because:
We use NotebookLM to share proposal materials and enable multiple team members to conduct AI-assisted self-reviews in a shared environment.
.png)
How it works:
This is a prompt.
When "Check "****"" (where **** is the name of the file uploaded to the source) is sent to the chat, follow the instructions below.
-
Point out the following based on the relevant slide in the source. Think carefully.
- Point out typos or obvious grammatical errors.
- Point out any obvious inconsistencies in the story of the slide content.
- Point out obvious numerical inconsistencies in the numbers or tables on the slide.
- Point out inconsistencies in the months (December, November, etc.).The AI checks for:
This approach is faster than ad-hoc AI chats and enables the shared environment to build on the context of previous reviews for more consistent feedback.
We found that NotebookLM (Gemini 1.5 Pro) reduces hallucinations compared to standard Gemini, making it particularly effective for reviewing English materials created by non-native speakers.
To explain web advertising mechanisms to clients and team members, we use Mermaid, a text-to-diagram syntax that generates visuals from simple code.
.png)
When sharing AI-generated content:
If clients already have access to a Google Drive folder, simply adding the Google Site file enables seamless sharing.
.png)
We stay updated on advertising platform changes through:
.png)
This ensures we receive regular updates on:
To move beyond simple prompt engineering, we leverage Google Workspace Studio to build custom, context-aware AI workflows that autonomously handle routine agency operations. This represents a significant shift in how we work.
FFV operates with:
For meetings:
We then:
Since notebooks are shared, team members can review content in their preferred language, regardless of the meeting language.
Generative AI enables non-native speakers to create near-native-quality slides.
We:
instruction_type: "Generic Slide Deck Generation based on Corporate Brand Guidelines"
target_brand: "Feedforce Vietnam Style"
input_source: "User provided Google Doc content"
# -----------------------------------------------------------
# DESIGN SYSTEM
# Fixed design rules to replicate the original Feedforce Vietnam profile look and feel
# -----------------------------------------------------------
design_guidelines:
visual_tone: "Professional, Data-Driven, Clean, Trustworthy, Growth-Oriented"
typography:
style: "Arial (system-default sans-serif for maximum compatibility)"
readability: "High contrast, minimal text density per slide, prioritized hierarchy"
color_palette:
primary_brand: "#6EBB44" # FFV Green (Use for headers, key emphasis, arrows)
primary_text: "#131C4B" # Deep Navy (Use for main text, title slide backgrounds, footers)
background_main: "#FFFFFF" # White (Standard background)
accents:
- "#FF8017" # Orange (Use for Call to Actions, 'Act' phase in PDCA, or Highlights)
- "#04A5D5" # Light Blue (Use for secondary data points, 'Plan' phase)
- "#FFC600" # Yellow (Use for icons, 'Check' phase, or notifications)
- "#FEA400" # Gold (Use for achievements, rankings, or premium features)
layout_preferences:
# ... [content omitted for brevity] ...Part of the YAML for creating Feedforce Vietnam slides
Using machine-friendly formats like:
is essential for effective collaboration between non-engineers and AI.
With well-designed prompts, non-native speakers can produce high-quality first-draft ad copy, sometimes exceeding native-level output.
Our approach:
Example prompt:
You are Japanese.
Use the following English copy as a reference to generate ad text for a Japanese Meta ad.
Note that a direct translation of English into Japanese sounds very unnatural, so make sure that the output is natural Japanese and does not contain any unnatural words.
Your audience is also Japanese. The ad copy must be perceived as natural and thought up by a Japanese person.
Also, your audience must never realize that the ad copy is Japanese translated from English or created by a generative AI. Take your time and think carefully.This ensures:
Although Google AI is our primary tool, we use ChatGPT and Claude for:
This helps bridge understanding across languages and contexts.
We "vibe-coded" lightweight internal tools using AI (GAS + Slack APIs):
This approach is cost-efficient and tailored to our needs.
Due to Vietnamese accounting requirements, we must track time allocation by project.
Workflow:
Purpose:
We also experiment with:
We are still in the experimentation phase, continuously updating and refining our AI stack while actively evaluating and integrating new tools into our workflow.
Generative AI is no longer just a productivity tool—it is the foundation for future organizational transformation and the key to redefining our core value proposition as an agency.
By combining:
We aim to:
Related Articles
.png)
GA4 engagement rate and bounce rate are often misinterpreted because they no longer mean what they did in Universal Analytics. This guide breaks down GA4’s exact definitions, explains when a high bounce rate is completely normal, and shows how to use these metrics to make smarter marketing and UX de
January 22, 2026
.png)
High CPM, unstable CPA, and no conversions after 1–3 days? Before you kill the campaign, understand the Learning Phase. Here’s how Meta learns, what the 50-conversion guideline actually implies, and the setup changes that stabilize performance instead of resetting learning.
February 3, 2026
