Matching products to business scenarios

The last session before the holidays was a test. Farah had written twelve short scenarios on index cards, each describing a business need at a real or imagined company, and dealt them face down on the table. "The Generative AI Leader exam is mostly scenarios like these," she said. "You read a situation, spot what actually matters, and choose the offering that fits best. Not the most impressive one. The one that fits." Owen turned over the first card, and Jules, who had decided to sit the exam too, took notes.

This lesson pulls the course together. It gives a method for scenario questions, a summary of which products fit which needs, worked scenarios, and the traps that catch candidates. It does not reproduce real exam questions; the scenarios are original and written for practice.

A method for scenario questions

Farah's method had five steps, the same ones she used at work when a team came to her with an idea.

• Find the user. Is the scenario about employees, customers, developers or the platform team? The user often decides the layer of the stack.
• Find the job. Writing, searching, answering customers, building something custom, choosing a model, grounding, securing, or running at scale?
• Find the constraint. Speed to value, data location, privacy, cost, skills available, or the need for control. Scenarios usually include one clue that rules out options.
• Choose the highest layer that fits. Prefer buy over assemble and assemble over build, unless the constraint demands more control.
• Check the controls. Whatever the product, does the answer respect access, data protection and safety? A choice that ignores security is rarely the best answer.

The product map in one table

Worked scenarios

Owen worked through Farah's cards aloud. Here are six of them, with the reasoning that led to each answer.

Reading the clue words

Farah taught Owen to look for clue words in scenarios. "Employees", "staff" or "internal" point towards the Gemini app, Workspace or Gemini Enterprise. "Customers", "website" or "mobile app" point towards the Customer Engagement Suite or a custom build on Agent Platform. "No developers" or "quickly" favour prebuilt products. "Custom", "integrate with our systems" or "full control" favour the platform. "Current" or "recent public information" suggests grounding with Google Search, while "our documents" or "our data" suggests Agent Search or Gemini Enterprise. "Sensitive", "regulated" or "in our country" signal that the answer must include security and data controls.

Clue words are a guide, not a rule. A scenario might mention customers but actually ask about the contact centre agents who serve them, which points to Agent Assist. Reading the question to the end, and asking what is actually being decided, matters as much as spotting the words.

Owen made a habit of crossing out options on scratch paper as he read. An option that serves the wrong user goes first, then any option that breaks a stated constraint, then any that needs much more building than the scenario calls for. What usually remains is one clear answer, sometimes two, and the final choice comes down to which one meets every requirement with the least effort and risk.

Traps that catch candidates

After the twelfth card, Farah listed the traps she had seen colleagues fall into.

A practice scorer

Owen built a tiny script to quiz himself with the product map. Each run picks needs and checks whether he names the right product first. It is a study aid, not an exam simulator.

# A self-check drill: does each answer match the product map?
product_map = {
    "employees search across many company systems": "Gemini Enterprise",
    "staff help inside Gmail and Docs": "Gemini for Google Workspace",
    "custom customer-facing agent": "Agent Platform",
    "compare partner and open models": "Model Garden",
    "screen prompts and responses": "Model Armor",
    "answer customers on chat and phone": "Customer Engagement Suite",
    "train a huge model from scratch": "AI Hypercomputer",
}
my_answers = {
    "employees search across many company systems": "Gemini Enterprise",
    "staff help inside Gmail and Docs": "Gemini for Google Workspace",
    "custom customer-facing agent": "Gemini Enterprise",
    "compare partner and open models": "Model Garden",
    "screen prompts and responses": "Model Armor",
    "answer customers on chat and phone": "Customer Engagement Suite",
    "train a huge model from scratch": "AI Hypercomputer",
}
right = 0
for need, correct in product_map.items():
    ok = my_answers[need] == correct
    right += ok
    if not ok:
        print(f"Review: '{need}' -> {correct}, not {my_answers[need]}")
print(f"{right} of {len(product_map)} correct")

Review: 'custom customer-facing agent' -> Agent Platform, not Gemini Enterprise
6 of 7 correct

His one wrong answer was the most common trap: putting a customer-facing agent in Gemini Enterprise, which serves employees.

Where Birchline landed

At the end of the session Farah asked Owen to summarize Birchline's decisions for the steering committee in a table. It is a plan for 2027, so every date in it is still ahead.

"You're ready for GCP-112," Farah said, collecting the cards. "Knowing what each product does is half of the exam. The other half is knowing how to make a project succeed with people, money and risk. That's next."