Improve customer discovery and service

Study guide for Generative AI Leader, Unit 2 · Topic 4. This is the topic's lecture in reading form — every slide's teaching, figures and worked examples, in order — followed by the official Google Cloud pages its claims rest on.

What the exam guide asks, quoted. Recognizing the functionality, use cases, and business benefits of Google Cloud’s external search offerings (e.g., Agent Search on Gemini Enterprise Agent Platform , Google Search). Recognizing the functionality, use cases, and business value of Google’s Customer Engagement Suite (e.g., Conversational Agents, Agent Assist, Conversational Insights, Google Cloud Contact Center as a Service).

This hive's learning objectives for the topic:

1. Distinguish enterprise search, public Google Search and grounding use cases.
2. Choose conversational agents, agent assistance, conversation insights or contact-center services for a service need.

Customer discovery and service

Help customers find answers, then help the people who serve them

Discovery — search your own sites and data, ground answers on the public web, or ground them on your documents. Service — a virtual agent first, a human agent with live assistance when needed, and analysis of every conversation afterwards, on a contact center platform that routes them all.

This topic covers how Google Cloud's generative AI offerings improve the customer experience, and it splits into two decisions. The first is discovery: when a customer or a model needs an answer, where should that answer come from? Agent Search, formerly Vertex AI Search, builds Google-quality search over your own websites and data; Grounding with Google Search connects a model to world knowledge and up-to-date information from the public web; and grounding on your enterprise data ties a model's answers to your own documents. The second decision is service: which part of the customer engagement products fits a given need? Google now documents these as the Gemini Enterprise for Customer Experience products — an agent builder for self-service, Agent Assist for the human representatives, Customer Experience Insights for analysing conversations, and a contact center platform that queues and routes every interaction. We take discovery first, then service, and close on a naming slide, because the exam guide still uses the older names.

Three ways to find an answer

Your data, the public web, or a model grounded on one of them

• Agent Search: Google-quality search over your own sites and data
• Grounding with Google Search: world knowledge and current web information
• Grounding on your data: answers tied to your own documents
• Grounding connects output to verifiable sources and reduces hallucinations

Worked example (synthetic). A retailer wants shoppers to search its catalogue in plain language, and wants its help bot to answer from the returns policy. The first is enterprise search; the second is grounding on its own documents. Neither needs the public web.

The first objective asks you to tell three answer sources apart. Enterprise search is Agent Search: Google says it lets you build AI-enabled search and recommendation experiences for your public websites or mobile applications, and it brings natural language understanding, semantic search, synonym understanding, spell correction and auto-suggest out of the box. Google names two opportunities for it — improving search across your intranet and customer-facing websites, and improving generative AI applications by grounding them in your enterprise data. Public Google Search is different. Grounding with Google Search is the choice when you want to connect your model with world knowledge, a wide range of topics, or up-to-date information on the internet, and it uses publicly available web data. Grounding itself is the common idea: Google defines it as the ability to connect model output to verifiable sources of information, and one of its stated benefits is reducing hallucinations, which are instances where the model generates content that isn't factual. So the question to ask is always where the trustworthy answer lives: in your own content, or on the public web.

Match the need to the answer source

Ask first where the trustworthy answer lives

Worked example (synthetic). Four requests from one company: three of them are answered from content the company owns, and only one needs the public web.

Lay the needs side by side and the choice becomes mechanical. When shoppers or staff search a site or intranet, the answer is enterprise search with Agent Search — Google describes moving from frustrating keyword matching to modern conversational search experiences, and says this can be as easy as adding a search widget to a webpage. When a chatbot must answer from policy documents, the need is grounding on your data: Google's grounding guidance uses retrieval-augmented generation to connect a model to your website data or your sets of documents stored in Agent Search. When a model must talk about what is happening in the world today, its training data cannot know, so Grounding with Google Search connects it to up-to-date information on the internet. Notice the pattern in the exam's distractors: public Google Search offered for a question only your own content can answer, or a site search offered where the model needs world knowledge.

Same question, two different right answers

Grounding source follows the question, not the product

Figure. Two cards. The first, What is your refund window, is answered only by the company's own documents, so it is grounded on your data with Agent Search. The second, What did the new consumer law change this week, is public and recent, so it is grounded with Google Search. A band beneath says either way grounding ties the answer to verifiable sources.

Worked example (synthetic). A fictional travel insurer's assistant gets both questions in one afternoon and needs both grounding sources configured.

Here are two questions a fictional insurer's assistant receives on the same day. The first — what is your refund window? — has exactly one correct answer, and it lives in the company's own policy. Grounding on your data, through Agent Search's retrieval over your documents, is the only source that can know it; the public web might even carry an out-of-date copy. The second — what did a new consumer law change this week? — is public and recent, which is exactly the case Google gives for Grounding with Google Search: up-to-date information on the internet. The lesson is that the grounding source follows the question rather than the product. In both cases the purpose is the same: grounding connects the model's output to verifiable sources of information, which is what reduces hallucinations.

What enterprise search gives you without building it

Agent Search arrives with the hard parts of search already done

Worked example (synthetic). A fictional garden centre replaces its keyword search; "plants for shade" now finds hostas and ferns although neither page uses the word shade.

Enterprise search is easy to underestimate, so look at what Agent Search provides before anyone writes code. Google lists natural language understanding and semantic search, synonym understanding, spell correction and auto-suggest, all out of the box. For a customer, that means a plain-language question finds the right page even when the words differ, a typo still lands, and suggestions appear while typing. Google describes the shift as going from frustrating keyword matching to modern conversational search experiences, and says getting started can be as easy as adding a search widget to your webpage. This is why, in an exam scenario about customers struggling to find products or articles on the company's own site, the answer is enterprise search rather than a model grounded on the public web.

Four products for customer service

Self-service, live assistance, analysis, and the platform underneath

• CX Agent Studio: build AI agents for self-service conversations
• Agent Assist: real-time help for human representatives
• CX Insights: patterns, sentiment and call drivers across conversations
• CCAI Platform: contact center as a service that queues and routes

Worked example (synthetic). A bank's phone queue is overloaded. Routine balance questions go to a virtual agent, complex disputes reach a person who sees suggested answers, and managers review weekly what drove calls.

The second objective is choosing among the customer engagement products, and each one answers a different need. For self-service, Google describes Customer Experience Agent Studio as a minimal code conversational agent builder that also uses AI to have controlled conversations with end users at runtime — the agent that answers customers before any person is involved. For the humans, Agent Assist provides in-the-moment coaching and next-best action guidance to customer care representatives, helping them resolve issues faster and more accurately, and when an interaction is complete it automatically summarizes the interaction, key takeaways and next steps. For managers and analysts, Customer Experience Insights helps detect and visualize patterns in contact center data, running machine learning analytics for information such as agent and caller sentiment, entity identification and call topics. And underneath sits the contact center platform: Google describes the Contact Center AI Platform as an AI-driven contact center as a service, or CCaaS, that queues and routes customer interactions across voice and digital channels, purpose-built to work alongside customer relationship management systems.

One conversation, four products

Each product owns a stage of the same customer conversation

Figure: A left-to-right flowchart. A customer contacts the business; CCAI Platform queues and routes the interaction to a virtual agent built in CX Agent Studio. If the virtual agent resolves the issue the conversation is done; if it cannot, it is handed off to a human agent who receives Agent Assist suggestions, and then it is done. A dotted arrow from done leads to CX Insights, which analyzes drivers and sentiment afterwards.

Worked example (synthetic). A fictional utility routes 10,000 contacts a week this way; the virtual agent resolves most outage reports, and Insights shows that billing confusion drives the rest.

Follow one conversation through the products. The contact center platform takes it first, because queuing and routing customer interactions across voice and digital channels is what it does. Google describes designing the system so that end users initially interact with a virtual agent before being escalated to a human agent — that virtual agent is the self-service agent you build in CX Agent Studio. If it cannot resolve all of the customer's issues, the conversation is handed off to a human agent. At that point Agent Assist supplies real-time document and response suggestions to the human agent that are relevant to the conversation. And after the conversation, Customer Experience Insights helps detect and visualize patterns in contact center data — it can import conversations from the virtual agents and from Agent Assist. Read an exam scenario by asking which stage it is about: before a person is involved, while a person is helping, or after the conversation is over.

Choose by the need, not the brand

Who is helped, and when, decides the product

Worked example (synthetic). Four requests from a fictional airline's customer operations team, each answered by a different product.

Turn the products into a decision table by asking who is being helped. If the need is to answer routine questions without a person, the product is a self-service agent from CX Agent Studio, which has controlled conversations with end users at runtime. If a representative needs suggested answers in the middle of a call, that is Agent Assist, which provides in-the-moment coaching and next-best action guidance. If a manager needs to know why customers call and how they feel, that is Customer Experience Insights, which analyzes caller sentiment and call topics across conversations. And if the need is the plumbing of the contact center itself — queuing and routing interactions across channels — that is the contact center platform. A common exam trap pairs a manager's reporting need with Agent Assist, or a self-service need with Insights; the who-is-helped question rules both out.

The guide's names and today's names

Expect the older names on the exam and the new ones in the docs

Worked example (synthetic). A candidate who has only read current documentation meets the phrase Conversational Insights in a question and recognises it as CX Insights.

The exam guide was written with the earlier product names, and Google's documentation has since moved on, so learn both. What the guide calls the Customer Engagement Suite is documented today as the Gemini Enterprise for Customer Experience products. The guide's Conversational Agents are built today in CX Agent Studio, which Google describes as the evolution of Dialogflow CX. Conversational Insights is now Customer Experience Insights, and its own page still says it can import conversations from Conversational Agents and Agent Assist, which is a useful bridge between the two vocabularies. Contact Center as a Service is the Contact Center AI Platform, which Google calls an AI-driven contact center as a service built natively on Google Cloud. And Agent Search was formerly Vertex AI Search. The capabilities did not change with the names, so answer the question the scenario asks, whichever name it uses.

What this topic actually tests

Where does the answer live, and who is being helped?

Where does the answer live? your own sites and data → Agent Search; the public, current web → Grounding with Google Search; your documents inside a model's answer → grounding on your data. Who is being helped, and when? customers on their own → CX Agent Studio; a person mid-conversation → Agent Assist; managers afterwards → CX Insights; the whole queue → CCAI Platform.

Two questions carry this topic. First, where does the trustworthy answer live? If it is in your own websites and data, Agent Search builds Google-quality search over it; if it is on the public and current web, Grounding with Google Search connects the model to world knowledge; and if your own documents must shape a model's answer, ground the model on your data. Second, who is being helped, and when? Customers helping themselves need an agent built in CX Agent Studio; a human representative mid-conversation needs Agent Assist; managers looking back across conversations need Customer Experience Insights; and the contact center as a whole runs on the CCAI Platform, which queues and routes every interaction. Hold those two questions and the older and newer product names stop mattering.

Official sources for this topic

• Agent Search on Gemini Enterprise Agent Platform — Google Cloud
• Grounding with Google Search — Gemini Enterprise Agent Platform
• Customer Experience Insights — Google Cloud Documentation
• About Agent Search — Gemini Enterprise Agent Platform
• Grounding overview — Gemini Enterprise Agent Platform
• CX Agent Studio — Google Cloud Documentation
• Agent Assist — Google Cloud
• CCaaS (CCAI Platform) — Google Cloud Documentation
• Agent Assist basics — Google Cloud Documentation