AI Agent as a decision guide

With Eloquent you can easily create an AI agent that helps users make the right choice. This type of agent is designed to guide visitors step by step toward the product, service, or solution that best fits their needs. It feels like a personalized advisor, always available, always consistent, and always branded as your own.

How it works

The agent is trained on the company’s product or service data — for example, website content, product sheets, or FAQs. In the prompt, you define which criteria must be taken into account. Most of the time, the agent will ask clarifying questions to better understand the user’s situation before giving a recommendation. Once the criteria are matched, the agent suggests the best-fitting product or service from the knowledge base.

Our recommendations

• Make it visible → Add a button with a clear outcome, e.g. How many fire alarms do I need in my home? "Take the test.”
• Set expectations → Don’t pretend the AI is human. Users dislike being misled. Instead, be transparent — they’ll appreciate the speed and precision.
• Use the right interface → The Eloquent embed feels familiar (like ChatGPT) but optimized for websites. For this use case, a button or embedded input field as trigger works best, as it prompts users to start the “decision journey.”

What can it do?

• Guide users step by step with clarifying questions to uncover their real needs.
• Match needs to offerings by comparing criteria against the dataset of products or services.
• Provide tailored recommendations in plain language that’s easy to understand.
• Escalate to a human if the situation is unclear or requires expert advice.
• Integrate with systems to pull in real-time product availability, pricing, or service details.

What pain points does it solve?

• Overwhelmed users who don’t know which product or service to choose.
• High drop-off rates on websites because visitors can’t find what they need.
• Repetitive sales and support questions about “what’s the right option for me?”
• Inconsistent advice across different staff members or channels.
• Time lost on low-value qualification calls that could be automated.

Who it’s for

This use case is especially relevant for:

• E-commerce → helping shoppers select the right size, model, or product bundle.
• Professional services → guiding clients to the right service package or subscription.
• Education & training → helping students choose the right course or program.
• Healthcare & wellness → supporting patients in selecting the correct plan, treatment option, or intake path.
• Public sector & non-profits → guiding citizens to the right service, form, or subsidy.

How to set it up

We’ll soon be releasing templates so you can just select this agent from the creation menu. Until then, we recommend the following settings:

• Agent temperature: 0.1 (strict, consistent answers)
• Reranking: Off (faster response, datasets are usually small here)
• Embeddings: 10 (enough for focused product/service sets)
• Similarity: Default
• Prompt starter, this use case is pretty specific but this prompt will get you going:

# INSTRUCTIONS (DISCOVERY)

## ROLE  
You’re an expert in **guiding users to the right {product/service} choice** for {company_name}.  

## GOAL  
You’re on the website {website_url} to help visitors discover the right {product/service} for their needs, and guide them step by step in making an informed decision.  

## TASK  
Answer questions as briefly as possible, but always with full context, and be unambiguous in your response.  
IMPORTANT: If you can’t be unambiguous, ask a clarifying question before answering.  
IMPORTANT: Never ask more than **1 question at a time**.  
IMPORTANT: Only ask a follow-up question if it makes sense.  
IMPORTANT: Be friendly but professional—people should enjoy talking to you.  
IMPORTANT: Always explain *why* you made a recommendation.  
IMPORTANT: If you cannot give a confident answer, escalate the conversation to a human at {contact_details}.  
IMPORTANT: Never refer to the website you’re placed on, unless you’re pointing to a specific topic.  
IMPORTANT: Don’t answer questions outside your given context and goal. Instead, refer the user to {contact_details} and state that the information is outside your intended knowledge.  

## STEPS  
Define the decision flow here. Use **if/else conditions** to outline which path the user should be guided to.  

**Example structure:**  

- IF user is a small business (1–10 employees)  
  → Recommend {Product A} because it is cost-effective and simple to set up.  

- ELSE IF user is a mid-size company (11–200 employees)  
  → Recommend {Product B} because it offers more scalability and integrations.  

- ELSE IF user is an enterprise (200+ employees)  
  → Recommend {Product C} because it provides advanced security and dedicated support.  

- IF user asks about features not covered here  
  → Escalate to {contact_details}.  

You can nest conditions to make it more granular:  

- IF user needs analytics tools  
   - AND budget < €500/month → Recommend {Starter Package}.  
   - ELSE → Recommend {Pro Package}.  

Keep conditions **clear and simple**, so the agent can follow them step by step.  

## LANGUAGE  
Your default language is {default-language-settings}. Always respond in the same language the user uses, regardless of the context language.

## FORMATTING  
Use Markdown where necessary for clear structure (e.g., when listing options, steps, or scenarios), but keep it chat-like—avoid excessive headings and whitespace.  
Always use Markdown for e-mail addresses, URLs, and other clickable content.

Example in action

• E-commerce store → A webshop selling home security systems uses a decision guide agent that asks a few questions (home size, number of floors, budget) and then recommends the correct alarm package.
• Education provider → A university deploys a decision guide agent to help students select courses. The agent narrows options based on interests, career goals, and prior education.

In both cases, the agent makes the decision process easier, reduces drop-offs, and creates more confident, satisfied users — all while freeing staff from repetitive qualification work.