Working with Agents

How to interact with Quotient's AI assistants

Agents are AI assistants powered by large language models. Agents are designed to be easy to work with. Like a human coworker, you can talk to them in plain English and tell them exactly what you want them to do.

However, working with AI agents is different from working with human coworkers in a few key ways. The most important concept to understand is context.

Context and Tools

"Context" refers to the information that is available to an agent. Unlike humans, agents start each conversation fresh - they don't remember what you told them yesterday. They have to be explicitly programmed with the information they need to do their jobs.

This means you should be explicit about what you want and provide all necessary context in each conversation. Knowing what context agents have and don't have will help you interact with them.

When you talk to an agent in Quotient, their context will automatically include…

• Information about your brand stored in the Knowledge Store
• The conversation history within that chat thread
• An understanding of the Quotient platform itself
• Specialized knowledge about that agent's area of marketing (e.g. the Blog Agent is an expert in writing SEO-optimized content)

However, the agent's context does not include...

• Knowledge of current events, like today's front page news or what's currently trending in your industry
• Any information about your brand or business that is not reflected in the Knowledge Store
• Memory of past interactions with you (if you tell the Blog Agent about a new product feature in one conversation, it won't remember that information in future conversations unless you add it to the Knowledge Store)

To get additional information, agents can use tools to look things up. Tools allow agents to interact with the outside world, including the Quotient platform, the open internet, and other integrated systems.

Agents can use tools to find information to add to their existing context, such as…

• Searching the web and visiting web pages
• Looking up a person or company from your Quotient audience
• Looking up all currently active campaigns in Quotient

In addition to looking up information, tools can also perform actions, such as…

• Creating a new campaign
• Editing a blog
• Deleting an email broadcast

Finding the Right Agent

Each agent within Quotient has a different job and specialized context and tools for doing that job. That means that it's important to talk to the right agent for the task you're trying to accomplish.

This is no different from working with human coworkers. If you want a quarterly financial report, you need to go to the finance department, not the marketing department. The same principle applies in Quotient.

Here's a quick reminder of the different agents in Quotient and what each of them can help you with:

• Brand Agent: Manages your Knowledge Store and brand information
• Campaign Agent: Plans campaign calendar and strategy
• Blog Agent: Writes and researches blog content
• Email Agent: Creates email broadcasts and templates
• Flow Agent: Builds automated marketing workflows
• Event Agent: Plans webinars, trade shows, and events
• Audience Agent: Creates customer segments and lists
• Design Agent: Generates on-brand images and visuals
• Web Agent: Builds landing pages and forms
• Help Agent: Answers questions about Quotient

@ Mentioning Agents

To control which agent you're talking to, you can simply @ mention them, just like you would on Slack or another chat app.

This will guarantee that the correct agent responds. If you @ mention multiple agents, the last one you mentioned will respond. If you don't @ mention anyone, Quotient will try to guess which agent you were looking for based on context clues, like what page you're on and which agent responded last.

If you accidentally @ mention the wrong agent, don't worry. Agents can route messages to each other, so they will typically re-route you to the correct agent if you get it wrong. For example if you ask the Blog Agent to write an email, it will hand you off to the Email Agent.

If you don't know which agent is the right one to talk to for your task, simply ask the Help Agent.

@ Mentioning Objects

In addition to mentioning agents, you can also @ mention objects in Quotient, such as Campaigns, Email Broadcasts, Blogs, Segments, and Email Templates.

Mentioning objects will automatically include all of the data about that object into the agent's context window. You can think of this like sharing a URL with a coworker. Your coworker can click into the URL, give it a quick read, and understand what you're talking about. But without the URL, they might be confused and not understand the context.

Here are some examples of when you might @ mention an object:

• If you want the Flow agent to use a specific email template as part of a flow
• If you want the Email agent to send an email to a specific segment
• If you want the Blog agent to use a specific asset as the thumbnail image for a blog

Managing Threads

Each conversation in Quotient is called a “thread”. Within a given thread, agents will recall the entire conversation history, but they won't recall conversations from other threads.

When a thread gets too large or covers too many topics, it tends to confuse or overwhelm agents. Long threads can be overwhelming for humans too. (No one likes looking at a Slack thread with 150 responses, or an email chain that's 80 replies long.)

To get the best performance out of agents, it's best to keep threads scoped to a single task or set of related tasks. A good rule of thumb is to create one thread per deliverable - i.e. a single thread for each blog, campaign, email broadcast, etc.

Sometimes it makes sense to group related deliverables into a single thread. For example, you might be working on a blog post as well as an email broadcast announcing the new post to your subscribers. In this case, it might make sense to create both deliverables in the same thread.

On the other hand, if you have a completely new, unrelated request for an agent, it's best to start a fresh thread to begin working on it.

Memories

Although agents don't naturally remember past interactions the way humans do, they are able to record memories which will be stored as context for future interactions. This is a helpful way for agents to learn your business's particular preferences and workflows. If you want an agent to remember something for future interactions, simply ask it to “please remember this going forward”.

Common examples of memories include…

• Which email templates the Email Agent should use as a starting point when creating new broadcasts
• Which segments or lists should be used in flows and email broadcasts
• Preferences about campaign management (e.g. “campaigns should always contain one email broadcast per blog post”)

Note: Memories and Knowledge Documents serve a similar purpose and are mostly interchangeable. The key difference is that only the Brand Agent can create Knowledge Documents, while any agent can record memories. Typically memories reflect workflow preferences, whereas Knowledge Documents contain core information about the brand.