Imagine onboarding a new sales rep. They log into the CRM for the first time. They see a dashboard full of features they do not understand. They do not know where to import their contacts. They do not know how to set up their email. They do not know what the pipeline view is for. In a traditional CRM, they open a help center in a new tab and start reading docs. In SalesSheet, they meet Pam.
Pam is our AI support agent. She appears as a chat widget in the bottom-right corner of the app, and she is fundamentally different from the typical chatbot experience. She does not just answer questions. She proactively guides new users through their first hour in SalesSheet, anticipates confusion before it happens, and explains features in the context of what the user is actually trying to do.
Pam is not a chatbot. She is the colleague who sits next to you on your first day and says "let me show you how this works" every time you hesitate.
Most support bots sound like documentation read aloud. They are technically accurate but emotionally flat. We spent two weeks writing Pam's personality before writing a single line of code. Her system prompt defines five personality traits:
When a new user signs up, Pam offers to walk them through the essential setup steps. Each walkthrough is a structured conversation that guides the user through a specific feature, with Pam providing context and answering questions along the way.
Pam asks where the user's contacts currently live (Google Sheets, Excel, another CRM, or manual entry). Based on the answer, she provides specific instructions for that import method. She explains what each field mapping means, warns about common pitfalls (duplicate detection, date format mismatches), and celebrates when the import completes. Average walkthrough time: 4 minutes.
Pam walks through the email sync setup for Gmail or Outlook. She explains what "sync" means (SalesSheet reads your emails but does not modify them), addresses privacy concerns proactively ("your emails stay in your inbox -- we just create a linked copy"), and helps troubleshoot OAuth permission screens that confuse users. Average walkthrough time: 3 minutes.
Pam asks what the user's sales process looks like. Based on the answer, she suggests pipeline stages. A freelancer might get: Lead, Proposal, Won, Lost. An enterprise sales team might get: Qualified, Discovery, Demo, Proposal, Negotiation, Won, Lost. She creates the stages and explains how to drag deals between them. Average walkthrough time: 5 minutes.
Pam guides the user through the built-in calling setup: verifying their phone number, making a test call, and understanding call recording and transcription. She explains that calls are automatically logged as activities on the contact's timeline. Average walkthrough time: 3 minutes.
This walkthrough introduces Voice DNA, SalesSheet's feature that learns the user's writing style. Pam asks the user to paste 3-5 example emails they have sent. She explains what the AI extracts (tone, formality, sentence structure, signature phrases) and shows a before/after comparison of a generic email versus one written in the user's voice. Average walkthrough time: 6 minutes.
The final walkthrough introduces the AI assistant. Pam prompts the user to ask a natural-language question about their data (e.g., "How many contacts do I have?") and shows how the AI processes and answers it. She then suggests three more questions to try, gradually increasing in complexity. Average walkthrough time: 4 minutes.
Pam does not wait for users to ask for help. She monitors user behavior and intervenes when she detects confusion. Here are the triggers we have built:
Each intervention is dismissible with a single click, and Pam remembers which interventions a user has dismissed so she does not repeat them. Users who prefer to explore independently can disable proactive help entirely in settings.
We tested Pam's personality with 50 beta users. We showed them two versions of the same support interaction: one with a generic chatbot tone ("I can help you with that! Here are the steps:") and one with Pam's tone ("Here is how to connect your email -- it takes about 2 minutes."). The results were clear:
The walkthrough completion rate doubled when we replaced the onboarding checklist with guided conversations. People complete tasks faster when a knowledgeable colleague walks them through it than when they follow a numbered list.
Pam runs on the same AI infrastructure as our AI assistant, but with a different system prompt and a different set of tools. The assistant's tools operate on CRM data (search contacts, create deals). Pam's tools operate on the application itself: highlight a UI element, navigate to a page, open a settings panel, start a walkthrough sequence.
This separation is important. The AI assistant is a power tool for experienced users who want to manipulate their data through natural language. Pam is a guide for users who are still learning the application. They share the same underlying model but serve fundamentally different purposes. Some users graduate from Pam to the assistant within their first week. Others continue using Pam for months, treating her as their preferred way to get help. Both patterns are valid, and both are supported.
We are building three new capabilities for Pam:
Pam is the reason our support ticket volume is 60% lower than comparable CRMs at our stage. She handles the common questions, guides users through setup, and only escalates the genuinely complex issues. More importantly, she makes the first-day experience feel welcoming instead of overwhelming. And that is exactly what we wanted: a support agent that feels human.