Every sales team has questions that no existing report answers. "Which of my accounts had no activity in the last 30 days?" "What is the total open pipeline per owner?" "Compare won and lost deals this quarter." The data is in Salesforce, but getting it out means building a report, adjusting filters, or asking the one admin who knows the report builder.
A natural-language chat inside Salesforce removes that step. You type the question and get the answer as a table, cards, a chart or a single number. This article explains how that works in AI Engine, and why it does not bypass your security model.
What you can ask
The Sales Agent chat understands everyday questions about the objects your admin enabled. For example:
- "Show my open opportunities above 50k"
- "How many leads did I get this week?"
- "Which of my accounts had no activity in the last 30 days?"
- "Total amount of open opportunities per owner"
- "Compare won and lost opportunities this quarter"
Answers start with the key numbers and add a sentence on what stands out. When the data shows a real problem, such as overdue items or deals past their close date, the answer names it and suggests one next step.
You can follow up without repeating yourself. After "show my open opportunities", type "only the ones closing this month". The chat keeps the last few messages as context.
How a question becomes an answer
Behind the scenes, three things happen. Two of them are AI requests:
- The question is turned into a Salesforce query. The AI is given the list of objects and fields your admin enabled, and builds a query from your words.
- The query runs in Salesforce, with your permissions. It uses your sharing rules, object permissions and field-level security, so the result contains only records and fields you could open yourself.
- The AI writes the answer from the query result and picks a format: a table, cards, a chart or a single number.
Because the data comes from a real query and not from the model's memory, the numbers match what you would see in Salesforce. You can click Show all to see the full result or export it to CSV or Excel.
Why it stays secure
Letting an AI "talk to your CRM" sounds risky. Three design decisions keep it safe:
- Admin allowlist. The AI can only query the objects and fields an admin selected in AI Engine's Agent Objects settings, plus tasks, events and users. Anything else is out of reach, whatever the user types.
- User permissions, not system access. Queries run as the user. A rep cannot see a colleague's private opportunities through the chat if they cannot see them in Salesforce.
- Asking never changes data. Records, tasks, emails and files are created only when a user clicks a button that does it, such as + Task or ✉ Send. A question is read-only.
On top of that, personal data such as email addresses and phone numbers can be masked before anything is sent to the AI provider. AI Engine does not store the full text of questions and answers; the activity log keeps usage metadata, plus a short error message when a call fails.
Tips for better answers
- Be specific. "Open opportunities in stage Negotiation closing this month" works better than "how is my pipeline".
- One object at a time. Start with one object, then ask a follow-up.
- Name the time frame. "This week", "last quarter" and "in the last 90 days" all work.
- Check before you act. The AI can misread a question. Check anything you are about to send to a customer.
If you want information from the internet instead, ask for it directly: "news about Example Ltd" or, on a lead, "research this lead". The chat decides when a question needs a web search instead of a Salesforce query.
Reports still matter
A chat does not replace dashboards your management reviews every week. It covers the long tail: the one-off questions that come up before a call, during a pipeline review or on a Friday afternoon, and that nobody will build a report for. Those questions are where most CRM data goes unused.
Where the chat lives
Admins decide where the Sales Agent panel appears: on record pages such as Account, Lead, Opportunity, Contact and Case, in the utility bar, or on an app page. On a record page, the chat knows which record you are looking at, so "summarise the last activities on this account" or "who is the decision maker here?" work without naming the record.
Summary
- Type a question and get the answer from live Salesforce data as a table, cards, a chart or a number.
- The AI builds a real Salesforce query from the objects your admin allowed, and it runs with your own permissions.
- Asking is read-only. Nothing is created until you click a button.
- It covers the questions no report was built for.
Want to try it on your own data? AI Engine is a native Salesforce package that works with your own OpenAI, Claude or Azure OpenAI key.
See AI Engine on your own data
Native Salesforce package. Your own OpenAI, Claude or Azure key. 5-day free trial, no credit card.
Book a 30-min demo