Here's what the build actually looks like
1. Read the ticket
Ishaan starts with the customer's question exactly as they wrote it, straight from Atlas. Nothing gets shortened or rephrased at this point; it works from what the customer actually typed.
2. Pull the account context
Next, it gathers the customer's plan, MRR, credit usage, renewal date, and health score from Chargebee and Vitally, along with any past tickets from that same account.
3. Find the right help article, and note where it came from
Ishaan searches the help center for articles that match what the customer actually means, not just the words they used.
A customer might describe a problem in completely different language than the docs do, so matching on exact wording alone would miss the article that actually answers it.
Every article it pulls gets attached to the draft, so a CSM can check the source in one click instead of trusting it blind.
4. Sort the ticket by type and urgency
Every ticket gets sorted into a category (a how-to, a billing issue, a feature request, or something that needs a human) and a rough sense of urgency.
This is also where the line gets drawn on what it's allowed to touch. If something feels sensitive, like a refund or an angry customer, it flags it and hands it off.
5. Check its own draft before sending it forward
Before the reply goes anywhere, Ishaan runs an initial check on its own draft, comparing it against the account data and the source article it pulled.
This catches cases where the two don't actually line up, a mismatched plan detail, or a claim the source article doesn't actually support.
6. Post it to Slack for review
The draft lands in Slack, where a CSM reads it, edits it if needed, and approves it before sending it to the customer.
