Here's what happened — and how. One ticket. One tech approval. One KB article. The bot grounded itself on that article when the next similar ticket arrived 18 hours later, and proposed the exact same runbook with 97% confidence.
A customer emails support: their print queue is stuck and an invoice is jammed. The bot's pipeline fires within seconds of the inbound webhook. It classifies the ticket type, searches the KB for grounding context, and drafts a reply — but at this moment, no KB article on print spooler issues exists yet. The bot's confidence sits at 82% — high enough to escalate with a quality draft, but not high enough to ground the response in known, approved guidance.
DTC's senior tech works the ticket: stops the Print Spooler service, clears the stuck files, restarts the service, sends a test print. They click Mark Done in Live Feed. The dashboard surfaces the 3-branch close-out loop: did you follow the bot's suggested resolution? They pick "Yes, used as-is" — the bot's draft already mirrored what they did. The runbook is captured automatically, drafted into a structured KB article, and lands in the Autopilot Queue for tech review.
Bot's runbook was correct. Promote it directly — one click.
Bot was on the right track but needed adjustments. Annotate inline.
Open Train Bot Q&A flow. Capture what actually worked.
C:\Windows\System32\spool\PRINTERS\.The next morning, a different customer emails about a stuck print queue. The bot's pipeline fires again — but this time, the KB search returns the approved "Clearing Stuck Print Queue" runbook. The bot grounds its reply on the verbatim steps the senior tech approved yesterday. Confidence jumps from 82% to 97%. The proposed reply pulls the exact action sequence from the article, including the constraint the tech respected (the customer mentioned they'd need someone to remote in — the bot drafted "via remote session" language instead of forwarding the script).
DTC's KB now holds 1,075 articles grounded for production lookup. The bot has run 1,722 inferences across 143 tickets since shadow mode started. Every Mark Done with "Yes, used as-is" is one more article in the corpus. Every Mark Done with "Yes, with edits" trains the bot to draft tighter next time. Every junior tech who joins later inherits the entire corpus on day one — they ramp like seniors because they're working from senior-approved playbooks.
The trust ladder makes this safe. At L2 Drafter (the default for new customers), every reply waits for a human approval. At L3 Coverage, the bot keeps customers warm when the tech is buried — without ever resolving on its own. Only at L4 Co-pilot does the bot start auto-sending on the ticket types you trust, with a circuit breaker that reverts to dry-run after 2 consecutive overrides. You set the pace. The bot earns its way up.
30-day free Starter trial for solo MSPs. Pro waitlist for growth-stage teams who want Coverage + Co-pilot from day one.