Agent Case Study · Fintech
Instant answers,
with receipts
A fintech scale-up doubled headcount in a year and its IT and HR teams became human search engines — the same policy questions, all day, every day. We deployed an employee helpdesk agent in Slack that answers from their approved docs with citations, respects role permissions, and escalates the rest with context.
Industry
Fintech (payments)
Company size
~300 employees, hybrid
Region
North America
Time to live
4 weeks including security review
Client details anonymized under NDA — industry, size band, and timeline are accurate; identifying details are not shared.
Where they started
The challenge
Growth broke the informal answer network: new hires didn't know who to ask, so they asked IT and HR everything — expense thresholds, VPN setup, parental-leave policy, whether the office closes for a holiday. Simple tickets sat behind complex ones; a password reset could wait a day while work blocked on it.
Being a regulated fintech raised the stakes: answers about compliance policy had to be right and traceable, and role-restricted information — compensation bands, disciplinary process — had to stay restricted. A chatbot that guessed was worse than no chatbot.
The build
What we built
01Approved sources only
The agent answers exclusively from their Notion wiki, policy PDFs, and curated past tickets — every answer carries a citation to the source document, and 'I don't know' triggers escalation, never a guess.
02Native to Slack
No new portal. Employees ask in a DM or the #help channel and get answers in-thread within seconds.
03Permission-aware by design
Answers respect their existing access controls: managers see manager-level policy, everyone else sees the public version, and designated topics always route to HR humans.
04Actions, then escalation with context
Password resets, access requests, and ticket creation happen end to end through integrations; everything else lands with the right owner, conversation attached.
Ninety days later
The results
The unanswered-question log became the sleeper feature: every 'I don't know' exposes a documentation gap, so the knowledge base improves weekly instead of rotting. Resolution settled at 68% — inside the 60–80% band we publish — and the compliance team, initially the most skeptical, now points auditors at the citation trail.
Want this outcome in your business?
Start with a free AI audit: we map your workflows and show you where this agent pays for itself — before you commit to anything.