Four-agent AI assistant for HR
A live assistant that answers recruitment, payroll, onboarding, and compliance questions inside the HRIS itself.
Context
By this point, Repute already had automations running quietly in the background, but people still had to go find their own answers. If an HR person wanted to know whether a candidate's background check had come back, why a payroll run hadn't gone through, what the leave policy was for someone still on probation, or whether a compliance filing was overdue, they had to hunt through the right screen themselves, or open a support ticket and wait. Recruitment, payroll, onboarding, and compliance are four very different parts of HR, and each one needed its own separate back and forth before someone actually got an answer.
What was needed
What we needed was one assistant that could actually handle all four areas correctly, not a chatbot that sounded confident and got the details wrong. In HR and payroll, a wrong answer isn't just annoying. Telling someone the wrong leave balance or missing a compliance deadline has real consequences. So this had to be something people could trust, not just something that looked impressive in a demo.
Work I did
I shipped a live assistant made up of four specialist agents, one each for recruitment, payroll checks, onboarding, and compliance. Think of it as four subject matter experts on call instead of one generalist trying to answer everything and occasionally guessing outside its lane. I worked with our CTO to define the orchestration layer, which is really just the traffic control system that reads a person's question and sends it to the right specialist agent. I made sure every answer was grounded in Repute's actual live data instead of the assistant making something up, and I set accuracy benchmarks with a human checking its answers before we let it go live for customers.
What helped
I pushed to get the accuracy benchmarks in place before we scaled the assistant up to more customers, instead of shipping fast and hoping it held. It slowed the launch down by a few weeks, but it meant we caught the cases where the assistant was wrong before customers did, not after.
Results
The assistant has been live across all four areas for a while now, and it's helped grow HRIS annual recurring revenue by 2X.
Want to discuss this in person?
Send me a note about what you read here and I'll get back to you myself.