How a lean RecOps team ran its Soreon console by asking, not clicking - and gave each coordinator back 50 minutes a day
The coordinators lived inside Soreon's admin console all day - pulling up a candidate's applications, checking interview and scorecard status, spinning up positions, chasing what was still outstanding - and every task meant hunting through screens. Soreon's AI assistant let them do the same work by asking in plain language, and because it drives the product as the person themselves - inside the same permissions, with every change gated by a confirm card - ops trusted it to actually do things, not just answer. Coordinators got about 50 minutes of their day back, and a new hire was productive in days instead of weeks.
The challenge
The coordinators were the engine room of hiring. All day they lived inside Soreon's admin console - pulling up a candidate and their applications, checking which interviews still needed a scorecard, spinning up a new position for a role that just opened, chasing the handful of loops that were stuck. Every one of those was a small hunt: the right screen, the right filter, the right tab.
None of it was hard, but it added up. A coordinator lost real chunks of the day to navigation, context-switching between screens for answers that were one query away. And every new coordinator spent weeks just learning where everything lived before they were genuinely fast - so the team's throughput was capped by how quickly people learned the interface, not by the work itself.
What Soreon did
Soreon's AI assistant let the coordinators drive the console by asking. 'Which of Dana's applications are still open?', 'what interviews this week are missing a scorecard?', 'create a backend engineer position from this rubric' - the assistant streams back the answer or does the task, so a lookup that used to be four clicks became one sentence. It reads and acts inside the product, not in a separate report.
The reason ops trusted it to actually do things is how it is built. The assistant treats the model as an untrusted planner: every action it takes replays a real Soreon route as the coordinator themselves, so the same role-based permissions and org limits that guard the UI guard the assistant - it can never do something the person couldn't do by hand. And every data-changing action stops at a confirm card the coordinator has to approve before it runs, so nothing happens to real data without a human saying yes. Each person's conversations stay private to them, and the whole capability is off until an admin turns it on.
The results, by the numbers
- 50 min
- Coordinator lookup time saved per day
- 68%
- Routine console tasks now done by asking
- 4 days
- To ramp a new coordinator
- -58%
- 'How do I...' questions to the ops lead
Ask, don't click
The shift people felt first was that the console stopped being a place you navigate and became a place you talk to. A coordinator clearing the morning's queue no longer opened five tabs to answer 'what still needs attention today' - they asked, read the list, and moved on. The small, constant tax of finding the right screen mostly disappeared.
Because the assistant streams its response like a chat, it also fit the way coordinators already worked - a running conversation alongside the task, not a context-switch into a reporting tool. The routine lookups and status checks that filled the day, the ones nobody enjoyed, were the first to move over to asking.
“My coordinators used to know the console by muscle memory - which is another way of saying it took months to get fast. Now they just ask for what they need. The work that used to be navigation is a sentence, and a new hire is useful in their first week instead of their first month.”
An assistant we could trust to act
Plenty of tools can answer a question. What made this one usable for real operations is that the coordinators trusted it to do things - create a position, not just describe how to. That trust came from the guardrails, not from hoping the model behaved. Every action the assistant takes runs as the coordinator, through the same permissions and org limits as the normal interface, so it has no special powers and can never reach data the person couldn't reach themselves.
And nothing changes without a human. A request that would create or edit real data stops at a confirm card that spells out exactly what is about to happen, and the coordinator approves it before anything runs. Each person's conversations stay private to them - not readable by a manager or an admin - and the whole assistant is opt-in, off until the org deliberately switches it on. Those bounds are what turned it from a demo into something the team leaned on all day.
What changed
- Coordinators drive the console by asking, instead of hunting through screens
- Routine lookups, status checks, and quick creates moved to plain-language requests
- New coordinators became productive in days, not weeks
- Every assistant action ran within the coordinator's own permissions - no special access
- Every change was gated by a confirm card, and each person's conversations stayed private
“I used to keep a mental map of every screen. Now I just ask, and when it is about to actually change something it shows me a card to approve first - so I never worry it will do something behind my back. It got me close to an hour of my day back, and it is the part of my day I miss least.”
Turn your next interview into your competitive advantage.
See how Soreon gives your team evidence-linked follow-ups and editable scorecards - live, on your real interviews.
Book a Demo