Agents propose and interpret. The deterministic engine calculates. You decide. Every agent below sits on one side of that boundary — never in the middle.
Assist agents work at the beginning, turning a blank configuration screen into a proposal you review. IQ agents work at the end, turning a finished calculation into a narrative you can trust. In between, the numbers run on rules — not a model, not a guess — so the same inputs produce the same output every time.
Not a general assistant bolted onto your workflow — a specific proposal or a specific explanation, tied to a specific place in the process.
A new study starts with a protocol and not much else. Planning Assist reads it and proposes your initial study plan — cost methodology, procedure-level assumptions, phased across the trial — built from what your own studies have actually looked like.
Reads the shape of your trial and proposes a starting cost methodology and regional settings, drawn from patterns in your own forecasting history. You review every setting before anything runs.
Chases down outstanding vendor estimates — the reminders, the status checks — then reads what comes back and maps it onto your budget structure, line by line. You review and approve before it becomes part of your plan.
A new study's clinical data arrives from EDC with its own field names and its own visit structure, nothing pre-mapped. Clinical Data Assist reads it and proposes how it maps to your study's visit schedule and data fields.
Reads a change order and proposes the budget reallocation it implies — a diff against your current mapped budget, line by line. You approve the mapping before it touches a number.
Reconciliation only works if the matcher starts with clean, comparable data. Site Payments Assist maps incoming site payment data into clean buckets and proposes the matching parameters, so the deterministic matcher has what it needs from day one.
A forecast moves from one version to the next, and someone has to explain why. Forecast IQ pieces together what drove the delta — driver changes, change orders, enrollment shifts — with every claim pointing back to a number you can trace.
Most sites reconcile without a second look. Reconciliation IQ triages the ones that don't — filtering noise from what actually matters and surfacing the pattern: four sites haven't billed in six months, here's the accrual exposure.
Reads what came out of your completed close — including which lines moved and why — and drafts the narrative you need to defend it, every claim traceable back to the calculation.
Reads the audit trail — who changed what, when — and drafts a plain-language summary for the period you're reviewing, flagging the changes that were material. "Walk me through this change" stops meaning scrolling a raw log.
The names change. The shape doesn't — and that consistency is the point.
Faces a blank configuration screen or a messy inbound file — and needs a real input before it can propose anything.
Proposes a starting configuration, grounded in your own historical data.
Waits for you to review and approve. Nothing is set without you.
Reads the finished calculation — the close, the forecast, the reconciliation.
Drafts the explanation and translates it into a number — accrual impact, forecast impact. Every claim traces back to the calculation.
Waits for your review before anything goes out. It might draft the follow-up email to a site that hasn't billed in months — but you're the one who sends it.
Good questions for any AI vendor. We'll go first!
See exactly where each agent touches your process, and where it stops.