AI at the Edges: Meet The Auxilius Agents

‍

In a recent article on AI in clinical finance, we covered where AI belongs in trial finance, why complex clinical grants reconciliation matching is the last place it should live, and seven questions to ask any vendor. Now, we’re excited to share with you exactly how we’re building in accordance with these principles at Auxilius.

‍

We're rolling out two families of agents, one on each side of a deterministic core.

‍

Assist Agents work at the beginning, where the job is parameter completion. They take the manual, tedious tasks at the front of the process (mapping lines, filling in inputs, incorporating a new file, etc.) and turn hours of work into minutes of review. Assist proposes settings, matches, and rules that you approve; nothing runs without you.

‍

IQ Agents work at the end, where the job is sensemaking. They read what the engine produced and surface what's most meaningful. They identify what moved, delve into why, and highlight areas that deserve your attention. Every statement is grounded in the actual calculation, so your time goes to judgment instead of hunting.

‍

In between sits the engine, deterministic and fully traceable, which does exactly what it did yesterday and will do exactly the same thing tomorrow.

‍

Assist Agents: give the tedious hours back

‍

The front of trial finance is full of piecemeal, manual work. An Assist takes it on with two advantages: speed on the tasks you could do yourself, and knowledge on the ones you shouldn’t have to, such as decoding a site’s procedure descriptions. Modern LLMs generally know clinical terminology, but knowing the terminology only gets you a plausible guess. What makes a proposal specific and verifiable is that it maps onto the real taxonomies we've built at Auxilius across protocols, visit schedules, procedures and CPT-level data, and more.

‍

The first is Forecast Assist: AI that reads your study setup and proposes forecast configurations, grounded in our forecasting knowledge base, that you review and approve before anything runs. It takes a “let me go bug my clinical lead for two hours on when these things happen in clinical trials” to an educated starting point.

From here, we’re building an Assist anywhere our customers face a blank page or a messy file:
‍

• ‍Vendor Estimate Assist reads the monthly or quarterly estimate your vendor sends and proposes how every line maps back to the budget you already have in the system, including when the naming has drifted. ‍
• Change Order Assist does the same for change orders: it reads the new change order, identifies missing or new items and activities, and proposes how each line maps back to your existing budget, so the engine can compute exactly what changed.‍
• Clinical Data Assist proposes field and visit mappings when a new study's clinical data arrives, so that your clinical data pipeline is up and running in days instead of weeks.‍
• Site Payments Assist maps incoming site payment data and proposes the matching parameters for reconciliation, so the deterministic matcher starts with clean buckets instead of raw descriptions.

Everything an Assist proposes (and you approve) runs through the Auxilius engine. Our second agent family takes it from there.

‍

IQ Agents: point your attention where it matters

‍

The engine can produce more output than anyone has time to read, which is where our sensemaking agents come into play. Every IQ generates the same three things: a narrative of what happened, the exposure, and recommended next steps.

‍

The first is Close IQ, which reads your near-completed close and drafts a narrative: what moved and why. It digs into the flux and double-clicks into the data behind it to trace numbers down to their drivers, with every statement grounded in the calculation. Your close still runs on deterministic, auditable logic. Close IQ just helps you tell the story of what it found.

‍

The family grows the same way: an IQ belongs anywhere a run finishes and a decision follows.

• Reconciliation IQ reads your payments reconciliation run and tells you which exceptions matter and which are noise. It spots the patterns behind them, like the four sites that haven't billed in six months and what that's hiding in your accrual. It can draft the outreach to those sites and you can send directly from the Auxilius platform.
• Forecast IQ explains what changed between forecast versions and what's driving the delta, piecing together drivers, change orders, and enrollment shifts into one answer instead of six tabs.
• Audit IQ reads the audit trail, who changed what and when, and drafts a plain-language summary for the period you're reviewing. This agent is only possible because our engine keeps a complete, structured lineage of every number. "Walk me through this change" stops meaning an hour of scrolling raw logs.

‍

Why we can build this

‍

Forecast Assist and Close IQ are the first of each family, not the last, and we have a lot more on our roadmap. An Assist belongs anywhere someone faces a blank page of inputs or a messy file, and an IQ belongs anywhere a model run finishes and a decision follows.

‍

The reason we can build agents like this is that we spent years building the two things they stand on:

‍

1. The first is the ontologies. We've mapped thousands of vendor budgets, tens of thousands of site budgets, and years of payments files and EDC exports, and out of all of it built the classification and mapping layer that sits at the core of the Auxilius data model. This allows us to turn a messy site description or a renamed vendor line into something with a real, verifiable meaning.
2. The second is the engines. These are purpose-built algorithms with the depth to capture the mechanics of clinical trials and the configurability to achieve precision on yours. Our investigator reconciliation model lets you tune matching parameters, group line items, and loosen or tighten criteria round by round. Our forecast model lets you set custom regional cost rules for future patients, and schedule invoiceable procedures and site fees either against the visit schedule or directly on the calendar.

‍

The ontologies are what let an agent understand your data, and the engines are what make its output worth trusting.

‍

Our mission at Auxilius has always been to make clinical trial and R&D costs more efficient, transparent, and predictable. Our agents are a natural next step in that mission, and we're so excited to get them into your hands.