Series Keeping Up With AbbaDox · AI Operations

Good AI Is Invisible AI

Why a working system shows you only its mistakes, and what that does to a team that was never told to expect it.

With Nick Avossa, VP of AI Operations Published

For operations and imaging leaders, front-line managers, and teams preparing for a go-live.

Read the recap
Jump to a chapter8 chapters
Episode highlights

What you’ll learn in this episode

Set a baseline before judging a tool, because with nothing to measure against nothing will look good enough.
Tell a deterministic rule apart from a probabilistic one, and give each the work it suits.
Prepare staff for a queue that now holds only the exceptions, so working software does not read as broken.
Keep a person in the loop wherever the answer has to be certain, instead of forcing AI into it.
Design around the whole workflow, so today's fix does not become next year's inherited workaround.
From the talk, 22:40

Why you only ever see the errors

Nick on what he hears after go-live: the queue holds only errors, so it “just keeps messing up.”

Episode context

AI gets judged by fear before it gets judged by results

In radiology operations, AI often enters the conversation carrying baggage it did not create. Teams worry about job loss, black-box decisions, bad data, fragmentation, and failure rates before they have a clear baseline for the work already happening today. That makes adoption harder, even when the current process is overloaded, inconsistent, or already creating avoidable errors.

Nick Avossa has seen that pattern up close. As VP of AI Operations at AbbaDox, he works with leaders and front-line teams trying to modernize workflows while carrying years of habits, workarounds, and skepticism into the conversation. His point is not that fear is irrational. It is that fear alone is a poor benchmark for deciding what should change.

As Nick puts it, a system doing its job is the one you stop noticing, and that is exactly what makes it hard to trust.

  1. Know what the work costs today, done by hand.
  2. When AI handles the routine, the routine disappears from view.
  3. Only the exceptions surface, so a working system can look broken.
  4. Count what was handled, not only what surfaced, and the invisible work shows up.
About this episode

Measure the workflow, then make the logic visible

In this episode of Keeping Up With AbbaDox, VP of AI Operations Nick Avossa argues that operational AI works best when it is grounded in context, transparency, and comparison. Teams need to know what the current process costs in time, throughput, and accuracy before they decide what good enough looks like.

They also need to understand where AI fits, where it does not, and when human review should stay in the loop. Real adoption happens when leaders show the data, prepare staff for what will change, and focus AI on the parts of the workflow where speed, consistency, and scale matter most.

Nick Avossa, VP of AI Operations at AbbaDox
if it's doing its job, it is invisible, and all you surface are the errors.
Nick Avossa VP of AI Operations, AbbaDox

Runs AI operations at AbbaDox, working with centers on the floor rather than from a slide.

What that changes about adoption
“If you do not prepare your staff for the idea that the things on their list are only errors now, they will give you the perception that it does not work.”

Nick's argument is that the tool is rarely the hard part. What decides adoption is a measured baseline to compare against, and staff who were told in advance that their queue would fill with exceptions only, so a system that is working does not arrive looking like a system that is broken.

Operational AI at AbbaDox

How AbbaDox introduces AI into an imaging operation

Six things that decide whether it sticks. Select one to explore it.

Put it to work

Take it to your team

Three questions for your next leadership meeting, each with the moment in the episode that answers it.

  1. What does the work cost you today?

    Measure accuracy and throughput now, then hold AI to the same benchmark you hold your staff to.

  2. Will your team know what an empty queue means?

    Once AI clears the routine work, the list holds only errors. Staff who are not told that will decide it does not work.

  3. Which tasks need an answer that is always right?

    Keep a person on those, and give AI the routine work, where its errors are more consistent and easier to fix.

Want these answered for your center?

See what AI you stop noticing looks like in your operation.

Watch the full talk, or bring your own numbers to the team that would be measuring against them.