Even Radiologists Don’t Catch Everything
- Mar 27
- 3 min read
One of my very good friends in Houston is a renowned radiologist, Dr. Walid Adham. A few years ago, he joined Voxel Systems as their CEO. At Voxel, Dr. Adham is focused on creating an AI-powered software product, now patented, with an extensive dataset used to effectively read MRIs to predict and diagnose certain brain disorders, including cognitive decline. His work and very detailed conversations with him brought this unique comparison to my attention.
Over the last year, I’ve spoken with numerous companies about payment accuracy. In practically every case, I find the current solution involves several AP clerks manually reviewing every invoice to ensure accuracy of payment. Aside from the obvious labor cost involved with this approach, on the surface it sounds reasonable to have people check the numbers and interpret contract language to make the right call. However, research and observations in the radiology space indicate this approach is likely creating a false sense of security hinged on the expectation that AP clerks perform at a 0% error rate.
What Radiology Teaches Us About Human Error
Radiology is a profession built on expert review, with highly trained physicians interpreting complex data under time pressure, at scale, with serious consequences for mistakes. Sound familiar? Even in this environment, the error rate is not zero.
In a widely cited review in the American Journal of Roentgenology, Dr. J. Nathan Itri notes that the error rate for practicing diagnostic radiologists generally ranges from 3% to 6%. Dr. Adrian Brady makes a similar point in Insights into Imaging, describing day-to-day error/discrepancy rates on the order of 3–5% of studies reported. Many studies, both new and old, go on to confirm these findings.
Now consider that radiologists spend four years in undergraduate education, four years in medical school, another few years in residency, and often go on to do additional fellowship work. These are elite, expert-level professionals with extensive education, training, and experience. Even with all of this, radiology still experiences error rates as high as 30% in certain subspecialties like chest radiographs, according to Dr. Leo Henry Garland’s classic work with the Radiological Society of North America.
How Does This Happen?
Given how critically important healthcare is, what is the cause of this? It’s actually very straightforward, according to the research. Nobody’s perfect. People make mistakes as a result of fatigue, cognitive bias, distractions, workload, pattern recognition errors, and simple oversight, with all of this playing a role. The more images they review, the more opportunities there are for small misses.
AP clerks are working as hard and as fast as they can every single day. Companies certainly don't require, expect, or provide a level of training anywhere near what we see with radiologists, yet AP clerks are expected to perform even better, at a near 0% error rate.
Most likely, at certain oil and gas companies, management recognizes that manual reviews are indeed not catching everything, as evidenced by the audits being performed at the end of the year. Third-party firms are engaged to review contracts, invoices, and supporting documentation in an effort to find overpayments. With this approach, however, issues still exist. The obvious ones are cost and timing, which impact cost of capital. Add to that the counterparty risk, and no service provider wants recognized revenue (and likely spent cash) from a year ago clawed back due to a billing mistake that should have been caught before payment was made.
The Better Approach
In radiology, the answer hasn’t been to simply work longer hours or add more people. Instead, radiologists are upgrading their tools with AI-assisted image analysis and workflow improvement. That’s the kind of approach Dr. Adham is taking at Voxel Systems, building technology that radically improves read accuracy and speed.
Invoice compliance reviews can now follow a similar principle. When companies rely on adding more people to manually check invoices as they grow, it acts as a self-imposed governor on the business’s ability to scale. More importantly, however, it’s not catching all of the issues. The better approach involves using AI-powered software like Prolaus to handle the overwhelming majority of compliance checks and leave the edge cases and judgment calls to people.
—---------------