Promise Okeke is CEO of NovoPath, a premier Laboratory Information System (LIS) for AP, Molecular and Veterinary labs.
The conversation about artificial intelligence (AI) in pathology is almost entirely about models. Foundation models are trained on millions of slides. Algorithms flag mitoses, grade tumors or quantify a stain in seconds. It is an exciting moment, and the science is real.
But an algorithm cannot read a piece of glass. Before any laboratory gets value from AI, it has to become digital. That means doing the unglamorous infrastructure work first.
As the CEO of a company that specializes in digital pathology solutions, I’ve become well acquainted with the work that turns a physical operation built on paper and glass into a digital system an algorithm can plug into.
The Scanner Is Just The Starting Point
Most labs assume going digital starts with choosing a scanner. The scanner matters, but it is the narrowest decision in the project. The harder work, making sure every image is tied to the right patient, happens before a slide reaches it.
A lab has to know, with certainty and without anyone retyping a label, which case, block and slide an image belongs to. A specimen passes through several hands between accessioning and the microscope, and every handoff where someone reads a label and writes it again is a chance for a mismatch.
On glass, a pathologist holds the physical slide and sees its handwritten markings and the tray it arrived in. In a digital workflow, that context disappears. The scanner reads the barcode, and the barcode becomes the slide’s identity. If the label is wrong, the image becomes confidently attached to the wrong patient.
The fix is less exciting than any scanner spec: labels generated from a single source of truth and printed at the moment each block or slide is created, not batch-printed and applied later. It is tedious work, and it’s the first thing a lab should get right.
It is also the thing AI depends on most. An algorithm is only as reliable as the specimen identity chain feeding it. A model that produces a brilliant result on the wrong patient’s slide is not a breakthrough. It is a liability. Get the identity chain wrong and every downstream promise, including AI, inherits the error.
The Open Loop Is Where Value Leaks Out
Labs that go digital tend to see value first at the pathologist.
Remote sign-out lets a specialist read a case from anywhere. Consults and tumor boards happen without shipping glass across the country. That value is real, and it is the right place to start, because it changes a pathologist’s working life rather than just their throughput.
But most labs stall, and the reason is structural. The case goes out to a viewer and never comes back into the system of record. The diagnosis gets entered somewhere else. Images are attached by hand. The pathologist ends up carrying a case between two systems in their head.
When the case does not come back, the lab loses the ability to see itself. It cannot know whether a case is complete, who is holding what, or how long anything took.
Those questions do not stop being asked. They get answered by people walking around and building spreadsheets instead of by software, and that labor never shows up in a business case as a line item. It shows up as headcount.
This is also where AI can fail to deliver. An algorithm that runs on a slide, but has nowhere to send its result, is orphaned. AI output is only useful if it flows back into the case record, where the pathologist’s decision is documented and the lab acts on it. That takes interoperability; the LIS, the image system and the AI have to exchange results in both directions.
A closed loop is not a nice feature to add later. It is the precondition for everything, including the AI you are excited about.
Reframe What You Are Actually Buying
Leaders often expect digitization, and by extension AI, to cut cost per case. It usually does not, at least not at first. Glass still gets cut and stained. Scanning adds a step and adds expense. A business case built on cost reduction tends to fall apart at the second budget review.
The honest returns are capacity and coverage. Pathologists who can work across sites and state licenses. Consult cycles that no longer wait on a courier. Recruiting and locum spend avoided in a brutally tight labor market. AI extends exactly that logic. Its real promise is augmenting scarce expertise, not replacing it. The leaders who frame AI as a way to reduce headcount will lose the room. The ones who frame it as a way to extend the experts they already have, and cannot hire more of, will be right.
The Edge Is The Boring Foundation
The tempting move is to wait for the algorithms to mature and buy them when they are ready. But you cannot bolt AI onto an analog lab. The infrastructure that makes AI usable takes years to get right: digital specimen identity, a closed loop back to the system of record and a workflow designed around images rather than paper.
That is precisely why it is the edge. The competitive advantage in pathology over the next several years will not go to whoever adopts AI first. It will go to whoever built the foundation that lets AI work at all. Much of that foundation is plumbing: workflow interoperability that carries an AI result into the case, the report and the bill without anyone retyping it. It isn’t the part anyone demos, but nothing else works without it.
The best time to start that groundwork is before the algorithms you want are clinically ready, because when they arrive, you do not want to be starting from zero.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?