Research note · October 2026
AI readiness is a question about evidence, not about tools
A firm is not ready to hand work to AI because it bought a model. It is ready when it can say which steps were seen, which were only reported, and which nobody observed.
01
Two methods, and a job that fits neither
Before an organization hands work to a system, someone has to establish how the work is done. The methods on offer have split in two. One asks the people who do the work. The other records their screens and treats the recording as the correction to whatever they said. Said, Seen, and Unseen argues that the split is a false choice, and that it fails first in operations where most of the work never appears on a screen.
In oil and gas, three of every four jobs in the industry that includes drilling and well services are field occupations. A screen recording of the office cannot see the yard, the shift, or the exception that was settled on a radio. An interview that asks for the official procedure tends to get the official procedure back. Readiness, if the word means anything operational, is the ability to tell those gaps apart before a system is asked to act inside them.
02
Authority belongs to the question, not to the source
The research on verbal reports does not say that interviews are unreliable. It says that some questions produce a report of what the person did, and others produce an inference about what they usually do. A question about a specific recent episode is in the first class. A question about how the process works in general is in the second, and it returns the canonical description even when the work is noncanonical.
The authors of process mining do not say that a step missing from a log did not happen. They require the opposite assumption: a log is evidence of what it recorded, not of the complement. Treating absence as absence of the work is how an unobserved step gets counted as if it had been seen and found empty.
From those two results the paper derives a rule. Do not rank sources. Assign authority by the question. A recording is the better source for the duration and sequence of a step that happened on a screen. The person who did an off-screen step is the better source for what that step was. Neither source is a source for the step no one asked about and no log contained.
03
Three states, so the unseen is not smuggled in as evidence
Every step in a process map can be marked in one of three states. On-screen: a recording or a system log observed it. Off-screen: someone reported a specific episode, and no recording could have seen it. Unobserved: no source covered it. The third state is the point of the partition. It stops a gap from being filled by the procedure, by a manager’s summary, or by a model’s guess, and then presented as a finding.
An organization that collapses those states into one narrative — “we mapped the process” — does not yet know what it would be delegating. Readiness is that partition, written down, with the unobserved steps still labelled as unobserved. The paper is explicit about what it does not settle, including which source is wrong, and how often, when an interview and a recording disagree about a step that did occur on a screen.
The sources, the partition, and the open questions are in Said, Seen, and Unseen. A map that keeps those states distinct, on one workflow, is what the audit is for.