The wrong lens
Where an IT risk matrix goes wrong on a plant floor
The method is not wrong in itself; it was calibrated for a different world, and each of its habits misleads here in a specific way.
- 01
It puts confidentiality first in a place that runs on availability
The classic triad reads confidentiality, integrity, availability — in that order — and IT scoring inherits it. On a plant the order inverts: a leaked setpoint is an embarrassment, a falsified one is a hazard, and a control system that stops responding is the event the whole operation is organised to prevent.
- 02
It values assets by their price tag rather than their process role
An asset register priced from the balance sheet puts the ERP servers at the top and a few-hundred-dollar controller at the bottom. The process reads it the other way: that cheap controller may hold the valve on which the whole train depends, and its loss is measured in production days, not replacement cost.
- 03
Its likelihood arithmetic assumes an estate that patches
IT likelihood models lean on exposure windows closing monthly as patches land. A control estate holds versions for years because the vendor qualification, not the bulletin, decides what runs — so likelihood here has to be argued from reachability and from who could plausibly get there, not from patch age.
- 04
Its evidence habit is to scan, which here is a hazard
The IT assessor's first instinct is a vulnerability scan. Pointed at older controllers, the same probe can hang the device it is measuring — the assessment itself becomes the incident. The instinct has to be replaced, not restrained: everything the scan would have told you can be learned another way.
Consequence first
The question is what happens to the process, not to the machine
A consequence-driven assessment starts at the end. Before asking how a system could be compromised, it asks what the site would experience if it were: which unit trips, what spills or overheats, whether the safety layer is engaged, how long restart takes and what a day of lost throughput costs. Those answers come from operators and process engineers, not from a security tool, which is why the core of the engagement is a structured workshop with the people who run the plant rather than a scan of it.
Only then does the technical layer earn its place, because now it has something to rank against. A compromised historian and a compromised engineering workstation look similar in an asset list; traced to consequence they are different orders of event — one degrades reporting, the other can rewrite the logic in every controller its project files describe. The register that results is short where a generic one is long, and severe where a generic one is bland, because every line has been argued through to what it does to the process.
The method
Reading the estate through everything except a probe
Each source replaces something a scanner would have claimed to measure, and together they see more than the scanner would have — without one packet aimed at a controller.
Listen to the traffic
Passive collection from switch mirror ports reveals what exists, what speaks to what, and over which protocols — an inventory built from observation, with nothing transmitted toward the equipment. An asset and flow map the plant did not have to risk anything for.
Read the configurations
Firewall rule bases, switch configurations and controller project files are exported and reviewed offline — exposure judged from what the rules actually permit, and versions read from records rather than probed from the wire. Reachability established from evidence, not assumption.
Interview the people who run it
Operators and control engineers know the undocumented paths — the vendor modem behind a cabinet, the laptop that bridges networks on night shift, the workaround everyone uses and nobody wrote down. Interviews surface what no export contains. The paths an attacker would find and a tool never would.
Walk the site
A physical walk-down verifies what is actually cabled to what: the unmanaged switch added during a turnaround, the USB port nobody disabled, the cabinet that does not match its drawing. Some findings only exist at eye level. A drawing corrected against the metal, not the archive.
Our role
A register the control room and the board read the same way
- 01
Scenarios in operational language
Each register entry is a named scenario — which path, to which system, with which process consequence — so a shift supervisor recognises it as true and a board member understands what is at stake without translation.
- 02
Treatments the plant can execute
Every risk carries options the operation can actually take — a network control, a procedural change, an item planned into the next window — costed against the consequence they reduce, never a generic instruction to patch what cannot be patched.
- 03
A method your team keeps
The scenario library, the consequence scales agreed with operations and the evidence sources are handed over documented, so the next review updates the register rather than starting again.
Which risks the operation accepts, and what any treatment may touch on the control path, are decisions that stay with your leadership and the plant's engineering authority; we supply the evidence and the options.
Common questions
What buyers ask
Why not extend our enterprise risk assessment to cover the plant?+
Because the calibration is wrong in three places at once, and each error compounds the others. Enterprise scoring puts confidentiality first, values assets by book price, and computes likelihood from patch cadence — and a plant inverts all three: availability and safety lead, a cheap controller can carry the whole train, and the estate holds software versions for years by design. Extending the enterprise method produces a register that is formally complete and operationally misleading, which is worse than no register because it directs spending at the wrong lines. The plant needs the same governance but its own method underneath, and the two registers can then roll up honestly.
Does the assessment ever touch a live controller?+
No. Nothing in the method transmits toward control equipment: discovery is passive listening on mirror ports, versions and rules come from configuration exports read offline, and the rest is interviews and walk-downs. This is not caution for its own sake — older controllers can hang when probed by tools written for IT equipment, so an active scan risks causing exactly the outage the assessment is meant to price. The trade-off is honest: a passive method can miss a vulnerability a probe would confirm. We accept that, because the missing datum is recoverable from vendor advisories and version records, while a tripped unit is not recoverable at all.
How do you judge likelihood on systems with no patch cadence at all?+
By replacing patch age with reachability and capability. The useful questions are structural: can anything hostile form a session with this device at all; how many network boundaries, credentials and physical doors stand between the internet and it; does exploiting it require commodity tooling or process knowledge an outsider rarely holds. A controller with a decade-old firmware build behind a well-kept conduit that admits two named stations can be a lower likelihood than a fully patched server exposed to a flat network. That reasoning is arguable and auditable — each rating carries the reachability evidence behind it — which is more than a formula pretending to precision can offer.
In this cluster
Next step
Scope the assessment with us
A scoping call that agrees the units in scope, the evidence sources available without touching the process, and who joins the consequence workshop.

