BeHAVIOURAL INDUSTRIALISM
Consulting inherited its shape from the era it was built for.
Diagnose, analyse, recommend, invoice, leave. It assumed the hard part was knowing what to do.
That assumption expired.
Information is abundant, benchmarks are commoditised, and any competent team can now generate a defensible strategy in a fortnight.
The hard part is, and always was, getting an organisation of adults with competing incentives to actually do the thing.
Behavioural Industrialism treats that as an engineering discipline rather than a communications exercise.
THE CONTRAST
TABLE
Unit of work
Evidence
Assumption
Role of AI
Ends with
Failure mode
Priced on
Traditional consulting
The recommendation
Benchmarks and analogues
Rational, informed, willing
A workstream
A deck and a departure
Adoption
Hours billed
Behavioural Industrialism
The decision
Experiments with control groups
Biased, busy, incentivised elsewhere
The operating substrate
An installed decision system
Uncomfortable findings
The decision improved
THE PRINCIPLES
-
Two. Falsify before you scale.
-
Corporate pilots are rarely tests. They are demonstrations, staffed by volunteers who want the thing to work, measured against no control, and evaluated by the person whose budget depends on the answer. A pilot that cannot fail has told you nothing, at considerable expense.
-
An intervention that cannot be disproved is a belief. Beliefs are inexpensive at the scale of one team and ruinous at the scale of an enterprise, because organisations do not abandon them quietly. They defend them, fund them, and eventually build an operating model around them.
-
The discipline is unglamorous. Define the measure before the intervention, not after. Hold something back as a control, even when it feels wasteful. Write down what result would cause you to stop, and circulate it, so that the decision to stop is already made before anyone has become attached.
-
The cultural difficulty is real. Falsification requires an organisation to publish its own failures, and most reward structures punish exactly that. Which is why this principle is usually adopted as a process and abandoned as a practice within two quarters, unless someone senior visibly kills their own idea first.
One. Design for the person who is tired.
-
Most organisational change is designed by people at their sharpest, in a room booked for the purpose, and then handed to people at their most depleted, on a Thursday afternoon, between two other things. This is the single most reliable predictor of whether a change survives.
-
The evidence for willpower as a depletable fuel tank is weaker than the popular literature suggests, and we do not lean on it. The robust finding is duller and more useful: behaviour follows the path of least resistance almost regardless of intent. Automatic pension enrolment moved participation rates further than three decades of financial education. Nobody's convictions changed. The default did.
-
So the test we apply to any new process is not whether it is sensible. It is whether it survives indifference. If a step requires someone to remember, to care, or to override the easier option, we assume it will not happen, and we redesign until the desired action is the lazy one.
-
This produces less impressive-looking programmes. It also produces programmes that are still running in month nine, which is the only measure that has ever mattered.
-
Three. Separate the decision from the outcome.
-
A good decision can lose money. A reckless one can make a great deal of it. Everyone accepts this in the abstract and almost no organisation is structured to act on it.
-
Performance reviews look at results. Promotions follow results. So the acquisition that worked is treated as evidence of judgement, and the one that did not is treated as evidence of failure, when in both cases the process may have been identical and the difference attributable to the market. The organisation learns to copy the lucky and avoid the unlucky, which is not learning at all.
-
The correction is a decision log. Before a material choice, record what is expected, with what confidence, and on what basis. Afterwards, review the reasoning separately from the result, and be willing to say that a decision was sound and the outcome poor.
-
Two things follow. Judgement becomes visible and therefore improvable, and the cost of being wrong falls, which is what allows people to take the intelligent risks that the strategy assumed they would take.
-
The obstacle is never the mechanics. It is that senior people must be prepared to be assessed on reasoning rather than results.
-
Four. The machine decides, the human is accountable.
-
Most organisations have deployed models faster than they have redrawn decision rights. The result is a recommendation with no owner: a system that produces an answer, a manager who did not generate it and cannot fully explain it, and a chain of accountability that dissolves precisely at the point where it is needed.
-
This creates two failure modes, and they look nothing alike. In the first, nobody acts on the model, because acting on something you cannot defend is career risk without career reward. Expensive infrastructure produces advice that is politely ignored. In the second, everyone defers to it, and the human in the loop becomes what researchers have called a moral crumple zone: present enough to absorb blame, powerless enough to prevent the error.
-
The remedy is structural, and it comes before the technology, not after. Classify each recurring decision as automated, augmented or deliberately human. Give every automated decision a named owner with the standing to override it and the obligation to review it. Calibrate trust deliberately, because a workforce that over-trusts a model is as dangerous as one that ignores it.
-
Automation without a named owner is not efficiency. It is diffused liability, arriving later and larger.