Reorganise the work, not the boxes

Near the front of most restructuring business cases sits a familiar sentence: seventy per cent of change initiatives fail. It is the most-quoted number in this market, and it appears to have no empirical source. Hughes (2011) traced the claim through the change-management literature and found it propagating from citation to citation without a study underneath it. Where an origin is offered it is usually Kotter's 1995 Harvard Business Review article, which opened with an observation drawn from the author's own consulting experience rather than from a dataset — no sample frame, no definition of failure, no measurement.

The point is not that Kotter was careless; his argument about the difficulty of transformation has held up rather better than the number attached to it. The point is that a board which accepts a seventy per cent failure rate has accepted in advance that most of what it is about to spend will be wasted, and has stopped asking the harder question of what would make this particular review one of the ones that works.

There is almost always a gap between the brief and the problem. The brief says: review the structure, recommend an operating model, reduce duplication, clarify accountabilities. The problem, described by the same people ten minutes into the first conversation, is different — decisions take too long, nobody owns the customer, two teams are doing the same thing and a third thing is not being done at all.

Those are statements about work, not about reporting lines. Galbraith's Star Model treats structure as one of five interdependent design choices alongside strategy, processes, rewards and people practices, and is explicit that changing one without the others produces misalignment rather than improvement. The brief should be treated as a symptom report rather than a specification.

Structure is the default intervention for good reasons: it is visible, it can be decided in a room, it produces an artefact the board can approve, and it can be announced as complete on the day it is announced. What it reliably fails to change is the work. Duplication is usually a symptom of unclear demand and undefined handoffs; move the boxes and the duplication migrates with them. Gaps in accountability are usually the residue of decisions nobody is authorised to make; create the role without the authority and the gap persists under a new title.

The sociotechnical tradition reached the same conclusion in a coalmine seventy-five years ago. Trist and Bamforth (1951) documented what happened when the longwall method reorganised mining work into specialised shifts: the technical logic was sound, the social system was destroyed, and productivity and wellbeing both fell.

The most reliable diagnostic of a structure-first review is the counterfactual. If the new chart were implemented tomorrow, which specific decision would be made differently, by whom, on what information? If that cannot be answered concretely for at least three decisions, the review has not yet found anything.

Meadows (2008) ranks twelve leverage points in a system. Adjusting numbers — budgets, headcount, targets — is the weakest. Structure and rules sit low in the ranking. The strongest interventions are the system's goals, the distribution of power over the rules, and the mental models the arrangement rests on. A review that changes reporting lines and leaves purpose and measures untouched has intervened at the weak end of that scale, at considerable cost.

Ashby's law of requisite variety (1956) adds a warning about simplification: a controller needs at least as much variety in its responses as the system it regulates. Consolidating decision points and removing local discretion reduces response variety, which is the mechanism behind the consolidation that saves cost in year one and quietly loses responsiveness by year three.

Deming (1986) argued that most of what looks like individual failure is produced by the system people work in, and that measures drive behaviour more powerfully than instructions do. Seddon's concept of failure demand — demand caused by a failure to do something, or to do it right, for the customer — is the practical extension: reorganise around that volume and you have built a structure optimised to process your own defects.

Datta, Guthrie, Basuil and Pandey (2010), reviewing three decades of downsizing research, found the relationship between workforce reduction and subsequent financial performance to be weak, inconsistent and frequently negative, with implementation quality mattering more than the size of the cut. The survivor literature documents the mechanism: the people who remain show measurable declines in trust, commitment and discretionary effort, strongly conditioned by whether the process was perceived as fair.

The Australian public evidence is more pointed, because it is auditable. The NSW Auditor-General's Machinery of government changes (2021) found anticipated benefits had not been articulated in sufficient detail, benefit achievement was not monitored, and costs were not tracked — and estimated a minimum direct cost of $23.7 million for a single 2019 departmental change, a figure the audit had to construct because nobody had counted it.

Bernerth, Walker and Harris (2011) developed and validated a change fatigue scale, establishing cumulative exhaustion from repeated organisational change as a measurable construct rather than a figure of speech. Rafferty and Griffin (2006) had earlier shown that the frequency of change and the uncertainty surrounding it each independently predict psychological strain.

The design implication is direct. How often an organisation reviews itself is a decision with consequences, in exactly the way that how it structures itself is. An organisation on its fourth restructure in six years is not simply one that has not yet found the right answer; it is one whose people have learned that answers are provisional and that engaging with them is not worth the effort.

Four things are worth diagnosing before anyone opens a drawing tool. Purpose and demand: what the unit is for, described by the people it serves, what demand actually arrives, and how much of it is failure demand. Flow and handoffs: follow a representative piece of work end to end and record where it waits, is re-entered, is checked by someone who adds no information, or goes back.

Decision rights: map what each level may decide without escalating and compare it with what it actually escalates — the gap between formal and practised authority explains more about organisational speed than the shape of the chart. Job design: Parker's (2014) review establishes that autonomy, skill variety, task significance and social support predict performance, learning and wellbeing, and that these are design choices.

Structure follows from these findings rather than preceding them. When the work is understood, most structural questions answer themselves: teams form around the flow of work, boundaries are drawn where handoffs are cheapest, and authority is placed where the variance occurs. The resulting chart tends to be less elegant than the one a design workshop would produce, and considerably more likely to survive.

Drawing the evidence together, a defensible organisational review runs in six stages: frame the problem in terms of the work; diagnose the work system; establish the baseline of cost and delivery; develop options with honest trade-offs; implement in stages, fairly, with genuine voice where influence is possible; and re-measure against the baseline at a defined interval. The last stage is almost always omitted, and it is the only one that produces organisational learning.

Before signing an engagement, a board should be able to answer six questions: what specific decision or piece of work will be done differently; what evidence says the current arrangement is the cause; what the change will cost, including disruption; what improvement is claimed and how it will be measured; how many changes this workforce has absorbed in the past three years; and who will report the results, to whom, and when. A review that cannot answer these is not ready to be commissioned — whoever is proposing it, including us.