Strategy making: reading a strategy against what the organisation does

A strategy states what an organisation intends. The organisation goes on deciding, spending and hiring in ways the strategy may or may not describe. This method reads the one against the other. It treats the scenarios strategists already write as fiction to be labelled and used for tests, not as evidence.

What is already known

Strategy has long been understood as more than a plan. Mintzberg and Waters separate the strategy an organisation intends from the one it realises, and the emergent pattern in between (Mintzberg and Waters 1985). Rumelt separates a strategy, which has a diagnosis, a guiding policy and coherent action, from a list of goals (Rumelt 2011).

There are working methods for reading a strategy as a text. Vaara and colleagues follow the making of a Finnish city’s strategic plan and name five features that give it force: self-authorisation, special terminology, discursive innovation, forced consensus and deonticity, the language of obligation (Vaara et al. 2010). Kornberger and Clegg read a city strategy as performative, something that does work in the organisation (Kornberger and Clegg 2011). Rouleau follows how middle managers interpret and sell a strategy in everyday routines (Rouleau 2005). The first of these reads what the text does to its readers and does not test it against what the city later did. That test is what Holagram adds.

The gap between what organisations say, decide and do has its own literature. Brunsson describes organisations in which talk, decisions and actions are decoupled, because each answers a different demand (Brunsson 1989). Meyer and Rowan describe formal structure adopted for legitimacy and kept apart from activity (Meyer and Rowan 1977), and Bromley and Powell review decoupling in the contemporary world (Bromley and Powell 2012). A strategy that does not match what the organisation does may therefore not be a fault to expose. It may be a document doing a different job.

The cascade, in which objectives become targets and targets become unit goals, assumes that levels add up. Its critics show what happens to a measure once it is a target: gaming in UK public spending numbers (Hood and Piotrowska 2020), metric fixation (Muller 2018), and rankings that reshape what they rank (Espeland and Sauder 2007). The second principle of Hola, that no level sums up another, says the same in general terms. Planning, on the other hand, does help on average: a meta-analysis of 87 correlations from 31 studies finds a positive, moderate effect of strategic planning on organisational performance, strongest for formal planning, in public and private organisations (George et al. 2019). A reading that calls a strategy ceremonial has to explain that.

Fiction already has a place in strategy, in scenario planning, and the evidence on it is mixed. With practising experts, using several scenarios did not change confidence overall, but judgement of a move shifted with how the move fared in each scenario (Phadnis et al. 2015). In a simulation, analysing several scenarios reduced strategic persistence through belief in a new strategy (Healey and Hodgkinson 2024). The route is belief, so a vivid scenario can persuade whether or not it is sound. One of the few published failures was a senior team that avoided the exercise (Hodgkinson and Wright 2002), and a later comparison with a successful intervention suggests checking how receptive the organisation is before starting (Wright et al. 2008). Causal layered analysis reads a statement at four depths, from the surface claim to the myth beneath it (Inayatullah 1998).

Practitioners already run forms close to this one. Policy steering rooms arrange material on four walls: the patterns, the proven, the promising and the possible (Mulgan and Quaggiotto 2023). Holagram’s registers can be laid over such a room, with a stamp for the register and the level on each item, and a separate hand for the test.

What Holagram adds to all of this is narrow: the four registers, a stated level on every claim, the rule that no reading judges itself, a test at a level below the document, and a place for synthetic hands.

Levels, and what each admits as a Trace

The layers of causal layered analysis measure depth of discourse and not scale. They belong inside the Reading register and are not mapped onto these levels (Inayatullah 1998).

The text and the ground

The text is the strategy document read with its siblings and not alone: earlier drafts, the mission statement, the annual report narrative, board papers, indicator sheets, budget and procurement notices, organisation charts, job adverts. Begin each study with a document sheet: who wrote it, for whom, from what template, and what job it does (to coordinate, to signal, to legitimate).

The ground is the pattern in what the organisation does (Mintzberg and Waters 1985), not a second document. Four public streams give it: what is funded (budget lines, grants), who is hired or appointed (job adverts, organisation chart changes), what is bought (procurement notices) and what is stopped (closures, withdrawn programmes). Each is a trace at move or unit level. The strategy’s claim is a trace at organisation level. The link between them is a Reading and cannot be a Trace. Indicator figures are the organisation’s own paper, with an author and an incentive (Hood and Piotrowska 2020), and are read as documents before they are used as traces.

Three tests

• Coupling. For a claim in the text, check whether a practice exists, and whether anything connects that practice to the stated end (Brunsson 1989; Bromley and Powell 2012). A Reading says which coupling it predicts: “this is talk” is held if the coupling is absent and broke if it is present; “this guides action” is held only if unit-level traces show it was followed.
• Another reader. Someone at unit level, or their documents, translates the same passage. Compare it with the organisation-level Reading (Rouleau 2005; Abdallah and Langley 2014).
• Fiction. Alter writes two or three worlds. A move is tried in each by another reader, and the test records whether the judgement of the move shifts with how it fares in each world, not whether anyone becomes more confident (Phadnis et al. 2015; Healey and Hodgkinson 2024). The worlds are never evidence. Their observables are leads that a person checks, and the Adversary’s test comes after the Fiction and is written by a different hand.

Agents

Boards, unit heads, staff, those the strategy is for, peers, regulators, and models that draft or summarise strategy documents. A recent strategy may be partly machine-drafted, so the document sheet asks which sections were written by whom, as a hands note would. For comparing passages, one early test of language-model judging found that pairwise comparison separated stronger from weaker corporate climate reports and held up better against text a model had been asked to wash (Chuang et al. 2025). It is a workshop paper, so treat it as a lead.

A worked entry

Invented, to show the shape.

Where it goes wrong

• A text-action gap read as a fault, or as proof of the Reading. Decoupling is often designed (Brunsson 1989; Bromley and Powell 2012). Name the document’s job first.
• The host avoids the reading. Scenario interventions have failed because the senior team avoided the exercise (Hodgkinson and Wright 2002; Wright et al. 2008). Before a study that involves people, run a short receptiveness interview, and give the outside reader a stop rule.
• The Fiction persuades. A scenario moves people through belief (Healey and Hodgkinson 2024), whether or not it is sound.
• The method’s own numbers get gamed. Never report held or broke counts as a score (Muller 2018; Hood and Piotrowska 2020).
• Front-stage selection. Public text is where talk aimed at legitimacy is most likely to sit (Brunsson 1989), so a gap there is expected and proves little. Require at least one lower-level Trace per entry.
• Depth taken for scale. See the note on causal layered analysis above.

Objections the literature raises

• “Strategy-as-practice and discourse analysis already do this.” They read strategy texts and follow them into practice (Vaara et al. 2010; Kornberger and Clegg 2011; Rouleau 2005). What Holagram adds is the rules and not the object: stated levels, separate hands for reading and testing, labelled fiction and a failure log. The first study is built so that it can be compared with a plain discourse analysis of the same text.
• “The strategy is not the document.” A strategy is a pattern in action (Mintzberg and Waters 1985), so reading the text first studies the layer that matters least. Hence the ground.
• “Planning works.” See the meta-analysis above (George et al. 2019).
• “The method creates its own metric.” Any count becomes a target once someone funds or ranks on it, and new categories made to stop gaming can open new ways to game (Hood and Piotrowska 2020). The counts are reported to calibrate the method and are not a score.
• “Public text and three weeks are the wrong conditions.” Strategy effects take years, public text is the front stage, and a retrospective design brings hindsight risks. The study claims a case and says so.
• “Labs and innovation offices cannot host this.” Policy labs are described as small, short-lived and dependent on patronage (Lewis 2020), and a reading that shows a unit’s own strategy as decoupled can read as an accusation. Readings are about documents and offices, and the body that owns a strategy reads the entry before it is published.

First study: a strategy whose term has ended, three weeks

Choose a strategy whose term has ended, so the outcomes are on record, from a public organisation that is not a client: a city, an agency, a university, or a national agenda with indicators set against benchmarks. Prefer one that has already been analysed in print, such as the strategic plan of Lahti (Vaara et al. 2010) or Sustainable Sydney 2030 (Kornberger and Clegg 2011), so that a published reading gives a second reader to set against Holagram’s. Public material only. Readings are about the organisation and its offices, not about named people.

What would show it works. The counts on the Research page, and at least one claim of the form “the strategy says X, the organisation does Y” that was tested at the unit level and either held or broke for a stated reason.

Read next

• Strategy making, governance innovation and systems, observers and levels in the literature review.
• The method, synthetic hands, and the neighbouring proposals: Cities and Mandates.