Rough Order of Magnitude (ROM)
The rough order of magnitude, or ROM, is a single, deliberately coarse estimate carried with a wide band, produced very early in a project, when almost nothing about the work is yet defined. Its value lies in timing: it gives a sponsor enough to settle an early decision, most often whether to fund the next, finer and costlier estimating step, before committing the effort a tight estimate demands. It reads at the wide mouth of the cone of uncertainty, where the band is widest, and tightens as the project takes shape. BABOK ranks it among the estimation methods, alongside top-down, bottom-up, parametric and rolling-wave estimation, three-point estimation (PERT) and the Delphi method.
Goal
The rough order of magnitude provides a single figure, coarse and openly so, wrapped in a wide band that plainly states the ignorance in which it is produced. It exists for exactly one moment: the one in which a decision has to be made before the work it concerns is known. Whether to open a budget line for a project, to fund a feasibility study, to keep one idea over another in a portfolio of thirty. These calls fall at the very start, when neither scope, nor requirements, nor architecture are set, and they cannot wait for the detailed estimate that would take weeks of work to produce.
The decision it supports is a gate: it releases the next step without committing a budget. A single-point figure offered at the same stage would carry a lie of omission, because it would hide its own fragility and read as a promise. The rough order of magnitude does the opposite: the band it shows carries the information. It says that at this stage the true cost may lie well below or far above the central figure, and that the only reasonable question is whether the matter is worth investing in a more serious estimate.
The deliverable comes down to four things: a base figure, an explicit band around it, the assumptions it rests on and the clear statement that it is a rough order of magnitude. The rough order of magnitude is one of the methods in the estimation family, the one produced earliest and with the least material, when all the others require a decomposition or a history not yet available.
Usage
When to use it
- Very early, before any analysis: pre-study, opportunity screening, portfolio triage, when a coarse figure is enough to decide whether to invest in a serious estimate.
- A go/no-go decision: it is the relative order of magnitude of the cost that settles the call.
- A credible analogue is available: a comparable past project or an expert who has done the work anchors the figure.
- Far-horizon component of a rolling-wave plan: definitive for the near phase, rough order of magnitude for the rest.
- Stakes modest against the cost of estimating: the decision does not yet warrant a detailed estimate.
When not to use it
- A figure meant to be committed or set as the budget: build a definitive estimate by bottom-up estimation, from a decomposition of the work.
- No analogue and no history to anchor it: do the minimum elicitation that enables a parametric estimation, rather than quoting a figure with no basis.
- A single-point answer expected with no tolerance: if the wide band is unacceptable, produce a defensible range with PERT or a consensus with the Delphi method.
Description
What makes it fast
The rough order of magnitude requires no decomposition of the work, no breakdown structure, no sum of elements estimated one by one. That is what separates it from bottom-up estimation and what makes it fast. The estimator takes an analogue of the whole of the work, scales it and attaches a wide band. It never descends to the task level. It is a top-down approach by nature, the coarsest and earliest of the top-down estimations.
BABOK lists, in its Estimation section, the sources of information from which a rough order of magnitude draws its material. Analogous situations supply a comparable element, a project or initiative like the one being estimated. Organisation history brings the experience of similar work already done, all the more useful when it was done by the same team and with the same methods. Expert judgment draws on the knowledge of those who have done the work before. These three sources share one trait: none requires the work to be defined. That is what allows a figure to be produced when there is almost nothing to hold on to.
The band convention
A rough order of magnitude is never given as a bare figure. It is quoted with a deliberately wide band. That band separates an honest estimate from false precision. The width chosen is a convention, and three bodies of knowledge propose three different ones, which must be attributed rather than blended into an invented average. BABOK states, in its Estimation section, that a rough order of magnitude is often no more accurate than plus or minus 50 %. The PMBOK Guide, in its sixth edition, associates a band of −25 % to +75 % with a rough order of magnitude and a band of −5 % to +10 % with a definitive estimate; its seventh edition dropped these fixed bounds and invites teams to derive their own from their context. The AACE International system, in its recommended practice 18R-97, classifies estimates into five levels and gives its class 5, that of the rough order of magnitude, a typical range of −50 % to +100 %.
| Convention (body of knowledge) | ROM band | Tightest estimate |
|---|---|---|
| PMBOK Guide, 6th edition (PMI) | −25 % … +75 % | definitive −5 % … +10 % |
| BABOK v3, Estimation (IIBA) | plus or minus 50 % | “10 % or less” |
| AACE 18R-97, class 5 (AACE) | −50 % … +100 % | class 1: −10 % … +15 % |
These three conventions diverge because three bodies each quote their own practice. Averaging them into a fourth figure would fabricate a datum no one stands behind. What matters for the analyst is to choose a convention, name it and stay with it. The upper bound of a rough order of magnitude is always farther than the lower one. The asymmetry is structural: the unknowns of poorly understood work almost always add effort and only exceptionally remove it, so that work overruns more often than it comes in light, and a symmetric band flatters optimism.
The cone of uncertainty
The band holds for one moment of the project. It tightens by degrees as the definition of the work advances. That progression has a name: the cone of uncertainty. The idea that the estimating error narrows as knowledge grows was popularised under this term by Steve McConnell, after Barry Boehm published the original curve, which he called the funnel curve, showing that an estimate at the start of a project can be off by a factor of four. AACE gives the numeric version: its five classes correspond to levels of project definition, from 0 to 2 % defined for class 5 up to 65 to 100 % for class 1, and the accuracy range tightens class by class. The rough order of magnitude sits at the wide mouth of the cone, the definitive estimate at its narrow end. The same project, re-estimated after a feasibility study, carries a tighter band: it has gained definition.
How to produce it
The approach comes in six steps. First fix what is being estimated and the decision the figure must serve, since a rough order of magnitude produced with no decision to inform has no recipient. Then find the closest analogue or the best-placed expert. Derive from it a single base figure, by scaling the analogue. Attach an explicit band to it according to the chosen convention. State the assumptions it rests on and name what would tighten it, typically the next analysis step. Finally label the result unambiguously as a rough order of magnitude, so that no one mistakes it for a commitment.
The pitfalls
False precision is the first. Quoting a rough order of magnitude at “CHF 1'237'400” destroys what gives it value, because the honesty of the technique lives entirely in the band; a figure to the franc suggests a certainty the stage of the project forbids. The second, and most common, is the rough order of magnitude becoming the budget. A sponsor keeps the central figure, writes it into a contract, and the +75 % tail is later called an overrun. The figure feeds a decision gate and commits no budget, a distinction that must be repeated every time it is passed on. The third is the absence of a historical base: with no analogue and no organisation history, a rough order of magnitude is only a guess wearing a percentage, and BABOK notes, among its limitations, that an estimate is only as good as the level of knowledge of the work estimated. The fourth is never revising it: the cone only tightens if you re-estimate, and a rough order of magnitude left untouched while the project matures becomes stale, which is the reason for rolling wave. The fifth is reliance on a single method, which BABOK notes breeds unrealistic expectations.
AI considerations
A rough order of magnitude involves no complex calculation: a figure, a band, a few assumptions. A model's contribution therefore lies upstream of the figure and in the upkeep of its assumptions.
Upstream, a language model speeds up the search for the analogue, which is the real cost of the technique. Asked about comparable projects, past cost records, market benchmarks, it proposes anchor points and documented orders of magnitude that an expert corrects, which beats a figure set from memory. It can also derive a plausible band from the variance observed on analogous work and draft the assumptions log that accompanies the figure, a tedious task nothing requires to be done by hand.
What the machine cannot supply is a matter of judgment. The relevance of the analogue is a business appraisal: knowing whether the portal delivered by another municipality resembles the one being estimated demands field knowledge that no general model holds. The width of the band must reflect real ignorance, and a model queried without context produces a plausible and baseless figure, exactly the guess-wearing-a-percentage pitfall, worsened by an appearance of confidence. The discipline of the label, holding the figure to be a rough order of magnitude and resisting false precision, remains a human decision that no tool imposes in the analyst's place.
Examples
A municipality in French-speaking Switzerland is weighing the replacement of its virtual counter, the portal through which its residents complete their administrative steps online. At the pre-study gate, before any requirements work, the business analyst needs a coarse cost to decide whether it is worth funding a feasibility study.
| Analogue chosen | portal delivered by a comparable municipality |
| Base figure | CHF 1'200'000 |
| ROM band (−25 % / +75 %) | CHF 900'000 … CHF 2'100'000 |
| Decision supported | open the pre-study and fund the feasibility study (go), without writing any amount into a contract |
The −25 % / +75 % band frames the base figure. At this stage the municipality commits only the few tens of thousands of a feasibility study, in view of a probable cost of that order. Re-estimated once the feasibility study is done, the same project will carry a much tighter band, because the requirements and framing work will have turned a scope 2 % defined into something the estimator can decompose.
Visualisations
The cone of uncertainty draws a funnel narrowing from left to right as the definition of the project grows, the rough order of magnitude at the wide mouth, the definitive estimate at the narrow end. It teaches the one thing specific to a rough order of magnitude: the band is wide because the estimate is early, and it tightens by design.
The cone of uncertainty
The band tightens as the project takes shape
The estimate table carries the deliverable itself: a base figure, its band and the decision it serves. The conventions table sets the three competing bands side by side, so that each attribution is legible at a glance and the temptation to average them falls away.
Cost
| Phase | Level | Justification |
|---|---|---|
| Preparation | Low | No decomposition of the work and no workshop to convene. The effort is limited to identifying the closest analogue or the best-placed expert and gathering the available history. |
| Execution | Low | Scaling an analogue and setting a band is a matter of a few hours. It is this speed that justifies the technique at a stage where a detailed estimate would cost too much for the decision at hand. |
| Documentation | Low to moderate | The figure fits on one line. The care goes into the assumptions and the rough-order-of-magnitude label, without which the band is misread and the figure slides toward the status of a budget. |
Tools
A spreadsheet is enough to carry a rough order of magnitude: a base figure, a band factor, a line of assumptions. At this stage, heavier tooling adds friction without adding anything.
The quality of the figure depends above all on the source of the analogue. A database of past projects, a repository of historical costs, a base of internal or market comparables are what turn a guess into an anchored estimate; the richer and closer the history is to the work being estimated, the better the anchor. Parametric estimation tools and portfolio-management platforms often carry the rough order of magnitude as an entry-level estimate, the one that feeds an arbitration gate before a project is worked up in detail. Their value lies in keeping the history from one estimate to the next and making the analogue easier to find.
Sources
- IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.19 Estimation: the placing of the rough order of magnitude among the estimation methods, its status as a high-level estimate with a very wide confidence interval, its plus-or-minus 50 % band, its sources of information (analogous situations, organisation history, expert judgment) and its limitations (an estimate is only as good as the level of knowledge of the work; a single method breeds unrealistic expectations). BABOK describes these methods and prescribes no single band; the numeric bounds come from the PMBOK Guide and AACE International.
- PMI, A Guide to the Project Management Body of Knowledge (PMBOK Guide), 6th edition: the −25 % to +75 % band associated with a rough order of magnitude and the −5 % to +10 % band of a definitive estimate. The 7th edition abandons these fixed bounds in favour of ranges derived from context.
- AACE International, Recommended Practice No. 18R-97, Cost Estimate Classification System, As Applied in Engineering, Procurement, and Construction for the Process Industries: the five-class estimate system and their accuracy ranges, including class 5, that of the rough order of magnitude, at −50 % / +100 % and class 1, definitive, at −10 % / +15 %.
- Steve McConnell, Software Estimation: Demystifying the Black Art, Microsoft Press, 2006: the cone of uncertainty, a term popularised by this book and by his Software Project Survival Guide (1997).
- Barry W. Boehm, Software Engineering Economics, Prentice Hall, 1981: the original funnel curve, which holds that an estimate at the start of a project can be off by a factor of four.