Bottom-up Estimation

Goal

Bottom-up estimation produces a defensible number and makes it traceable: rather than announce a total and justify it after the fact, it starts from the work itself, breaks it down into components small enough to be priced one by one, estimates each on a tangible basis, then sums those amounts back up the decomposition. Such a number is what every commitment decision calls for, the amount of a fixed-price offer, the budget of a phase, the quote handed to a client. The total is therefore the result of a calculation, and its justification is already written, line by line, in the decomposition that produced it.

The decision it supports is a commitment made on a known scope. Answering a tender, setting a price that will hold, choosing between two options whose costs you know in detail, all of this assumes a so-called definitive estimate, one whose expected deviation from actuals is measured in percentages rather than in multiples. That is the precision band bottom-up estimation reaches, where a rough order of magnitude applied to the whole settles for a factor of two. It reaches it by paying the price of inspection: you cannot sum leaves you have not first defined and estimated, and that effort is the direct counterpart of the accuracy. It is the most accurate of the estimation methods and the most costly, and the trade-off between those two qualities alone governs whether to use it.

The deliverable is a priced decomposition: one line per component, the amount of each, subtotals per work package, then the total and a contingency reserve that turns it into a usable budget. Where top-down estimation examines the work from above without decomposing it to the leaves, bottom-up estimation takes the opposite path and draws its traceability from it: every amount in the total points back to a component that can be discussed, corrected or challenged in isolation.

Usage

When to use it

• Deliverable fully understood, activities defined and measurable: each leaf is priced as a measured quantity times a unit rate, as in construction under SIA standards.
• Definitive estimate or firm budget required: a fixed-price offer, a quote, a phase budget, where accuracy of around ten percent justifies the cost.
• Confirming a rough order of magnitude or a top-down number: decompose one representative slice in detail, a marked gap revealing that the overall number is wrong.
• Stable scope: the content no longer moves enough for the decomposition to be obsolete before it is finished.
• History of actual costs available: comparable delivered projects supply unit rates and per-component amounts on which to calibrate each leaf.

When not to use it

• Deliverable not yet fully understood: in software development, research or early discovery the leaves cannot be defined, so the summed total is mere false precision, prefer a rough order of magnitude or a top-down estimate.
• An answer is needed quickly or the stakes are low: the cost of the decomposition exceeds the value of the answer, use top-down or parametric estimation where a per-unit model exists.
• Scope still volatile: the components change faster than they can be estimated, decompose only the stable near-term slice with rolling-wave estimation and defer the rest.

Description

The technique runs in four moves, in order: decompose the scope down to estimable leaves, price each leaf separately, consolidate the amounts upward, then add a reserve and record the assumptions. None of these moves belongs to BABOK in its own right, which names the method without laying out the procedure: the approach belongs to project-estimating practice, and PMI's standard gives the detail.

Decompose down to the estimable leaf

The starting point is a decomposition of the work, most often a work breakdown structure (WBS), which splits the deliverable into work packages and then into elementary activities. The decomposition stops when a component is small enough that one person can price it with confidence, on a tangible basis. That is the estimable leaf, the unit that will be summed. Without this decomposition there is nothing to add up: that condition decides the method before any other consideration.

The most destructive pitfall is the forgotten leaf. The total is worth only as much as the decomposition, and any activity missing from the WBS is silently costed at zero. An unlisted coordination task, a rework step, an ancillary service passed over in silence do not inflate the total, they cut into it, and the error is invisible because it leaves no line. The defence is methodical: run the decomposition against a reference checklist or the WBS of a comparable project and treat the completeness of the breakdown as a deliverable in its own right before pricing anything.

Estimate each leaf independently

Each component then receives its amount, obtained by the soundest route available: a measured quantity multiplied by a unit rate when the work is measurable, an actual cost drawn from the organisation's history, expert judgement or an analogy with work already done. A leaf may itself carry a range rather than a single point, and a three-point estimate per component, whose mechanics PERT sets out, is a common refinement when the per-leaf uncertainty is worth quantifying.

Two pitfalls lie in wait at this step. Double counting first: a shared component, estimated once in one package and once in another, inflates the total, and the usual cause is a blurred boundary between two overlapping packages. It is caught by checking that the package scopes are disjoint and cover the whole without overlap. False precision next: an amount stated to the franc inherits the uncertainty of each of its leaves, and a total presented as a certain point disguises that accumulated uncertainty. Keeping the leaves' ranges or confidence levels, rather than collapsing each estimate onto its median value, preserves the information that a single total erases.

Consolidate upward

The amounts then climb the hierarchy: activities sum into packages, packages into deliverables, deliverables into a total. This is the move that defines the method and gives it its traceability, since the total is built line by line. A quiet pitfall comes with it, the stale decomposition: summing leaves estimated against a scope that has since moved yields a total that is precise and wrong. The sum is therefore re-checked against the current version of the scope.

Consolidation reveals a pitfall of a different order, one that lies in what links the leaves together. Summing the parts ignores the cost of assembling them: integration, whole-system testing, coordination, site logistics, project management itself. The whole costs more than the sum of its leaves, and a decomposition that reserves no line for these cross-cutting activities produces a total that is structurally too low. These assembly costs are components of the WBS in their own right, to be decomposed and priced like the others.

Contingency reserve and assumptions

The consolidated total is not yet a budget. Adding raw estimates treats each leaf as its expected value, which, repeated across dozens of components, guarantees an overall underestimate: across a large number of leaves, the overruns do not all cancel out. A contingency reserve corrects this bias and covers the identified uncertainty. It is often set as a percentage of the total, or better, estimated per risky component when the risk concentrates on a few leaves rather than spreading evenly. The total plus the reserve, together with the assumptions and exclusions that bounded it, is the deliverable. A constraint or an assumption not communicated with the estimate, about the availability of a resource or the exact content of the scope, turns an accurate number into a misleading one the moment the reader takes it for granted differently from the estimator.

AI considerations

The calculation itself needs no artificial intelligence: summing amounts and adding a percentage fits in a few spreadsheet cells. A model's contribution lies upstream of the decomposition and in its control.

Upstream, a language model helps accelerate the decomposition. From a scope statement or comparable projects, it proposes a candidate WBS and a list of packages the estimator prunes, which replaces the blank page with a breakdown to correct. Over an organisation's project archive, it suggests unit rates and per-component amounts drawn from work actually delivered, mechanising the source BABOK calls "organisational history". Its most valuable use is control: checking a decomposition against reference WBSs is exactly the kind of pattern recognition that spots the missing leaf and the double count, the two errors that sink a bottom-up estimate, and that tests the plausibility of a consolidated total.

What the machine does not supply is judgement, and it bears first on the question that decides everything: is the scope complete and stable enough to be decomposed. The whole method depends on it, and no general model holds that business reading. Likewise, deciding that a leaf is genuinely estimable, sizing the reserve against the project's real risk and handling a component sensitive to safety or to stakeholders fall to the analyst. The model proposes leaves and rates, the human remains accountable for the completeness of the breakdown and for the decision to accept the total.

Examples

A physiotherapist is fitting out a new practice in French-speaking Switzerland. The scope is fixed and the works are defined, so each item is priced as a measured quantity times a unit rate. The decomposition follows the positions of the Swiss construction cost classification (CFC) and reads from the bottom up: each item amount feeds into the sum of items, to which a ten-percent reserve is added to give the estimated total.

The concept the priced decomposition makes visible is the construction of the total from the bottom. No summary line was guessed: the sum of items, CHF 38'900, is the exact addition of the six amounts above it, and each of those amounts recomputes from its quantity and unit rate, 45 m² at CHF 40 giving CHF 1'800, 220 m² at CHF 35 giving CHF 7'700. The ten-percent reserve, CHF 3'890, sits on this measured total and brings the estimate to CHF 42'790. Every franc points to a line the physiotherapist can discuss with the contractor in isolation: contesting a unit rate, revising a quantity, dropping an item does not rebuild the estimate, it corrects only one of its lines. The total stops before VAT, because it is the estimate of the work; VAT at 8.1 percent is a tax add-on carried on that total, and mixing it into the reserve line would blur what the decomposition demonstrates.

Visualizations

The technique produces two objects, and each calls for its own medium. The first is the estimate table: made of rows and columns, it is the deliverable itself, and its amounts are what a reviewer recomputes to verify the estimate. It stays in HTML, selectable and recomputable, rather than frozen into an image.

The second is the logic of the consolidation, which is better seen drawn than tabulated. The decomposition tree of the same fit-out carries that logic: the deliverable at the top branches into packages and then leaves, each leaf shows its measured amount, and the amounts climb the tree to sum, level after level, up to the total, above which the reserve is added. The tree shows at a glance what the table states in columns: the total is built from the bottom up, each branch carries its share, and nothing is placed on the whole without passing through the parts. The tree makes the direction of the construction visible, the table carries the reproducible calculation.

Cost

Tools

The spreadsheet is the honest choice and hard to beat. The items sit in rows, the quantities and unit rates in columns, the amount of each leaf and the subtotals recompute themselves the moment a value changes, and the reserve is set with a single formula. For a fit-out of a few dozen items nothing more is needed, and the transparency of the workbook serves the traceability the method seeks.

Beyond that, project-management tools carry the decomposition natively: you enter the WBS, assign each package its estimate, and the tool consolidates amounts and durations along the plan, which spares you keeping a spreadsheet in parallel. In construction, quantity-surveying and structured-quotation software organised by CFC positions calibrates each leaf against market unit-rate bases and produces the priced decomposition in the format a tender expects. This extra tooling is justified when the number of leaves, the frequency of revisions or the format requirement exceed what a workbook holds cleanly; below that, it adds a licence without contributing anything the spreadsheet does not already do.

Sources

• IIBA, A Guide to the Business Analysis Body of Knowledge (BABOK Guide) v3, §10.19 Estimation: the placement of bottom-up estimation among the estimation methods, the need for an agreed description of the elements to be estimated, often a work breakdown structure, the communication of constraints and assumptions with any estimate and the link between accuracy and the confidence interval. BABOK names the method and does not prescribe its procedure, which comes from PMI's standard.
• PMI, Practice Standard for Project Estimating, 2nd ed. (2019): the bottom-up estimating procedure based on decomposition into a work breakdown structure, the estimation of components at leaf level and their consolidation, the contingency reserve that turns the total into a budget and the accuracy classes that place a definitive estimate within a few percent of its actual, where a rough order of magnitude stays within a factor.