Software Estimation Techniques

* adapted from Miranda, 2014

1. What Estimation Answers

Estimation helps us answer two core questions:

  • How big is it? → Sizing the system (e.g., number of features, modules, function points)
  • How much effort is needed? → Effort and cost estimation (staff months, budget)

Estimation converts scope into resources and time.

2. Two Basic Estimation Strategies

Approach Process Strengths Weaknesses Example
Bottom-Up Estimate cost/effort for each component, sum them, check total. Accurate when WBS is detailed. Time-consuming; depends on detailed scope. Estimating each microservice and summing total developer hours.
Top-Down Start with total project budget → allocate to major subsystems → refine. Fast; useful early in planning. Risk of unrealistic breakdowns. Management fixes total budget ($500k) and allocates to phases.

Both require sanity checks: overhead, contingency, and lifecycle coverage (e.g., testing, maintenance).

3. Categories of Estimation Approaches

(a) Judgment or Expert-Based Estimation

Estimation based on experience and intuition.

  • Unaided: relies purely on expert intuition (e.g., senior engineer says “6 months”).
  • Structured: uses formal elicitation and combination methods:
    • Wideband Delphi – iterative group consensus.
    • Planning Poker – agile team estimation.
    • Paired Comparison – experts rank or compare tasks pairwise.

Example: Estimating sprint backlog effort using Planning Poker.
Use When: Project has little data, early stage planning.

(b) Engineering Approaches

Estimate by analyzing what drives effort—build from known process characteristics.

Example: Training project

Number of users × Hours per class × Locations + Fixed setup costs.

Used when no past project data exists.

Variants:

  • Process-based cost modeling (breakdown by activities such as design, coding, testing).

Use When: Understanding process mechanics is easier than collecting data.

(c) Counting Approaches

Quantify system size via measurable functional elements.

Technique Measure Example
Function Points User-visible inputs/outputs, files, interfaces 10 inputs × 4 + 5 outputs × 5 → FP total
COSMIC Points Functional data movement Used in modern distributed systems
Use Case Points Number and complexity of use cases UML-based
Web Object Points Pages, scripts, media Web apps
Test Points Test cases and complexity QA planning

Use When: Early requirement specs are available.

(d) Industry Norms

Based on historical distributions across projects.

Typical averages:

  • Requirements: 15%
  • Design: 10%
  • Coding: 35%
  • Testing: 40%

Also includes Function Point Backfiring: converting function points to LOC based on historical ratios.

Use When: Benchmarking effort allocation.

(e) Analogy-Based Estimation

Estimate by comparing with similar past projects.

  1. Find analog projects.
  2. Adjust based on similarity.
  3. Average or cluster results.

Variants:

  • Proxy-Based Estimation (PROBE) in the Personal Software Process (Humphrey)
  • Academic systems: ESTOR, ACE, ANGEL use clustering.

Example: A mobile app project estimated by comparing with a similar past app.
Use When: Reliable database of past projects exists.

(f) Parametric Models

Statistical models linking effort to measurable drivers (Cost Estimation Relationship, CER).

General Form:
Effort = a × (Size)^b × Π(cost drivers)

Famous examples:

  • COCOMO II (Boehm): Effort = a × (KLOC)^b × EAF
    (EAF = effort adjustment factors like reliability, complexity)
  • SLIM (Putnam) – uses Rayleigh curve of manpower distribution.

Example:
If estimated size = 20 KLOC, COCOMO predicts ~50 person-months.

Use When: Large dataset available; detailed drivers known.

4. Comparison Summary

Category Data Needed Typical Accuracy Example Methods Best Used When
Judgment Expert opinion Low–Medium Delphi, Planning Poker Early stages
Engineering Process decomposition Medium Activity-based New, data-poor projects
Counting Measurable features Medium–High Function Points Defined requirements
Industry Norms Historical ratios Low Effort % by phase Benchmarking
Analogy Similar projects High PROBE Stable organization
Parametric Statistical models High COCOMO, SLIM Mature data sets

5. Combining Techniques

In practice, estimators combine methods:

  • Use expert judgment to calibrate parametric models.
  • Use counting to size and COCOMO to predict effort.
  • Apply analogy for cross-checks.

6. Example Integration

Suppose a company estimates a CRM system:

  • Function points = 350 → size-based estimate.
  • Apply COCOMO → 75 person-months.
  • Experts adjust for complexity (+10%) → 83 PM.
  • Add 15% contingency for uncertainty → final = 95 PM.

Acknowledgments

This content is heavily inspired by and adapted from lectures by Eduardo Miranda and David Root on software project management. The structure, examples, and pedagogical approach reflect their teaching materials and frameworks.

Sources

  • Miranda, Eduardo. Managing Software Development. Lecture materials, 2014.

Disclaimer: AI is used for text summarization, explaining and formatting. Authors have verified all facts and claims. In case of an error, feel free to file an issue or fix with a pull request.

Table of contents