Reference Class Forecasting (RCF) vs nPlan
RCF benchmarks your project against similar completed ones; nPlan's AI-SRA forecasts every activity in your schedule. When to use each, in the UK and US.
.jpg)
Updated 29 October 2026
Short version: For owners, project controls leads and risk managers choosing a forecasting method: reference class forecasting (RCF) gives a top-down view of how similar completed projects performed and suits early budgets before a detailed schedule exists. nPlan's AI-SRA forecasts every activity in your own schedule, shows what is likely to drive delay and keeps working through delivery.
If you’re planning a major project, some of the first questions you’ll ask are: 'how long will this take?', and 'how much will it cost?'
Reference class forecasting (RCF) is a well-established method for answering these questions. It is based on large datasets of completed projects and provides a top-down view: “Projects like this tend to take X months and overrun by Y%.”
But what if you could take a bottom-up view instead? What if you could understand the actual risk in every activity, given the whole context of your specific schedule, and understand the key drivers of delay before you break ground?
Reference class forecasting and nPlan's AI-led schedule risk analysis (AI-SRA) take opposite routes to the question of how long the project will take, and each makes sense at a different point in a project.
INSIGHTS PRO · DRIVING PATHS
See the paths through your own schedule that are most likely to delay your key milestones, alongside the benchmark you already have.
See how Insights Pro worksWhat is the difference between reference class forecasting and nPlan?
Reference class forecasting predicts a project's cost or duration from the actual outcomes of a group of similar completed projects, then adjusts the team's own estimate to match. nPlan's AI-led schedule risk analysis (AI-SRA) forecasts each activity in the project's own schedule, using what our models have learned from more than 780,000 past project schedules, and rolls those forecasts up to key milestones and the completion date.
RCF therefore sets a realistic envelope for the whole project, while AI-SRA shows which activities and paths inside your plan are most likely to move the date.
What is RCF?
RCF comes from Daniel Kahneman and Amos Tversky's research on planning bias and the outside view. Bent Flyvbjerg developed it into a practical forecasting method for projects. It compares a project with a dataset of completed, similar projects, a reference class, to inform cost and schedule forecasts.
The idea is simple: if 80% of projects like yours overrun by 25%, there’s a good chance yours will too. So the forecast starts from what actually happened on past projects.
Big players:
- IPA (Independent Project Analysis) benchmarks capital projects, and an IPA benchmark can form part of an operator's stage-gate process, especially pre-FID. Its dataset spans tens of thousands of projects.
- Oxford Global Projects, led by Bent Flyvbjerg and Alex Budzier, provides RCF data and guidance across infrastructure, IT, mining, and more.
The value: RCF is meant to give decision-makers a reality check. It’s a great way to sanity check optimism bias and answer the question: “Are we being too ambitious?” (John Hollmann discusses schedule contingency).
The method's roots are academic. Flyvbjerg's 2006 paper in the Project Management Journal grounds it in the decision research that won the 2002 Nobel prize in economics and describes its first practical use, for cost forecasts on large transport infrastructure projects.
How is reference class forecasting used in the UK and the US?
How is RCF used on UK public projects?
In the UK, reference class forecasting is built into public-sector appraisal through optimism bias uplifts. HM Treasury's Green Book supplementary guidance on optimism bias says explicit adjustments should be made to estimates of a project's costs, benefits and duration, and it supplies cost and time uplift percentages for generic project categories.
The Department for Transport's TAG Unit A1.2 on scheme costs (May 2025) describes its optimism bias adjustment as an outside view and a top-down approach based on RCF techniques. Its recommended cost uplift for rail schemes, drawn from Oxford Global Projects (2020) data, is 56% at PACE Stage 1, 33% at PACE Stage 3 and 30% at PACE Stage 5 (for local authority and public transport schemes, these map to the Strategic Outline, Outline and Full Business Case), and the guidance requires these uplifts for the economic case.
Illustrative scenario, not a real project. A rail line upgrade has a base capital cost estimate of £400 million at PACE Stage 1. The economic case carries £624 million with the 56% TAG uplift. If the same base estimate held at PACE Stage 5, the 30% uplift would give £520 million.
What the uplift cannot tell the sponsor is which possession, consent or commissioning activity will drive that extra cost. TAG itself expects reliance on optimism bias uplifts to reduce as project-specific risks are better understood and quantified. An activity-level schedule forecast can add to that understanding.
How is RCF used on US capital projects?
In the US, the outside view reaches capital projects through benchmarking and parametric contingency methods. Owners in energy and process industries can benchmark against IPA before authorisation; IPA states that its proprietary databases cover more than 20,000 projects, and its Industry Benchmarking Consortium 2025 findings report that average schedule slip has roughly doubled and now hovers around 18 percent.
AACE International's recommended practice 42R-08 sets out parametric methods, generally empirically based, for estimating cost and schedule contingency. For the schedule itself, the US Government Accountability Office lists a schedule risk analysis as best practice 8 in its Schedule Assessment Guide (GAO-16-89G): it starts from a good critical path method schedule and predicts the level of confidence in meeting the completion date. AACE's 117R-21 covers combining parametric and CPM-based methods, provided steps are taken to avoid double-counting sources of risk. Our article on risk reviews in US transit programs looks at how this plays out for transportation agencies.
What is nPlan AI-SRA?
nPlan starts from the schedule itself and works up from each activity:
- You upload a schedule file (e.g. Primavera P6) with all activities, durations, logic, and constraints.
- nPlan uses AI trained on more than 780,000 past project schedules to forecast how each activity is likely to unfold.
- It considers over 160 contextual features for each task: lag types, float, resource density, calendar logic, and more.
- The result is a probabilistic forecast of the full schedule, from the ground up.
nPlan can be re-run every time the schedule is updated, from pre-FID to commissioning and close out, so teams can re-analyse and act on risk as work progresses.
nPlan learns how different types of work actually perform and applies that to each activity in the context of your schedule, which is why it can forecast first-of-a-kind projects that have no obvious reference class. Schedules can come from Primavera P6, Asta Powerproject or Microsoft Project.
The value: nPlan gives detailed, data-driven schedule assurance based on the schedule's own activity logic, so the project team can respond to change before delays compound.
In Insights Pro, those forecasts feed Driving Paths, which shows the paths most likely to affect a milestone, and Mitigation Strategies, which tests how much a proposed change would move the forecast. Our white paper From QSRA to AI-SRA explains how AI-SRA can take the place of a traditional quantitative schedule risk analysis, with the project team still validating the results and deciding the mitigations.
Top-down vs bottom-up: key differences
RCF answers: “How did other projects like this perform?”
nPlan answers: “Given the actual logic and context of this schedule, how is this project likely to unfold?”
| Aspect | Reference class forecasting | nPlan AI-SRA |
|---|---|---|
| Starting evidence | Outcomes from a relevant set of similar completed projects | Your schedule's activities, logic and constraints, with past performance applied to each activity |
| Level of analysis | The whole project, through a project-level reference class | Every activity, with forecasts that combine into milestone and completion forecasts |
| Comparable projects | Needs a sufficiently similar reference class | Learns how types of work perform, so no near-identical past project is needed |
| Project input | Enough project definition to select a relevant reference class | A logic-linked schedule representing the proposed work |
| Stage of use | Early planning and the business case, including before a detailed schedule exists | From pre-FID schedules through delivery and close out |
| Decision support | An outcome range or uplift for cost and duration | Probabilistic milestone dates and the activity-level risk drivers behind them |
| Review trigger | Stage gates in the business case, new project evidence or a change in the reference class | Every schedule update, with changes in forecast risk reviewed each time |
| Dataset context | The chosen reference class, which must fit the decision | AI that learns from more than 780,000 past project schedules |
When should you use each?
Use RCF when:
- You’re early in the planning process and need a reality check.
- You don’t have a detailed schedule yet.
Use nPlan when:
- You have an L2 schedule at pre-FID and need activity-level assurance.
- You want to pinpoint and understand what will likely cause the delay on your project.
- Your contractor revises the schedule frequently and you need to test each revision.
Can you use RCF and nPlan together?
Yes, and on major projects the two work best in sequence. RCF tests whether the overall budget and duration are plausible; an activity-level forecast then shows where in the schedule that risk sits and what to do about it. AACE's 117R-21, covered above, guards against counting the same risk twice.
Moving from one stage to the next needs a detailed schedule. Schedule Studio turns scope documents into a logic-linked first schedule with WBS, durations and critical path that your planners review and refine, so a bottom-up forecast can start earlier than it otherwise would.
What does this look like on a live project?
On LNG Canada, the export facility at Kitimat, British Columbia, with an initial capacity of 14 million tonnes per annum, the project team used nPlan during construction. In our case study on LNG Canada, the team describes nPlan as "part of our toolkit to help us anticipate risk and capture opportunity on the Project", adding that data from other project schedules helps it focus conversations on the areas that could have the biggest impact. A one-off benchmark is not designed for that kind of work during construction.
Why start sooner?
The earlier you use nPlan, the more confidence you gain in the risk and realism of your plan. You can:
- See where durations look optimistic against how similar work actually performed.
- See which activities are driving delivery risk.
- Stress test alternative schedules before committing.
Wrap-up
RCF and benchmarking have an important role to play in early project assurance: they set the stage, challenge optimism and are established parts of project controls practice in sectors such as transport and energy.
Treating forecasting as a continuous exercise, with an activity-level forecast on each schedule update, keeps the risk view current from the first schedule through delivery.
SCHEDULE STUDIO · SCOPE ASSUMPTIONS AND WBS REVIEW
No detailed schedule yet? Turn your scope documents into a first draft with a WBS and stated scope assumptions your planners can review and refine, ready to forecast.
Try Schedule StudioFrequently asked questions
Who invented reference class forecasting?
Reference class forecasting comes from Daniel Kahneman and Amos Tversky's research on planning bias and the outside view, work that contributed to Kahneman's 2002 Nobel prize in economics. Bent Flyvbjerg turned it into a practical forecasting method for projects, first applied to cost forecasts for large transport infrastructure projects.
Is reference class forecasting required in the UK?
For public projects appraised under HM Treasury guidance, an equivalent adjustment is expected. HM Treasury's Green Book guidance asks for explicit optimism bias adjustments to costs, benefits and duration, and the Department for Transport's TAG Unit A1.2 requires optimism bias uplifts, based on reference class forecasting, in the economic case for transport schemes.
How do US owners get an outside view of their project?
They can use benchmarking against completed projects, such as IPA's databases before authorisation, or parametric contingency methods like those in AACE International's 42R-08. They then test the schedule itself with a schedule risk analysis, which GAO's Schedule Assessment Guide lists as a best practice.
Does nPlan need a similar past project to forecast mine?
No. nPlan forecasts each activity using what our models have learned from more than 780,000 past project schedules about how different types of work perform, so it does not depend on finding a near-identical completed project.
Can nPlan replace a reference class forecast?
No, they do different jobs. Before a detailed schedule exists, RCF or a benchmark gives an outside view; once there is a schedule, nPlan adds an activity-level forecast. Where guidance requires an optimism bias uplift, it still applies. Project-specific evidence can inform any justified reduction under that guidance, with the appraisal team retaining the decision.
Sources
- HM Treasury, Green Book supplementary guidance: optimism bias (April 2013)
- Department for Transport, TAG Unit A1.2 Scheme Costs (May 2025)
- Bent Flyvbjerg, From Nobel Prize to Project Management: Getting Risks Right, Project Management Journal 37(3), 2006
- Independent Project Analysis, What is the key to delivering more predictable projects?
- Independent Project Analysis, resource constraints, staff turnover and project predictability
- AACE International Recommended Practice 42R-08: Risk Analysis and Contingency Determination Using Parametric Estimating
- AACE International Recommended Practice 117R-21: combining parametric and CPM-based methods
- GAO Schedule Assessment Guide: Best Practices for Project Schedules (GAO-16-89G), December 2015
- John Hollmann on schedule contingency (LinkedIn)
- nPlan, LNG Canada case study
- nPlan, From QSRA to AI-SRA (white paper)
- nPlan Insights Pro
- nPlan Schedule Studio
