Adam O'NeillLast updated 11 min readPart of Quantitative risk analysis (QRA)

What is Pre-mortem?

A pre-mortem is a risk identification workshop in which the team imagines the project has already failed and each person writes down, individually, why it happened. Gary Klein described it in Harvard Business Review in 2007. The written reasons are then grouped, tested against the risk register, and converted into quantified risks.

The two workshops answer different questions. Most programmes benefit from both.

DimensionConventional risk workshopPre-mortem
Framing questionWhat could go wrong?It is a year from now and the project has failed. What happened?
Output typeRisks sorted into prompt-list categoriesFailure narratives, often specific and unexpected
Group dynamicVocal participants can dominate; politeness can suppress concernsIndividual written input first, so quieter people contribute at the same point as everyone else
Best atCoverage across known risk categoriesBringing out knowledge the team holds but does not volunteer
WeaknessCan miss risks people sense but will not say in publicLess structured, so it can miss whole categories if the team's view is narrow
When to useBuilding the first registerBefore a major investment decision, or when a register looks too tidy

What a pre-mortem is

A pre-mortem is a risk identification exercise described by the psychologist Gary Klein in "Performing a Project Premortem" (Harvard Business Review, 2007). Instead of asking the team what could go wrong, the facilitator asks them to assume the project has already failed and to write down why. The facilitator then collects the reasons, groups them and tests them against the existing risk register.

The underlying research is on what Mitchell, Russo and Pennington (1989) called prospective hindsight. Their laboratory studies found that people produce more correct reasons for an outcome when they imagine it has already happened than when they consider it as something that might happen. Klein applied that finding to project planning. Two limits apply. The research base is laboratory work, and we are not aware of controlled field evidence that pre-mortems improve project outcomes. Treat the technique as a structured way to challenge a plan, not as a proven predictor.

In project controls the pre-mortem sits at the front of the quantitative risk analysis (QRA) chain. Its output feeds the risk register, the register feeds the QRA model, and the model produces the cost and schedule confidence ranges behind a funding decision. A risk found in a pre-mortem moves the P80 only if someone quantifies it and puts it into the model. Until then it is a note.

The wording to use

Klein's published prompt asks the team to imagine that the plan, as it now stands, was implemented and the outcome was a disaster, and to write down the reasons. The version below is our adaptation for infrastructure programmes.

Our script reads: "It is twelve months from now. The project has failed. Costs are significantly over, the programme has slipped and the client has lost confidence. Write down every reason you can think of for why this happened." It names the outcomes a sponsor cares about and leaves the cause open for the team to supply. The 12-month horizon is a default; use whatever period suits how quickly failure becomes visible on the programme.

"What might go wrong?" invites the same cautious replies as an ordinary brainstorm. "What is your biggest concern?" puts one person on the spot and encourages self-censorship. Klein's "disaster" wording is a fair alternative. What matters is that the failure has happened and the cause is left open.

Timed agenda for a 90-minute session

Run the pre-mortem as a standalone session. If it is the last item on a risk workshop agenda, the change of framing is lost. Six to twelve participants works well. With fewer than six you lose the range of viewpoints. Above twelve, the read-out takes too long, so split into two groups and merge the results. The numbers are our practice, not a published rule. Include people who will deliver the work as well as those who plan it, and at least one person who was not involved in writing the plan.

Minutes 0 to 5: context. State the scope, deliverable, budget and programme, then the assumption and the task. Minutes 5 to 15: silent writing. Each person writes up to five reasons, one per card or one per line in a shared form. The facilitator stays silent. Minutes 15 to 25: collection. Entries are typed into a shared sheet or the cards are shuffled and redistributed, so that handwriting does not identify the author.

Minutes 25 to 50: read-out and grouping. The facilitator reads each entry without attribution and groups similar ones on a board. With ten people and five entries each, that is up to 50 entries in 25 minutes, or 30 seconds each, so cap the entries per person if the group is larger. Minutes 50 to 70: register check. For each group, ask whether the register already holds the risk, and if so whether the pre-mortem adds specificity. If it does not, ask whether the risk is new or the register categories are too coarse to hold it.

Minutes 70 to 80: priority vote. Ask: "Which of these would you bet on as the actual cause if the project did fail?" Each person gets three votes. This forces a ranking and produces the shortlist for quantification. Minutes 80 to 90: actions. Name an owner and a date for converting each shortlisted item into a register entry.

Anonymity is our adaptation. Klein's description has participants offering their reasons to the group in turn. We use anonymous written entries because in a mixed-seniority room the junior view is easier to give when it cannot be traced. The cost is that you cannot ask the author to explain, so keep the discussion open for anyone who recognises their own point.

If a participant says they cannot know why it failed, reply once with "make it up - what story would explain it?" If they still decline, move on and do not press. Their silence is information about the room, and the others will still produce enough.

One-page participant handout

Give each participant this on paper or screen. Assumption: it is twelve months from now, the project has failed, costs are significantly over, the programme has slipped and the client has lost confidence. Task: working alone and in silence for 10 minutes, write up to five reasons why, one per card. Write what happened, not what might happen, and say who or what was involved where you can. Nothing is attributed to you.

If someone stalls: What did we assume that proved untrue? Who did we depend on who did not deliver? What did we know early on and not raise? What changed outside the project?

Common pitfalls

Running it too early. Before the scope is defined and the team understands the approach, the output is generic project management risk. The right time is after the team can describe what the project involves and before the plan has hardened to the point where challenging it feels disloyal.

Running it as a group brainstorm. Writing alone is what separates this from a conventional workshop. If people talk during the input stage, the vocal participants set the agenda and the rest defer. Discussion belongs after the writing.

Letting the output go straight into the register. "The contractor underestimated the scaffolding" is a narrative. It has to be rewritten as a cause, event and effect with a probability, an impact range and an owner before it can feed a QRA. Skipping that step is the usual reason pre-mortem output does not change the cost case.

Doing it once. Run another at the start of each major phase or after a significant scope change, because the team knows more by then.

Taking vivid stories at face value. A detailed failure narrative feels likely because it is easy to picture. That is an availability bias, and it can inflate probabilities if you copy them straight into the model. Calibrate against reference-class data or the programme's own history, and record the basis for each number.

Worked example: from narrative to QRA input

Narrative from the session: "The scaffolding contractor underestimated the access needed on the north elevation, and we found out when the first lift was due." Cause-event-effect statement: because scaffold scope was set from the tender drawings, which do not show the north elevation access, the contractor may price and programme for a smaller scaffold than the work needs, resulting in a variation and a delay to the first lift.

Quantification, using figures chosen for illustration. Probability 30 per cent. Cost impact if it happens: minimum £120,000, most likely £160,000, maximum £250,000. Schedule impact if it happens: 2 to 6 weeks, most likely 3. Owner: the package manager. Source of estimates: the contractor's rate build-up and the cost of similar variations on the programme's earlier packages. Record that source, because it is what a reviewer will ask for.

Arithmetic check. A triangular distribution with those cost values has a mean of (120,000 + 160,000 + 250,000) / 3 = £176,667. The expected value is 0.30 x £176,667 = £53,000. That is the average contribution to contingency. It is not what the risk costs when it occurs, which is why the Monte Carlo output, not the expected value, is used to set a percentile-based contingency.

Shared cause. Suppose two other entries say the structural steel arrived late and the cladding fix date slipped, and both trace to the same single fabricator. If the model treats them as independent risks with 25 per cent probability each, it understates the chance that both happen. The probability that both occur is 0.25 x 0.25 = 6.25 per cent when independent, but 25 per cent if they share one cause. Model the fabricator problem as one risk driver at 25 per cent that applies to both activities, with each affected activity responding to it with its own duration or cost range. Where a driver-based approach is not available, apply a stated correlation between the two risks and document the value used and why. AACE Recommended Practice 57R-09, Integrated Cost and Schedule Risk Analysis Using Risk Drivers and Monte Carlo Simulation of a CPM Model, describes the driver method.

Using the output in a business case and at review

The output is useful wherever a QRA supports an approval: the HM Treasury business case process (strategic outline case, outline business case, full business case), National Highways Project Control Framework stages, and Ministry of Defence approvals at Initial Gate and Main Gate. A pre-mortem shows the register was challenged for completeness. It does not replace the review criteria: a review team looks at register quality, the evidence for the ranges, the link to the schedule and how shared causes are modelled.

Keep a short record in the QRA documentation: session date, participant roles, the sanitised failure groups, and the risks added or recalibrated.

How we run it on a QSRA or QCRA

On a QSRA or QCRA we run the pre-mortem before the risk workshop, so its register entries exist before the model is built. We ask for the schedule and current register in advance and check each group against both. The session pack holds the script, the roles present, the sanitised narratives and the new or recalibrated risks, each with a stated basis for its numbers.

Strengthening your QRA function?

SOMA runs QSRA, QCRA and integrated risk analysis to AACE recommended practice: workshop facilitation, calibrated three-point inputs, risk-driver Monte Carlo and a report written so a board can act on the number. Send us your scope and we will tell you what it needs.