Quantifying equipment failure risk the way insurers do

This page writes down the method behind the RCM ROI and risk calculator and the reliability business case engagement. It is for maintenance and plant managers who have to defend a budget, and for the finance partners who will read the file. Every result is an estimate from stated inputs. It is not a guarantee of savings, and it is not financial, insurance, or legal advice.

1. Frequency

For a failure mode observed n times in T years of relevant exposure, a simple estimate of the annual rate is:

λ̂ = n / T

If a mode has been seen four times in two years, λ̂ = 2 per year. That point estimate is enough to start, and it is not enough to stop. With small n, the statistical uncertainty is wide. A standard Poisson confidence interval on the rate, at a 90 percent level, is:

[ χ²(0.05; 2n) / (2T) , χ²(0.95; 2n + 2) / (2T) ]

When n = 0 the lower bound is zero and the upper bound still depends on T, which is a reminder that “we have not seen it” is not the same as “it cannot happen.” Write down what period you counted, whether short outages were recorded, and whether the duty changed mid-way through T.

2. Severity

The cost of one occurrence, S, should be built from the event, not from a headline number copied from another industry.

S = (downtime cost per hour × hours lost) + fixed event costs

Downtime cost per hour should count only unrecoverable throughput at contribution margin, plus incremental labor (idle crew, overtime) and other time-based costs. Fixed event costs include things like a premium freight motor, a contractor mobilization, scrap that is already committed, or a cleanup bill. The working is on the downtime cost per hour page. If you already have a figure your finance team accepts, use that figure and cite it.

3. Expected loss

The annualized loss expectancy for a mode is the product of rate and expected severity:

ALE = λ × E[S]

For the synthetic unit-test case used throughout this site: n = 4, T = 2, so λ = 2. Six hours at $1,500 per hour plus $3,000 fixed gives S = $12,000. ALE = $24,000 per year. Add modes by summing ALE values. Do not hide a high-loss mode inside an average of many small ones.

4. Aggregate loss (optional)

A single ALE is a mean. Some finance conversations also want a sense of a bad year. If event counts are modeled as Poisson and each event has cost S (held fixed, or drawn from a severity distribution you can defend), the annual loss is a compound-Poisson total. A Monte Carlo sample of that total gives, for example, a P50 and a P90. The calculator offers this as an option, with severity held at S for transparency. It is a model, not a forecast, and it should not be presented as a plant-specific prediction without a review of the inputs.

5. Credibility weighting

When n is small, plants are tempted to import a “typical rate” from a handbook or a sister site. That can be useful, and it can be misleading. A simple credibility weight is:

Z = n / (n + k), λ_cred = Z × λ̂ + (1 − Z) × λ_ref

k is a judgment about how many observations you would want before you mostly trust your own data. λ_ref must be a documented, cited reference rate — from a named data set, a corporate history, or a published source. This site will not invent a reference rate for you. If you do not have one, say so and use your own λ̂ with its interval, rather than dressing a guess up as an industry number.

6. Pareto of expected loss

Once ALE is computed per mode or per asset, sort it. In many plants a small number of items will carry most of the expected loss. That is a Pareto view of risk, not a law that the split is always 80/20. Use the ranking to decide where the next analysis hour goes. The calculator draws this as a bar chart and as a table, so a reader who does not want the graphic can still see the shares.

7. Risk reduction

A program is worth funding only if residual risk, plus the cost of the program, looks better than the baseline. Residual ALE is written as:

ALE_after = λ (1 − r_f) × S (1 − r_s)

r_f is a fractional reduction in frequency; r_s is a fractional reduction in severity. Both are assumptions. They should be entered as a low, base, and high case. An example used in the calculator — 10 / 25 / 40 percent — is a starting illustration, not a recommended or typical result. For the unit-test case, r_f = 25 percent and r_s = 20 percent give ALE_after = $14,400 and avoided loss = $9,600 per year. Avoided loss is ALE − ALE_after. If you cannot defend a reduction, do not put it in the base case.

8. Finance metrics

Cash flows should reflect that a program rarely delivers its full benefit on day one. A simple ramp used on this site, which you can replace with your own, is 25 percent of steady-state avoided loss in year 1, 75 percent in year 2, and 100 percent from year 3 through the end of the analysis period.

Net benefit_t = (avoided loss × ramp_t) − annual operating cost

Payback (months) = one-time cost / steady-state net benefit × 12

NPV = −C₀ + Σ_t=1..N benefit_t / (1 + i)^t

Use your plant’s hurdle rate for i; the calculator defaults to 10 percent only as a placeholder. A risk-adjusted NPV can be the base NPV multiplied by a success probability p (the calculator defaults to 70 percent, labeled as an assumption), or simply the NPV of the low scenario. Report both if that helps the conversation. The analysis period N defaults to five years.

9. Tornado sensitivity

A tornado chart moves one input at a time — for example, plus or minus 20 percent on downtime cost, on each reduction factor, on program cost, and on the discount rate — and shows how NPV changes. The items with the widest bars are the assumptions worth arguing about in the meeting. If NPV changes sign when a single input moves, say that plainly in the memo.

A worked numeric example (synthetic)

The unit-test case used in the calculator is small enough to check by hand. Observation: n = 4 events in T = 2 years, so λ̂ = 2 per year. Severity: six hours at $1,500 per hour, plus $3,000 of fixed event cost, so S = $12,000. Expected annual loss: 2 × 12,000 = $24,000. A 90 percent Poisson interval on the rate uses χ²(0.05; 8) / 4 for the lower bound and χ²(0.95; 10) / 4 for the upper bound; the calculator prints those bounds next to λ so a reader can see how thin the history is.

Apply assumed reductions r_f = 25 percent and r_s = 20 percent. Residual ALE is 24,000 × 0.75 × 0.80 = $14,400. Avoided loss is $9,600 per year. If a program cost $14,000 once and $1,000 a year to run, the steady-state net benefit would be $8,600 a year and simple payback would be a little under twenty months, before any ramp. With a 25 / 75 / 100 percent ramp, year-1 net benefit is smaller, and NPV at a 10 percent hurdle would be computed from those ramped cash flows. None of those program-cost figures is a recommendation; they are there so you can see the arithmetic. The $24,000 / $14,400 / $9,600 chain is the one the automated test checks.

When you write the memo, put the unit table (n, T, λ, interval, S, ALE) in an appendix, the totals and the range in the first page, and the assumptions in a box a skeptic can find without hunting. If you use credibility weighting, cite λ_ref on the same page as Z. If you do not have a cited reference rate, omit the blend and say that the case uses the plant’s own rate and interval only.

10. Limitations

• This is a model of expected loss, not a guarantee of savings, uptime, or payback.
• It is not financial advice, insurance advice, or legal advice, and it is not an insurance policy.
• CMMS data is often incomplete. Under-counting events understates ALE; counting every short stop as a full event can overstate it.
• Reduction factors are assumptions. They are not empirical results from this practice unless a later, permissioned case study says so.
• Independence between modes is a modeling convenience. Common-cause and cascading failures need a separate discussion.

11. Standards context

ISO 31000 describes a process for identifying, analyzing, evaluating, and treating risk. The same structure can be applied at the asset level or rolled up to a line or a plant, provided you are clear about double counting (a shared spare, a common utility, or a single event that stops several units). Roll-ups are sums of ALE only when the modes can reasonably be treated as additive. ISO 55000 describes asset management as a coordinated activity that realizes value from assets. SAE JA1011 requires a consequence evaluation as part of any process that claims to be Reliability Centered Maintenance. This method is one way to express those consequences in dollars per year. It does not make a program “ISO-aligned” or “JA1011-compliant,” and no insurer endorses it.

When you want the arithmetic done for you, open the calculator. When you want a memo written with your team, see the business case service. For a plant-level first look, start with a Reliability Assessment.