Tender contingency is an explicitly approved allowance for uncertain costs the bidder expects to retain under the proposed offer. Build it from identified exposure and a stated estimation basis, after separating required work and funded risk controls. The work product is a risk-to-price decision record: each risk has an event, evidence, probability or scenario range, cost consequences, owner, treatment and a traceable place in the approved price. Some risks require clarification, different terms or a decision not to bid instead of a larger allowance.
A team adds ten percent to the estimate because the contract feels risky. Nobody can explain whether the figure includes extra testing, delivery delays, supplier inflation or a liability concern. Another reviewer removes the percentage to make the offer competitive. Both decisions change the price without establishing what the business will have to pay if the problem occurs. A spreadsheet containing precise decimals can conceal equally weak judgment when its probabilities were guessed.
Name the event before pricing it. Count planned work once, explain the uncertainty that remains, and give the approver both the proposed allowance and the loss it may fail to cover. The calculations below are fictional teaching cases, not market rates, measured failure frequencies or recommended reserves. Contract interpretation, insurance recovery and allowable-cost treatment need review for the actual procurement. An internal allowance neither changes the buyer’s terms nor authorizes an exception.
Cost boundary
Do not hide required work inside a risk percentage
Begin with the work the signed offer would require. Mandatory training, planned security testing and ordinary project management belong in the base estimate. Their effort may be uncertain, but their existence is not an optional event. If the estimator omitted a required test cycle, correct the base. Labelling its entire cost as contingency makes it too easy for a later reviewer to remove work the bidder still owes.
Distinguish uncertainty about the amount of planned work from a separate event. A migration may take between eight and twelve days under the agreed approach. A failed acceptance test may also trigger a second migration. Define whether the upper duration already includes a rerun. Otherwise the model can price twelve days as the base, add a rerun allowance, and charge twice for the same hours. Use a work breakdown detailed enough to expose the overlap.
A planned preventive action has a cost even if the feared event never happens. Put the approved action into the delivery estimate, then assess the remaining risk after that action. The IPA’s UK infrastructure estimating guidance distinguishes base estimates, uncertainty, mitigation and residual risk. Its public-project funding conventions should not be copied automatically into a supplier’s bid price.
An ambiguous obligation needs a question. If the specification does not establish how many sites must be supported, a large reserve cannot decide the scope for the buyer. Seek clarification through the permitted channel and record the response. Where no answer arrives, the commercial and legal reviewers must decide whether a permitted qualification, an explicitly authorized retained assumption or withdrawal is appropriate. Do not silently insert a quantity cap into an unconditional offer.
| Item | Initial treatment | Evidence needed |
|---|---|---|
| Mandatory onboarding | Base delivery cost | Required activities, effort and rates |
| Pre-delivery rehearsal | Planned control cost if approved | Owner, timing and priced resources |
| Possible extra repair visit | Assess residual event allowance | Trigger, likelihood range and incremental cost |
| Unspecified service territory | Clarification or authorized bid decision | Buyer definition and permitted response route |
| Exposure beyond approval limits | Escalation before price release | Contract analysis and competent decision maker |
Evidence
Give every number an event, a time window and a basis
“Implementation risk” is too broad to price. A usable event might be that the buyer’s source files fail an agreed import test during mobilisation, requiring a second preparation cycle. Specify the data condition, the affected milestone and the resources consumed. Establish responsibility from the proposed contract. A cost the buyer must bear under an enforceable mechanism is different from a cost the bidder merely hopes to recover.
Estimate the incremental consequence if the event happens. Separate extra labour, external purchases, standby and any applicable contractual remedy. Check whether the same staff hours appear in more than one consequence. State currency and price date, and use the same tax and inflation basis as the surrounding model. A delay of three weeks is not itself a monetary amount; it needs a supported link to the resources or charges that change.
For probability, retain the reference population and the reason it fits. Three failures among ten vaguely similar projects do not establish a reliable 30% forecast for this contract. Show differences in scope, testing, suppliers and observation period. Where specialists provide judgment, name the role, record a plausible range and explain what information would move it. Do not turn a five-point likelihood label into a percentage by dividing it by five.
Without defensible likelihoods, compare named scenarios and leave probability unassigned. A normal-delivery case and a delayed-delivery case can still identify an affordability boundary. Describe a low-to-high range as scenario bounds when that is what it is. It is not a statistical confidence interval. More simulation runs cannot supply missing evidence, and a model covering named events alone says nothing about losses outside those events.
Worked calculation
An expected loss is not the cash needed in a bad outcome
The fictional Norwick tender has EUR 200,000 of base cost before contingency, profit and tax. Event A adds EUR 20,000 of rework with an assumed probability of 25%. Event B adds EUR 30,000 of delay cost with an assumed probability of 20%. For this first model only, the events are independent, each can occur at most once, and their costs are additive without shared charges. These assumptions are stipulated for arithmetic, not estimated from a real project.
The expected additional loss is 0.25 × 20,000 + 0.20 × 30,000 = EUR 11,000. Yet none of the four possible outcomes costs EUR 11,000. An allowance of that amount covers only the no-event outcome, whose probability in this model is 60%. The mean is useful for comparing alternatives on the same basis. It does not promise that an individual contract will stay within the mean.
Define P90 here as the smallest loss amount for which cumulative probability reaches at least 90%. The loss distribution reaches 60% at zero, 80% at EUR 20,000 and 95% at EUR 30,000. Its P90 is therefore EUR 30,000. The discrete jump means that this amount covers 95% of the modelled outcomes, not exactly 90%. There is still a 5% modelled chance of EUR 50,000 of additional cost.
GAO’s cost-estimating guide, chapter 12, connects contingency choices to uncertainty analysis and management’s accepted risk; it does not prescribe one universally correct confidence level. A bidder must also consider its available funds and portfolio concentration. Do not label a model percentile a guarantee, or assume that a reserve alone supplies the cash or credit needed when a loss occurs.
| Outcome | Probability calculation | Probability | Extra cost |
|---|---|---|---|
| Neither event | 75% × 80% | 60% | EUR 0 |
| A only | 25% × 80% | 20% | EUR 20,000 |
| B only | 75% × 20% | 15% | EUR 30,000 |
| A and B | 25% × 20% | 5% | EUR 50,000 |
Dependence
Keep the same averages and test a different loss pattern
Now change one assumption in Norwick. Suppose delay B can happen only when rework A happens. Keep the marginal probabilities at 25% for A and 20% for B. A coherent joint model has 75% probability of neither event, 5% of A alone, zero of B alone and 20% of both. The same two individual risk entries now produce a different distribution of total loss.
The expected loss remains EUR 11,000: 0.05 × 20,000 + 0.20 × 50,000. But EUR 30,000 now covers only 80% of the modelled outcomes. P90 rises to EUR 50,000 because the combined loss occurs with probability 20%. A review that inspects only each event’s probability and cost, or only the sum of expected losses, misses that change.
Use this comparison to question independence, not to declare that every risk is perfectly correlated. A late buyer decision might trigger both standby and expedited procurement. A common subcontractor might affect several work packages. Conversely, alternative technical failures may be mutually exclusive. Ask delivery owners which combinations are possible and show the mechanism. Do not multiply separate probabilities unless the assumed relationship supports it.
A model can also exaggerate joint loss. If rework and delay both include the same extended project-management week, their impacts are not simply additive. Keep that shared week once in the combined scenario. Document whether dependence changes event frequency, severity or both. When the evidence cannot resolve the relationship, show more than one plausible case and route the resulting funding difference to the approver.
| Measure | Independent case | B occurs only with A |
|---|---|---|
| Expected additional loss | EUR 11,000 | EUR 11,000 |
| Probability of EUR 50,000 loss | 5% | 20% |
| Coverage by EUR 30,000 | 95% | 80% |
| P90 under the stated definition | EUR 30,000 | EUR 50,000 |
Treatment choice
Recalculate after a control, including its own cost
Return to the independent Norwick case. A proposed rehearsal costs EUR 4,000 and is assumed to reduce A’s probability from 25% to 10%, leaving B and the independence assumption unchanged. Expected residual loss becomes 0.10 × 20,000 + 0.20 × 30,000 = EUR 8,000. The control saves EUR 3,000 of expected loss while costing EUR 4,000. Base plus control plus expected loss is EUR 212,000, compared with EUR 211,000 without the control.
That is EUR 1,000 more on a mean-cost comparison. Do not present the control as a cost saving on these assumptions. It may still be required, or justified by an approved delivery benefit that the simple calculation omits. The Orange Book, section D, recognises that treatment decisions involve obligations and consequences beyond monetary savings. State the particular reason for approval rather than inventing a quantified benefit to make the arithmetic look favourable.
If the approver selects EUR 30,000 of residual contingency after the rehearsal, the modelled funding basis is 200,000 + 4,000 + 30,000 = EUR 234,000 before profit and tax. Do not add EUR 8,000 of expected residual loss again for those same events. It is an alternative description of the exposure already funded by the selected reserve. Any additional reserve must name different uncovered exposure.
Risk transfer needs the same discipline. A supplier warranty or insurance policy may reduce loss only within its conditions, exclusions, limits and recovery process. Count premiums and agreed supplier charges once. Retain deductibles, uncovered work, timing gaps and other supported exposure. An asserted right of recovery does not prove immediate payment, and allocating a risk to a partner does not remove the bidder’s customer obligations unless the contract does so.
Price approval
Separate internal funding from contractual permission
Reconcile the allowance against everything already in the estimate. Supplier fixed prices, hourly rates and planned overtime may contain protection for the same events. Ask what they cover without demanding commercially unnecessary detail. Explain each remaining addition. A blanket uplift on the final price can duplicate several carefully priced protections and conceal its own calculation base.
Keep optimism bias separate from a named-event calculation. The Green Book 2026 addresses historical forecasting bias and residual contingency in UK public appraisal. Its government reserve arrangements are not a universal supplier markup. If a company uses experience-based uplifts, explain which forecast errors they address and how those errors relate to the explicit risk model. Do not stack two allowances for the same uncertainty merely because their labels differ.
The procurement’s cost rules matter. US FAR 31.205-7 distinguishes foreseeable future-cost contingencies from effects too imprecise for equitable item-cost estimates; the latter require separate disclosure for contractual treatment. It generally excludes contingencies from historical costing, subject to stated exceptions. Check applicability instead of assuming every internal reserve is reimbursable or every contingency is prohibited.
A fixed-price offer may fund retained risk through the permitted selling-price structure, but the reserve does not itself change scope, liability or remedies. Disclose qualifications only through the procedure’s allowed route and obtain approval before relying on them. Keep internal risk appetite and confidential cost detail restricted except where disclosure is required and authorized. If an obligation exceeds authority, lacks a credible financial boundary or makes delivery unacceptable, stop price release pending a competent decision.
Approval record
Leave a record that survives the next price revision
For each material risk, store the source version, event window, owner, supporting observations and estimation range. Link the selected treatment to its planned cost and residual exposure. Where no probability is defensible, retain the scenario comparison and say so. The approver needs the decision and its basis, including what would make it invalid, rather than a single green status beside an unexplained number.
At total-price level, show base cost, controls, selected allowance and separately defined return. Identify which buyer lines carry the allowance, including allocation across options if relevant. The internal model may be detailed while the required customer schedule is simple. Both must add to the same approved offer without revealing protected working detail unnecessarily.
When an event materialises, transfer its now-known forecast cost into the appropriate delivery estimate and reconsider the reserve for what remains. For example, an agreed repair invoice is no longer merely a possible future repair. Do not add that invoice while retaining its full event allowance without explanation. A release of unused contingency also requires a decision; it is not automatically available for extra scope or a sales discount.
Assign separate authority for accepting risk, changing the price and spending reserve. Set review triggers such as revised buyer data, a changed deadline, a new supplier term or a failed control. The finished record should let a reviewer follow one event into one price decision and see the next action if its assumptions change. It does not authorize buyer contact, a contract exception or submission of the offer.
| Record group | Required content | Review test |
|---|---|---|
| Boundary | Bid version, period, scope and money basis | Is the allowance for this offer? |
| Event and evidence | Trigger, source, likelihood or scenarios, impact range | Can another reviewer challenge the assumptions? |
| Treatment | Control cost, transfer terms and residual exposure | What changes if the treatment fails? |
| Price decision | Amount, existing coverage, price location and uncovered outcomes | Is each cost counted once? |
| Authority and review | Approver, conditions, owner and change triggers | Who must act before the decision is reused? |
What good looks like
Useful outcomes from price risk contingency in a tender
- Required work and planned controls have their own cost lines before contingency is calculated.
- Each retained exposure has a defined event, estimation basis and responsible owner.
- The approver sees how dependence and severe outcomes affect the proposed funding.
- Every allowance is reconciled to the price without adding the same risk twice.
- Unpriceable or unauthorized obligations receive an explicit decision before submission.
Operating model
How to run the work
- 01
Fix the cost boundary
Identify the offer version, work, period, currency, price date and included costs. Separate required delivery, planned controls, uncertain events and unresolved contract meaning.
- 02
Describe retained events
Record what may happen, why, when and which incremental costs follow. Attach the relevant scope or term and establish who would bear the loss.
- 03
Challenge the estimates
Document probability and impact ranges with their evidence, missing facts and dependencies. Use scenarios without invented likelihoods when the evidence cannot support a distribution.
- 04
Compare treatment options
Price feasible controls, confirm any transfer and recalculate residual exposure. Escalate risks that exceed authority or cannot be made acceptable.
- 05
Approve the price treatment
Explain the selected allowance, uncovered outcomes and where the amount enters the buyer’s schedule. Check existing allowances, cost rules and exception requirements.
- 06
Reconcile later changes
Assign review triggers and spending authority. When an event becomes known work, update the cost forecast and remaining reserve together instead of counting both.
Evaluation
Questions that change the decision
- Is this required work, an uncertain event or an unresolved obligation?
- What evidence supports the likelihood and incremental cost?
- Can the event occur alongside other priced risks?
- Does a funded control or confirmed transfer reduce this exposure?
- Which losses remain outside the chosen allowance?
- Who may accept the residual risk and approve a changed price?
Failure modes
Where teams lose control
Known work is hidden in a percentage and removed during negotiation.
An ordinal risk score is treated as a measured probability.
A common cause creates simultaneous losses omitted from the model.
The same exposure appears in supplier prices, base rates and contingency.
A reserved sum is mistaken for permission to ignore a contractual duty.
An internal funding decision is treated as a right to invoice the buyer.
Measurement
Measure the finished job
Measure the completed workflow, including review effort and exceptions. Output volume on its own is not evidence of a better process.
- Retained risks with documented evidence and estimation ranges
- Allowance amounts traced to approved price locations
- Unresolved high-impact risks awaiting an authorized decision
- Exposure counted in more than one cost component
- Material assumption changes awaiting reserve reconciliation
Questions
Common questions
What percentage should a tender include for contingency?
There is no defensible universal percentage. Start with the offered scope, retained events and evidence. An approved percentage may summarize a supported calculation, but it must state its cost base, covered risks and exclusions. A number chosen only because it fits the target price does not establish adequate funding.
Can I use probability multiplied by impact?
For a defined event with a supported probability and conditional cost, that gives its expected loss. It does not identify the reserve required for a chosen level of loss coverage. Check joint outcomes, overlapping consequences and severe cases before turning the sum into an approval recommendation.
What if the probability cannot be estimated?
Show plausible scenarios, the evidence gap and the cost boundary each scenario creates. Do not manufacture percentages or call scenario endpoints a confidence interval. Assign the missing investigation or seek a commercial decision that explicitly recognises the uncertainty.
Does insurance remove the need for contingency?
Only the exposure demonstrably covered by the applicable policy can be treated as transferred, and recovery conditions still matter. Assess deductibles, exclusions, limits, delay in payment and remaining delivery costs. Include the premium once and obtain qualified advice on coverage.
Is contingency the same as profit?
No. The internal allowance funds specified uncertain costs; return is a separate commercial choice. Unused allowance may affect the eventual result, depending on actual costs and contract terms. Its existence does not prove a particular accounting profit or permit an unsupported cost claim.
Should the buyer see the entire risk register?
Follow the required disclosure rules and approved confidentiality boundaries. The buyer may require particular assumptions, breakdowns or qualifications. That does not justify releasing every internal cost floor, risk limit or partner detail. Ensure the submitted price and statements remain accurate.
When should pricing stop instead of adding more reserve?
Pause when mandatory scope is unresolved, the exposure exceeds delegated authority, a necessary qualification is not permitted, or delivery remains unacceptable despite funding. Name the unresolved decision and its owner. A higher price cannot create permission or make an impossible obligation deliverable.
Sources
Primary references
- Cost estimating: base, uncertainty and residual risk Infrastructure and Projects Authority
- Cost Estimating and Assessment Guide, chapter 12 U.S. Government Accountability Office
- Orange Book, section D: risk assessment and treatment HM Treasury
- Green Book 2026: uncertainty, optimism bias and contingency HM Treasury
- FAR 31.205-7: contingency cost treatment U.S. Federal Acquisition Regulation
About the author
Tony Kim
Tony writes about applied AI, dependable product engineering and the systems that turn complex response work into controlled delivery.
Zelius
Managed tender intelligence and bid execution for teams that want the commercial outcome.
Suppliers, founders and commercial teams pursuing public or private opportunities. Start with the workflow, constraints and evidence you already have.