What Is the One Assumption Most Likely to Make Your Project Unfundable?

Early in my time working across project finance in the Asia-Pacific, a colleague handed me a model for a renewable energy project that had already reached financial close and drawn its first tranche. Eighteen months later, its lender had called an event of default – not because the technology failed, not because the team was incompetent, but because a single offtake price assumption had been stress-tested too leniently. The project team had built a financial model that looked sound. What they had not done was ask honestly what happened when reality diverged from their base case by more than a comfortable margin.

That gap – between a model that looks fundable and a deal that actually is – is precisely where sensitivity analysis lives. Put simply, sensitivity analysis is the disciplined study of how changes in your input assumptions drive changes in your project’s outputs: IRR, DSCR (debt service coverage ratio), NPV, payback period. It answers the question capital providers are always asking, even when they are too polite to ask it directly: if your best case proves optimistic, does this project still work?

Strong projects do not fail because of weak fundamentals. They fail because capital and structure do not meet at the right time – and one of the most common reasons that meeting goes wrong is that nobody tested the model against the range of conditions the real world reliably delivers. This article walks through the methods, the common failure points, and the practical discipline that separates a fundable deal from a rejected one.

Why Sensitivity Analysis Matters More Than You Think

The misconception worth clearing up immediately: sensitivity analysis is not a post-hoc validation exercise performed after the model is finished and the IM (information memorandum) is being drafted. That framing relegates it to a compliance checkbox, and in that role it catches nothing useful. Lenders and capital raising advisors who have reviewed hundreds of project models know the difference between a sensitivity section built into the model architecture from day one and one appended at the end to satisfy a checklist. The former shapes the deal structure. The latter decorates it.

Capital providers – whether development finance institutions, infrastructure funds, or long-term strategic investors – ask for sensitivity analysis because they have seen deals collapse when a single assumption proved fragile. A copper project where offtake price assumptions, varied by fifteen percent in either direction, shift the project from bankable to distressed is not a hypothetical. It is a pattern repeated across commodity cycles in Chile, the DRC, and the Asia-Pacific. In each case, the model had a base case. What it lacked was honest interrogation of that base case.

The practical standard is straightforward: every assumption in your model that carries real-world variability should be tested. Revenue assumptions vary because commodity prices cycle, demand fluctuates, and offtake counterparties renegotiate. Cost assumptions vary because construction materials, labour, and logistics are inflation-sensitive. Financing assumptions vary because debt markets move. Sensitivity analysis is the mechanism by which a model built on a single set of inputs becomes a decision-support tool built on a realistic range of outcomes. That is what an investment-ready model actually means – not that the numbers look good, but that the model can withstand challenge.

Local Sensitivity: When One-Factor-at-a-Time Is Enough (and When It Isn’t)

The most common starting point for sensitivity analysis in project finance modelling is the OFAT method – one-factor-at-a-time. The logic is intuitive: hold all other inputs constant, vary a single input across a defined range, and observe how the output responds. The result is a clean, readable relationship between that variable and your project’s returns. Run this across your five or six most important assumptions and display the results in a tornado diagram – a ranked bar chart that places the most influential variable at the top and the least at the bottom – and you have a fast, stakeholder-friendly picture of where your risk is concentrated.

Local methods are fast, transparent, and particularly well-suited to early-stage model diagnostics. For a simple, near-linear revenue-cost model – a single-product facility with a fixed offtake agreement and predictable operating costs – OFAT analysis often tells you most of what you need to know. Sponsors presenting to a non-technical project committee will find a tornado diagram far more persuasive than a page of variance decomposition statistics. Understanding how different model types map to different fundraising contexts can help sponsors choose the right analytical framework before sensitivity testing begins.

The trap, however, is a real one. OFAT analysis misses interactions – situations where two inputs move together in reality but the model treats them as independent. Consider a commodity-linked infrastructure project where diesel price and shipping cost are correlated: when global energy prices spike, both rise simultaneously. An OFAT analysis that varies diesel price and shipping cost in separate runs, holding the other constant each time, will systematically underestimate the combined downside scenario. The model looks more robust than it is. Lenders who understand this will probe exactly this point.

The practical discipline is to start with OFAT and the tornado diagram – they are fast, readable, and will immediately identify your two or three most sensitive assumptions. Then ask honestly whether those assumptions are independent of each other in the real world. If they are not, the analysis needs to go deeper.

Global Sensitivity: Testing What Really Matters in a Complex Deal

Global sensitivity analysis (GSA) varies multiple inputs simultaneously across realistic ranges, capturing not just the direct effect of each input on the output, but also the interactions between inputs. For complex, non-linear project models – energy projects with variable generation profiles, mining projects with grade variability, infrastructure projects exposed to regulatory change – global methods are not optional sophistication. They are the standard expectation of institutional capital providers and development finance institutions.

The most rigorous global framework is variance-based decomposition, associated with the work of Andrea Saltelli and colleagues, which produces what are known as Sobol indices. These are numbers between zero and one that tell you what fraction of your output variance is driven by each input and its interactions. A first-order Sobol index of 0.60 for offtake price means that price changes alone account for sixty percent of the variability in your project’s IRR. The total-order index goes further: it adds the contribution of that input through all its interactions with other variables. When the gap between a variable’s first-order and total-order index is large, you are looking at a model where interactions matter – and where OFAT analysis would have given you a misleading picture of where the real risk sits.

What is equally important to understand is that global sensitivity analysis reframes the question sponsors should be asking. It is not merely "what is my most sensitive variable?" It is "which of my assumptions, in combination, determines whether this project survives a bad year?" For an energy project seeking long-term project finance, a Sobol analysis that reveals offtake agreement terms – rather than technology cost – as the dominant driver of IRR variance is commercially actionable information. It tells the sponsor where to concentrate due diligence, where to negotiate harder, and where the deal structure needs reinforcing before it goes to a lender.

A 2023 review of environmental and hydrological modelling practice noted that contemporary sensitivity frameworks increasingly emphasise comprehensive sensitivity matrices and statistics to identify sensitive regions and detect unusual model behaviour – a methodological discipline that translates directly to financial model governance for capital-intensive projects. The principle is the same whether the model is simulating rainfall runoff or a thirty-year power purchase agreement: understand which inputs drive your outputs, and understand which combinations of inputs can break your model.

The practical trade-off is real. Sobol analysis requires substantially more computational effort than OFAT and demands that the analyst define credible probability distributions for each input – a task that requires domain expertise and honest engagement with historical data. For a greenfield lithium project in Argentina or a biomass energy facility in Southeast Asia, that effort is appropriate. For a simple two-variable revenue model, it is not. The choice of method should match the complexity of the project and the expectations of the capital provider.

The Assumptions That Most Frequently Tank Projects in Capital Raising

Across project finance engagements in energy, mining, and infrastructure – from geothermal projects in LATAM to port developments across the Asia-Pacific – certain assumptions recur as the most common points of failure when capital providers stress-test a model:

• Offtake price and volume. This is the single most common culprit. Sponsors frequently assume stable demand or pricing in markets that are cyclical by nature – copper, LNG, electricity, agricultural commodities. A model that shows acceptable returns only when prices remain at or above the prevailing spot rate is not a fundable model. It is a best-case scenario dressed up as a base case. Sensitivity analysis must test offtake price across the range the market has actually delivered over the past decade, not merely the range the sponsor finds comfortable.

• Construction cost escalation. Projects in capital-intensive sectors routinely underestimate the combined effect of labour cost increases, materials inflation, and timeline extension. A ten to twenty percent upside scenario on total project cost is not pessimism – it is history. Any model that does not test this range explicitly will face hard questions during due diligence.

• Financing structure and available capital. Early-stage models frequently assume access to debt or equity at rates and leverage ratios that may not materialise at the time of financial close. Sensitivity to the cost of debt, the debt-to-equity ratio, and the availability of refinancing on reasonable terms is essential in any model being presented to a lender or a capital raising consulting panel.

• Sovereign risk and regulatory stability. Projects in emerging markets – across LATAM, parts of the Asia-Pacific, and Sub-Saharan Africa – often underweight the probability of policy shifts, tariff changes, or enforcement delays. The history of independent power producers across multiple markets provides ample evidence that regulatory assumptions embedded in a thirty-year model should be tested against scenarios where the rules change. Sovereign risk is not a background condition. It is a model input.

• Working capital and operational cost inflation. Sponsors frequently assume flat operating costs over multi-year horizons when energy prices, labour, and supply-chain costs have demonstrated persistent volatility. A sensitivity run that holds OPEX flat for twenty-five years is not a model. It is a wish.

In each case, the discipline is the same: define a realistic range based on observable historical data, test the model across that range, and show the results – including the uncomfortable ones. It is important to remember that when a sponsor demonstrates to a lender the model has been tested honestly against these scenarios and the deal still holds, that transparency is worth more to the relationship than a flawless-looking base case that has never been challenged.

Sensitivity Analysis as Part of Your Deal Governance and Due Diligence

The common practice of attaching a sensitivity table to an investment memorandum as a final appendix reflects a fundamental misunderstanding of what sensitivity analysis is for. It belongs in the model architecture, in the term sheet conversation, and in the independent assessment – not as a presentation slide, but as a governance discipline embedded in how the deal is structured and documented.

Best practice is to wire sensitivity workflows directly into the model so that when a term changes – offtake price, cost per tonne, debt tenor – the sensitivity outputs update automatically. The financial model is the single point of truth, and sensitivity analysis is not a separate document appended to it. It is a layer of the model itself. That is what structuring for capital actually looks like in practice. The strategic role a well-constructed financial model plays across the full investment lifecycle – from early structuring through to financial close – makes this integration a governance imperative, not merely a technical preference.

For M&A and refinancing transactions, sensitivity analysis often functions as the bridge between a sponsor’s base case and a lender’s conservative case – a middle path between competing projections that allows both parties to understand the range within which the deal can be renegotiated without breaking. Red flags that suggest weak sensitivity work include models that show the project is equally profitable across a fifty percent range of inputs (implying the model is insensitive where it should not be), and conclusions that hinge entirely on a single best-case scenario without any honest stress-testing.

The regulatory and transparency expectation in sustainable investment and green bond issuance has tightened considerably. Sensitivity to policy risk and carbon price variability is now standard documentation for green instruments and ESG-aligned project finance. Name your scenarios – base case, stress case, extreme case. Define the input ranges used and explain why those ranges are realistic. Show your working, and invite independent professionals to challenge it. That invitation, consistently extended, is what healthy financial discipline looks like in practice.

Practical Tools and Workflows: From Spreadsheet to Sobol

The choice of tool should be driven by model complexity and the expectations of the capital provider, not by the analyst’s preference for sophistication.

For most early-stage project finance models, Excel’s data table feature is the correct starting point. It allows the analyst to vary one or two inputs simultaneously and display the results in a matrix – fast, transparent, and directly presentable in stakeholder meetings without requiring any specialist software. Build the tornado diagram alongside it, and the key risk drivers are immediately visible to a non-technical board or investment committee.

When the project is complex, non-linear, or requires probability distributions rather than single-point estimates, Monte Carlo simulation becomes the appropriate tool. Tools such as @RISK and Crystal Ball integrate directly into Excel and allow the analyst to define probability distributions for each input – triangular, normal, log-normal – sample thousands of combinations, and produce output distributions showing the range of IRR or NPV outcomes the project might realistically deliver. The output is a probability distribution, not a number. That distinction matters to a lender who needs to understand the probability that the project’s DSCR falls below covenant levels in any given year.

For research-grade rigour or institutional capital raises where Sobol indices are expected, Python’s SALib (Sensitivity Analysis Library) and comparable R packages provide the technical infrastructure. These are appropriate for complex simulation models – geothermal resource models, integrated mine-to-port logistics models, large-scale grid-connected renewable energy projects – where the number of interacting parameters makes variance decomposition genuinely informative rather than decorative. Analysts seeking to deepen their command of the techniques that make Monte Carlo outputs genuinely useful will find that the principles underpinning effective financial modelling for fundraising apply directly to how sensitivity workflows should be structured.

The pragmatic workflow is straightforward: start with OFAT and the tornado diagram to identify the key drivers, validate the results against domain expertise (if the tornado diagram says your most sensitive variable is one the engineering team considers well-constrained, that is worth examining), and deepen with Monte Carlo or global methods if the model complexity warrants it or if the capital provider demands it. Define input ranges before running the analysis, not after reviewing the results. Define first, then test. The sequence matters.

When Sensitivity Analysis Reveals Your Deal May Not Be Fundable

Honest sensitivity analysis sometimes delivers uncomfortable conclusions. A project that looks viable in the base case reveals, under stress testing, that acceptable returns depend on commodity prices staying elevated, costs not escalating, and regulatory conditions remaining stable – simultaneously. Real projects face all three risks at once.

The response to this finding is not to obscure it. Sponsors who have tried to present a model with weak sensitivity outcomes beneath a strong-looking base case have almost uniformly discovered that lenders and investors conduct their own stress-testing – and their scenarios are typically more aggressive than the sponsor’s. Presenting a model that cannot survive those scenarios, without having addressed them yourself, damages credibility in ways that are very difficult to repair at the term sheet stage.

The more productive response is to treat poor sensitivity outcomes as design feedback. A mining project that shows poor sensitivity to ore grade assumptions in the early modelling phase – a situation that arises with some regularity across engagements in APAC and LATAM – presents a clear set of options: bring in a partner with geological expertise and a stronger resource database, structure a phased investment schedule that gates later tranches on confirmed grade, or reduce leverage to create more buffer against downside scenarios. In each case, the sensitivity analysis did not kill the project. It showed where the deal needed to be redesigned before it could be funded.

Sponsors who engage unfavourable sensitivity results early – and act on them with discipline and clarity – tend to arrive at financial close with stronger deals, better lender relationships, and more durable capital structures than those who present polished base cases and hope the lender doesn’t look too hard. Transparency about a model’s weak points builds earned trust. That trust, once established, is the foundation on which term sheets are negotiated and relationships endure beyond the closing dinner.

The forward question for any project where sensitivity analysis reveals marginal outcomes is always this: what term sheet changes, capital structure adjustments, or partnering arrangements would make this fundable? That is the question capital raising consultants should be asking from the very first model run – not as an afterthought, but as a governance discipline built into how the project is designed and presented from day one. A robust financial model, built with that question at its centre, is the only reliable way to ensure it gets answered honestly enough to matter.

Frequently asked questions

What is sensitivity analysis, in simple terms?

Sensitivity analysis is the study of how changes in your model’s input assumptions – price, cost, timeline, demand – affect the output, whether that is profit, IRR, or payback period. It answers: if the offtake price drops ten percent, does the project still work? Lenders and investors ask for it because they need confidence that the deal performs across a realistic range of conditions, not just the single best-case scenario the sponsor finds most comfortable. It is the mechanism by which a model becomes a decision-support tool rather than a presentation document.

What are the main types of sensitivity analysis?

The two broad categories are local and global methods. Local sensitivity analysis – including one-factor-at-a-time (OFAT) testing and tornado diagrams – varies one input at a time and is fast and intuitive. Global sensitivity analysis – variance-based methods including Sobol indices, and Monte Carlo simulation – varies multiple inputs simultaneously and captures interactions between parameters. Local methods suit simple or linear models and early-stage diagnostics. Global methods suit complex projects or situations where institutional capital providers expect research-grade rigour. Most project finance engagements warrant both, applied in sequence.

Why is sensitivity analysis essential for project funding?

Capital providers need confidence that a project will perform across a realistic range of conditions, not just the sponsor’s preferred assumptions. Sensitivity analysis shows which assumptions are critical, where risk is concentrated, and whether the deal structure has enough resilience to absorb downside scenarios without breaching covenant thresholds. Projects that arrive without rigorous sensitivity analysis are typically perceived as higher-risk – priced accordingly or rejected outright. The analysis also demonstrates governance discipline, which matters to long-term strategic investors who are evaluating a relationship, not just a transaction.

What happens if I skip sensitivity analysis?

The model may look fundable in the base case. But lenders and investors will test robustness themselves – and their stress scenarios are usually more aggressive than the sponsor’s. Skipping sensitivity analysis signals either overconfidence or weak governance, and either reading damages credibility at the term sheet stage. In practice, deals built on transparent and thorough sensitivity outcomes tend to attract better financing terms and faster closings than those built on unchallenged base cases. The analysis does not add risk to the deal. It reveals risk that was already there.

What is a Sobol index, and do I need it?

Sobol indices are numbers that quantify what fraction of your output variance is driven by each input and its interactions with other inputs. A first-order Sobol index of 0.60 for offtake price means that price variability alone accounts for sixty percent of your IRR variability. The total-order index adds interaction effects. Most early-stage project models do not require full Sobol decomposition. It becomes relevant when the model is non-linear, has many interacting parameters, or when development finance institutions or infrastructure funds expect variance-based sensitivity as part of their investment committee documentation.

How do I choose between a tornado diagram and a full Monte Carlo analysis?

Start with the tornado diagram: it is fast, visual, and immediately communicates which inputs drive the most risk. Use it in early-stage presentations and to guide due diligence priorities. Move to Monte Carlo if the model is complex, non-linear, or if lenders require probability distributions of outcomes rather than single-point estimates. The pragmatic path is: tornado diagram first to identify key drivers, then Monte Carlo if the deal complexity or the capital provider’s expectations warrant the additional rigour. Most sponsors underuse tornado diagrams and underinvest in the model architecture that would make Monte Carlo genuinely informative.

What if sensitivity analysis shows my project is marginal or unfundable?

Treat it as honest feedback and a design brief, not a failure. Use the results to restructure the deal: lock in longer-term offtake agreements to reduce commodity price exposure, reduce leverage to improve downside margins, or bring in partners who can hedge key risks more effectively than the sponsor alone. Sponsors who engage unfavourable sensitivity outcomes early and act on them with discipline and clarity consistently end up with stronger, more fundable deals. Transparency about model weaknesses builds earned trust with investors and lenders – and that trust is more durable than any base case projection.

Which software tools should I use for sensitivity analysis?

Excel data tables and tornado diagrams are sufficient for many local sensitivity tasks and remain the most transparent format for stakeholder meetings. For Monte Carlo analysis, @RISK and Crystal Ball integrate directly with Excel; SimulAr is a capable free alternative. For global variance-based methods, Python’s SALib library and comparable R packages provide research-grade capabilities accessible to analysts with basic coding experience. The choice depends on model complexity, available expertise, and whether the capital provider expects probability distributions or simply wants to see that the key assumptions have been honestly stress-tested.

Sensitivity analysis is not a technical nicety reserved for large infrastructure transactions or development finance engagements. It is the discipline that separates a model built to impress from a model built to survive – and in project finance consulting for complex projects, that distinction is the difference between reaching financial close and receiving a politely worded rejection. It is important to remember that the assumption most likely to make a project unfundable is not always the most obvious one. Sometimes it is the interaction between two assumptions that look safe in isolation but become dangerous in combination. The only reliable way to find it is to look honestly, early, and with the right tools.

If the financial model is the single point of truth – and for every project worth doing, it must be – then sensitivity analysis is the test of whether that truth holds under conditions the real world actually delivers.