Making the Right Decisions in Uncertain Times: How Many Futures Are Enough?

The traditional base/high/low approach to forecasting can’t absorb today’s uncertainty. Energy markets are more volatile and interconnected than ever, raising the stakes for every long-term resource decision. Load growth, shifting policies, fuel prices, and infrastructure challenges are all moving at once, often in conflicting directions.

If you get a 20-year resource plan wrong, the costs show up on customer bills, in system reliability, or as stranded assets. Confidence in a resource plan doesn't come from guessing everything correctly. It comes from watching a decision hold up even when some assumptions turn out wrong. So how many futures do you need to test before you can defend the decision you are making? 

Why Three Cases No Longer Covers It

The base/high/low convention was built for a world where a handful of variables moved slowly and mostly independently. It served resource planners well when load growth was relatively flat and traditional generation resources ruled the day, but running just three scenarios is increasingly inadequate for the range of forces now at play.

The Variables Multiplying at Once

With load, generation, and policy all shifting at once, today’s energy market is characterized by complexity and uncertainty.

• Skyrocketing demand: Data centers and large loads could drive peak load growth up between 5% and 25% in the next five years.

• Demand uncertainty: Load forecasts are high, but in reality, existing data centers are consuming less than their nameplate power capacity. In addition, some data centers in the connection queue may never materialize. This calls into question just how much supply will be needed to meet future demand, and it increases uncertainty about how much demand is real and how much is “phantom demand.”

• Changing generation mix: Aging fossil fuel plants are shuttering and being replaced, in many ISOs, with intermittent renewable resources.

• Fuel price volatility: Fluctuations in diesel and natural gas prices complicate planning.

• Policy uncertainty: Federal requirements, particularly around emissions and environmental impacts, can shift with each new administration, making it hard to predict future mandates. In September 2026, for example, a federal appeals court vacated the Department of Energy’s emergency order that had kept Michigan's J.H. Campbell coal plant running past its planned retirement date. Separately, several states have sued over federal agreements that cancel offshore wind leases, leaving that future supply uncertain.

Given today’s environment, three scenarios provide a useful starting point for planners, but not much flexibility around your reference case. When six or eight variables each have a plausible range, you can’t tell which variable drove the difference in your outcome. Plus, reality could just as easily land near the high or low end as it could hit your base case.

What Each Level of Analysis Tells You

Depending on the stakes and the time you have, a study might rest on one scenario, three scenarios, 20 sensitivities, or several hundred. Each level answers a different question.

One Scenario: Confidence in a Single Future

When the decision at hand isn't particularly sensitive to underlying assumptions, or when the deliverable itself calls for one trusted forecast rather than a range of possibilities, running a single scenario is legitimate. In these cases, building out multiple scenarios would add complexity but not much value.

The limitations, though, are significant. A single scenario tells you what the optimal path looks like if your assumptions hold, but it tells you nothing about what happens if they don't. There's no sense of exposure, no stress-testing, no visibility into how fragile or robust that plan actually is. You're optimizing for one version of the future and hoping it's the one that shows up.

It doesn't prepare you for being wrong — and in long-term planning, being wrong about at least some assumptions is nearly guaranteed.

Answers the question: “What happens if we’re right?”

Three Scenarios: Comparing Alternatives

The traditional base/high/low convention lets you bound a decision between a defined set of alternative futures, rather than betting everything on a single forecast. It provides a sense of range and shows you what the resulting portfolios have in common. Shared resources or strategies across scenarios often point to resilient choices.

The limitation is that those bounds are only as good as your own assumptions. Since you're defining the three scenarios, the analysis is inherently limited by your imagination. If you fail to anticipate a plausible future, it simply won't be captured. With three scenarios you learn the spread between your own assumptions, not the shape of your actual risk.

Answers the question: “Which of these options looks better?”

20-Plus Futures: Understanding Risk

When you move individual levers, or small sets of levers, rather than swapping out the entire baseline scenarios, you can see how different futures affect your outcomes. Variables could include demand, fuel prices, political changes, renewable capacity, and more. For example, you could isolate a single variable, like renewable capacity, and model a high-renewables future to see how that shift alone changes your outcomes.

However, levers rarely move in isolation, and capturing their interconnectedness is challenging. For instance, if you assume a large build-out of solar resources, you also have to consider whether that surge in demand for solar panels would drive up their costs. Many drivers need to move together to reflect a realistic future, but in practice, the assumptions behind them can conflict, making that kind of coordinated modeling difficult to get right.

Given that complexity, ranking uncertainties by consequence becomes essential, as it enables you to further hone your risk assessment and future priorities.

Resource planning teams often get bogged down in runtimes and manual output comparisons at this stage.

Answers the question: “Which assumptions actually change the answer?”

Hundreds of Futures: Finding Resilient Decisions

When running hundreds or even thousands of sensitivities, the focus shifts to risk assessment: you run your resource plan against many different possible futures, including those it wasn't originally optimized for, to see how it performs under different conditions. For example, if the portfolio ended up gas-heavy because it was optimized under a low-gas-price scenario, the key question becomes: what's the biggest risk to that plan if the actual future turns out to be a high-gas-price environment instead?

If a resource performs well in 10 out of 20 sensitivity analyses, you may have reasonable confidence in it. But once you increase your sample size and run 200 sensitivities and it performs well in only 10, that plan looks much weaker. A larger sample also surfaces the tail risk. At this point, the output stops being a forecast and becomes a decision test.

Answers the question: “Which decision holds up across most futures?”

More Sensitivities Isn’t the Point

More sensitivities generally improve the accuracy and reliability of your risk assessment, but only if they are well designed. Volume alone isn’t a virtue. A thousand poorly designed runs are worse than 20 well-chosen ones, because sheer quantity can create false confidence and bury the signal.

Rather, the goal should be uncovering which uncertainties materially affect the decision. Scale removes the need to guess which variables matter in advance; instead, you can find out empirically.

When you're evaluating the resulting resource plan, you typically run multiple candidate plans through the same set of sensitivities. Naturally, each plan responds somewhat differently to the same sensitivities, but the ones that show the least variation and remain the most stable are worth paying attention to.

A portfolio might look like the best choice when you're only looking at a single future outcome. But realistically, no single outcome will play out exactly as modeled over a 20-year horizon; it's more likely some combination of many different factors will occur. That's why the goal is to find the portfolio most resilient to a range of volatile changes. The more sensitivities and variables you test, the more thoroughly you can stress-test the portfolios, giving you confidence in the resource plan.

Economics aren't the only factor at play. As intermittent resources make up more of the grid, making sure the lights stay on during a cloudy, windless stretch matters just as much as cost when you're evaluating resource plans. So run sensitivities across reliability metrics for each candidate plan as well. The deeper reliability work, like an effective load carrying capability (ELCC) analysis, is a separate study, but the question it answers belongs in the integrated resource plan.

What Used to Make Broad Exploration Impractical

Until recently, this type of broad exploration wasn’t practical for many teams.

Workload was often the main constraint. Teams would have liked to study more scenarios but didn’t have the time or staffing to do it. If you wanted to run multiple scenarios with hundreds of sensitivities, for instance, someone would need to set up each one manually, adjust the model inputs, hit run, and then pull and review all the sets of results one at a time. This process often relied on fragile handoffs between forecasting, data, and modeling tools, where a single misstep could mean redoing hours of work.

Scaling up also came at a price. Run times on local or on-premises systems could stretch a single sensitivity out for hours, and license or core limits often capped how many runs could be executed at once. This forced teams to work through scenarios sequentially rather than in parallel, and the only solution was to add dozens of servers at significant expense.

How Cloud Modeling Removes the Constraint

Cloud computing, which lets teams access virtually unlimited computing power on demand, has rewritten the playbook for resource planners. Many of the resources required to build and maintain infrastructure are reduced or entirely eliminated with the cloud.

Power system modeling software, such as Yes Energy’s EnCompass, now allows teams to run hundreds of futures simultaneously without hitting the hardware limits that once forced sequential, batch-by-batch execution. Whether a study calls for 10 sensitivities or 400, the process and the time required to create these are the same. The system handles the added computational overhead behind the scenes rather than forcing teams to split runs across multiple machines.

Building scenario inputs is also faster. Creating 200 different gas price distributions once meant a statistician had to build each one by hand and enter it into the model manually. API-driven workflows can now automate that work within the platform itself, creating and inputting the distributions for you.

Cloud-based energy market simulation tools also reduce friction between forecasting, data, and modeling systems that once operated independently, requiring teams to rebuild inputs at every handoff. By integrating data and assumptions in one system, teams can join and study the results of numerous cases, spending less time reconciling models and more time analyzing results.

The outcome is faster studies, fewer errors, and decisions you can confidently defend. And, because the process is automated, integrated, and transparent by design, results stay traceable, giving teams defensible proof that a plan holds up across many possible futures, not just a single reference case. That kind of evidence is increasingly what regulators and stakeholders expect before approving major investments, and it reduces the risk of committing capital to a plan that falls apart outside the future for which it was built.

Questions to Ask Before Your Next Planning Cycle

Before locking in your next resource plan, it’s worth stress-testing the process that produced it.

1. Where does our key exposure to uncertainty lie?

2. How many futures would we actually need to run to feel confident in our results, and are we currently running enough?

3. If we're under a time crunch, would moving to the cloud let us expand the number of scenarios or sensitivities we test, rather than settling for fewer than we should?

4. Which assumptions in our last plan would have changed the recommendation if they'd moved by 20%?

5. What extreme (but plausible) stress events could affect reliability?

6. Could we defend our chosen portfolio to a regulator who asks why we didn't test a specific alternative future?

7. Is our current level of confidence really about how many scenarios we run, or about how much we trust the ones we've already built?

12 Questions to Ask Before Choosing a Power System Modeling Platform

Make Decisions You Can Depend On

You can't predict the next 20 years. But you can make decisions that hold up across many possible futures. Find out which assumptions would change your recommendation if they moved, and build a plan that survives them moving. That's what makes a portfolio defensible.

Explore the EnCompass interactive demo.