From ranges, to knowing what's actually driving them.
Analysis is still an iteration process. Same as it always was.
What changes is how much of the space you can see before you decide what to test next.
Still building the concepts, refining your communication or considering sequencing your decisions over time instead?
Progressive search, at a different scale
Refining a decision to find where it actually breaks always meant iterating. Test a budget at 100 and 150, see a meaningful gap, then test at 120 and 135: finding exactly where the slope changes over that short delta.
That's true whether you're evaluating each case by hand, or you already have the whole space pre-calculated in front of you. Only the scale changes: how much context you have available before you decide what to test next, and how many cases you can test at once.
Before: one, then a handful
Test a variable. Watch what happens. Adjust, test again. Step by step.
Once there's some automation behind it, that becomes a small batch: run a handful of combinations together, review, learn, and guess the next batch. Same process, you're learning just slightly more on each step as you try to inch towards an optimum without an overall perspective.
Now: hundreds or thousands, and a shape to start from
When you're not limited on batch size, you get to map the whole space, calculated upfront. Hundreds or thousands of combinations at once.
The iteration changes shape too. You can start with the broad picture, then refine the specific elements that matter. Or go looking for a mechanism you hadn't previously considered, and understand how that new element changes the entire shape of the map in one go.
It is still iteration. Just starting from a shape instead of a blank page.
When the broad shape of possible outcomes are already defined, finding the optimum is just a matter of zooming in on the right area and adding the right definition, rather than hoping you'll stumble into it via iteration.
The map is weighted
Test decision A against decision B at a base case, add a high and low price, and you've got a handful of combinations. Sample every uncertain outcome against every other instead: production, price, cost, whatever's genuinely unknown. That's where the hundreds or thousands actually come from.
And every point in that space carries a likelihood, alongside its value.
Weight each point by how likely it is, and the average across the whole map is your expected value. In constrained systems, it's most likely lower than your base case, or your best guess at what will happen, because downside is typically uncapped while upside runs into real limits once things go well. That delta is worth watching closely: it keeps expectations aligned, and it's often exactly where a debottlenecking opportunity is hiding (and how it can be valued).
The map may hold two other key areas worth naming. One is the exceptional case: the region where things go well and you achieve success. The other is the tail: the region where enough goes wrong at once to lose money. Chasing down what drives each one, and how to accentuate or mitigate them, is where the real insight sits.
Both sit on the same map as everything else here. Finding them is the same exploration as the rest of this page, aimed at two specific corners of it.
Every outcome holds more than one number
A scenario carries more than an NPV. Run it through and you get a cashflow profile over time, the capital it actually requires, how long payback takes, and the risked range around all of it.
That's a genuinely multi-variable outcome. Which of those numbers matters, and how much, depends on what you're actually optimizing for.
What you care about shapes the map
Those multi-variate points hold still. Same scenarios, same combinations, same options on the table.
Which of these outcomes actually matters, and how much, depends on the lens you're using. Growth rates achievable under different capital investment amounts is one lens. A constraint on emissions is another.
Same map. Different contour lines drawn over it, depending on what you're optimizing for, and which of a scenario's several outcomes that lens actually cares about.
That's why the best-looking answer can flip without a single input moving. The valuation underneath it shifted.
A case that looked immaterial under one aim can turn out to be the whole decision under another.
Shape the map as you test it
That's what testing is for. You're checking whether a case is material to what you currently care about.
Pull a case out. See whether it was doing any of the work under this lens. Add a new one when a stakeholder raises a scenario you hadn't modelled: a facility outage, a regulatory delay, a price case nobody wanted to say out loud. Reweight a probability, and watch whether the shape of the map changes, or just the shading on top of it.
It's the same map, redrawn live, in the room. No rebuild required. The question that prompted it, and what you're optimizing for, are both still on the table.
What actually moves the number
Whether it's a decision worth getting right or a risk that matters, picking it out from the rest is what focuses the conversation.
Testing each lever means watching for where the reaction stops being proportional. Where a small change suddenly matters a lot more, or a lot less, than it did in the previous step. Finding that boundary deliberately (and defining it within the uncertainties that matter), instead of stumbling into it in front of a stakeholder, is the point.
The breakeven is [$X/bbl] on an Expected Value basis.
Where that lever is an uncertainty that matters, cataloging its impact helps ground the stakeholder's expectations in reality: transforming the unspoken into a shared, quantified outcome, whether that's a dollar impact or the decision itself changing.
Increase annual FCF by [$Xmm] for every $10/bbl change in oil price.
Refining a number is only worth it if a tighter estimate could actually change what gets selected. One of the most time-expensive parts of an analysis is chasing data you don't have.
Where you don't have it at your fingertips, put in a few rough estimates first and test whether alternative cases move the answer. If they don't, don't worry about over-specifying it.
Oil price ($/bbl)
Test whether refining the input would change the decision.
In all cases, focus on what is critical over what might be interesting, where critical means it would change what they decide. A number that never changes the answer doesn't need extra source modelling effort or better source data behind it.
Finding the upside is the same search
The exceptional case comes from the same map, read for the region where things go better than plan instead of worse.
What it's worth is a probability and a value, same as any other point: the chance of landing there, and what that outcome is worth if you do. Multiply them, and that's what the upside actually contributes to expected value.
There's a 15% chance of hitting the stretch case, worth $40M if it happens: a $6M contribution to expected value.
Failures are usually combinations
Most of the time it takes more than one thing going wrong at once. Production underperforming and price falling together, say: the two variables that reliably drive the biggest swings.
But not always. A single variable can carry enough weight on its own. $20/bbl oil pushes plenty of companies below breakeven regardless of how new wells perform. $40/bbl still catches a few.
What usually decides which of those you're looking at is whether a mitigation was ever put in the way. Hedge production volumes while oil is still at $60/bbl, and you've given up some upside to remove the scenario where price alone sinks the case. Now it takes a combination, the hedged volumes not being enough and something else going wrong too, to get there.
Skip the hedge, whether by choice or by not having looked closely enough at the exposure, and one variable is all it takes.
That's what the map is actually for: finding the region where combinations of decisions and uncertainties collide, absent mitigation, leading to that bad outcome.
Once you've found it, the work is quantifying what a mitigation actually does, and what it costs. That turns risk appetite into a number you can weigh, instead of a feeling.
[Mitigation] removes [X%] of the loss scenarios in this region of the map, at a cost of [Y] to expected value.
Example: "Hedging 60% of production removes roughly two-thirds of the loss scenarios below $50/bbl, at a cost of $14M to expected value."
The space is understood now. What you do with that is next.
You've found the shape, you now know which decisions are worth getting right, and which risks actually matter. You know the key boundaries, how each step impacts the overall outcome, and what could sink it. All of it priced.
Ready to land this with stakeholders?
Knowing what matters and getting a room to agree on it are two different problems.
Working through a staged decision?
The same exploration applies to a phased commitment: which of your own numbers would actually flip the path.