P10, P50, P90 and the quiet violence of point estimates

Here is a pattern we see in every second IC pack on a merchant battery. The revenue line reads: 'Year-one revenue: A$72M. Source: Aurora / Cornwall / [vendor].' One number, two decimal places, confidence implied.

It is the wrong shape of answer. Even a well-run merchant dispatch model on a reasonable NEM node will produce a revenue range across its input scenarios where the P90 is 40 to 55 percent of the P10. Ship a single number and you have deleted that cone from the conversation.

What distribution-first underwriting actually looks like

The shape we ship in an IC memo is not a dramatic reinvention. It is the same shape a mature project finance team uses for solar, applied to the merchant case:

1. 1.A base case. P50 revenue, built from the consensus price curve and the project's dispatch optimisation under realistic SOC and SOH constraints.
2. 2.A downside. P90 revenue, built from the stress-scenario curve (cannibalisation pulled forward, volatility compressed, FCAS cap tightened) under the same dispatch.
3. 3.An upside. P10 revenue, built from the high-volatility scenario (higher price formation frequency, fatter tails, inertia-adjacent ancillary products valued higher) under the same dispatch.
4. 4.A sensitivity table. How each dimension (curve, cannibalisation, FCAS mix, SOH degradation, cycling cost) moves the cone.
5. 5.A cone diagram. Revenue by year, P10 to P90, for the first 15 years.

Worked example: a 200 MW / 400 MWh NEM site

Take a real-shape example. 200 MW / 400 MWh, single-cycle merchant BESS in NSW, COD 2027, assumed SOH degradation at 2.5% per year, no capacity contract. Point estimate from a consensus curve: year-one revenue around A$65M. Plausible, repeatable, underwritten with a straight face across dozens of these assets.

Run the same asset under distribution-first underwriting and the picture changes:

• P50 around A$68M. Very close to the point estimate. That is the point: the P50 is not where the insight is.
• P90 around A$38M. That is where merchant exposure bites. If cannibalisation accelerates on the NSW node and volatility compresses, the project's IRR drops below hurdle.
• P10 around A$92M. Upside is real but is not a base case.
• The widest sensitivity is not the curve. It is cannibalisation assumptions across the NSW BESS cohort. A seven percent shift in assumed fleet entry drops the year-five P50 by 18 percent.

None of this is knowable from the point estimate. All of it is decision-relevant at the IC.

The case against point estimates

There is an honest argument for point estimates. They are simpler to communicate. They compress well. They make cross-project comparison easier in a crowded pipeline meeting. All fair.

None of that survives first contact with an IC that has seen one merchant BESS underperform its base case by 40 percent. After that meeting, the point estimate is a liability. The IC will want to know what you did not tell them, and the answer is always: the shape of the cone.

Batteries are revenue products. Underwrite them like revenue products.