Project finance model review checklist: when the returns look too good, check the model first

A project finance model review checklist works in five layers, run before a lender sees the model. Check that it holds together, then the finance logic, the formulas and the scenarios, and last whether the answers make sense. When the IRR looks too high, trace it back to the assumptions that drive it.

The founders were planning a fuels project in the Gulf, with two plants sharing one site. Their model ran for two decades and had grown through many versions. Before it went to a lender, they asked us to check it from top to bottom.

The biggest number in the model was the problem. It said the project would earn an IRR above 45%. Projects like this usually earn much less. A lender who sees a number like that doesn't get excited. They assume something is broken and start digging.

How we reviewed the project finance model

We ran the review in five layers. Here's the checklist we used, in order.

1. The model holds together: the balance sheet balances and the cash flow ties to cash on the balance sheet
2. The project finance logic works: loan repayment, reserve accounts and covenants
3. The formulas are clean, with no numbers typed straight into them
4. Every scenario runs, and none of them breaks the loan's rules
5. The answers make sense, and any outlier traces back to its inputs

A lot of it was good. The balance sheet balanced every year to the cent. The cash flow tied to the cash on the balance sheet. The model had its own 53 checks, and 52 of them passed. The loan was paid off on time and the reserve account worked the way it should.

What we found

We still found 20 issues, and five of them were serious. Here are the ones that mattered most.

The interest paid while the plants were being built showed up in three places, with three different values. The model used a shortcut that came out higher than the month-by-month figure it already worked out on another tab. That made the plant's value on the balance sheet too big.

An old copy of some revenue and profit rows had been left behind after a reshuffle. Halfway through the forecast, those rows dropped to zero. Anything still reading them would quietly pick up a zero.

We counted 875 formulas with numbers typed straight into them, mostly the inflation rate. If someone changed inflation on the inputs page, those cells wouldn't move. Two summary tables also gave very different answers for the same scenario. One of them had to be wrong.

The model tested 20 scenarios, and in four of them the project couldn't cover its loan payments with enough room to spare. That breaks the loan's rules. Then there were the returns. We traced the high IRR back to two assumptions: the selling price and how hard the plants run in their first years.

What we handed back

• A written review of every issue, with the cell-level fix for each one
• A tracker so the team could work through the list and mark each item done
• Two loan structures that fix the covenant breaks without changing the mix of debt and equity, with our pick
• A plan for the returns: explain them on one page before a lender asks, or trim the price and slow the early ramp

Later the project needed one lender for each plant. We rebuilt it as two standalone models and a combined view that ties back to the original. While doing that, we found 421 broken references the model's own checks had missed.

Why it matters for your raise

A big return feels like good news. In front of a lender it's a question, and you want the answer ready before they ask it. The founders here had a careful model. It had still picked up the kind of mess that builds up over many versions, and the headline number was where it showed.

You can run this checklist yourself. If you'd rather hand it to someone else, see how our financial model review works.