How to calculate burn rate without lying to yourself

I told a board we had eighteen months of runway. The real figure was fourteen and a half, and I was wrong for most of a year before anybody caught it.

Nobody had lied, and no line in the file was wrong. The spreadsheet was correct to the cent. It was answering a question about a company that had stopped growing, and we had not stopped growing, and that one unstated assumption was worth four months.

I should be precise about my own part in it, because the way I have told this story to people since has a tendency to drift towards the spreadsheet being at fault rather than me. I built the model, so the fault is mine and not the tool's. I chose the flat formula because it was the one I had seen used everywhere, and I never went back to ask whether the assumption underneath it still held, and there were at least four quarterly reviews in which I presented that number to people who trusted it and in which asking the question would have taken an afternoon and would have changed what we did next. Nobody stopped me because nobody had reason to. That is not the spreadsheet's failure.

Four months is a fundraise. That is the part I still find hard to sit with, because the decision it changed was when to start talking to investors, and we started late on the strength of a number I had built.

What the formula gets right

Most of it, which is why the mistake survives so well.

Net burn is money out minus money in over a month, and the two words in front of burn matter more than people allow. Gross burn is everything leaving the account. Net subtracts what came in. A company spending 180,000 a month and collecting 60,000 has a gross burn of 180,000 and a net burn of 120,000, and the second number is the one eating the balance.

Runway is cash divided by net burn.

1.8 million against 120,000 gives fifteen months, and that is the number that goes in the deck, and for a company that is not growing it is entirely correct and there is nothing else to say about it. The trouble starts the moment the company is growing, which is most of them, and which is the condition the formula was never asked about.

Use a three month average rather than last month. One annual insurance premium, one shifted payroll run, and a single month swings the answer by two or three.

Strip the one offs out too. A 90,000 dollar recruitment fee is real money that will not repeat, and burying it in the average tells you the wrong thing about every month after it.

Where it breaks

Every guide on how to calculate burn rate gives you those two lines and stops. The stopping point is the problem.

Cash divided by burn assumes next month looks like this month. Fine for a company holding steady. Wrong for almost anyone who has just raised, because raising is generally followed by spending more on purpose.

Our own numbers, 78,000 to 120,000 over 12 months. Net burn in January was 78,000. In December it was 120,000. Nobody decided to grow burn by 54 per cent, it happened one hire and one tool and one office at a time, and compounded out it is 3.7 per cent a month.

Run both against 1.8 million. Flat, about eighteen and a half months. With the growth term in, fourteen and a half.

It gets worse the safer you feel

Same 1.8 million, burn growing four per cent a month. At 65,000 a month the flat formula says 27.7 months and the compounding one says 19.0, an overstatement of 45.7 per cent. At 95,000 it is 18.9 against 14.4, so 31.8. At 125,000, 14.4 against 11.6, which is 24.2. At 140,000, 12.9 against 10.6, or 21.7 per cent.

Read the 4 per cent column downward and the slope is the whole argument.

The company with two and a half apparent years is being told a story nearly half too optimistic. Compounding needs time to do damage, so the further out you look the more of it there is, and the further out you look is exactly when people relax.

The spreadsheet was part of it

A short aside about the file itself, which has nothing to do with finance and which I have now watched happen to three other companies.

Ours had cash as an input cell, burn as an input cell, and runway as a formula. That made runway look like a fact rather than a projection. Once a number looks like a fact people stop asking what sits behind it, including the person who built the model, who in this case was me and who had known perfectly well what assumption he was making on the afternoon he made it and had entirely forgotten by the following quarter.

So: put the growth rate in as a visible input cell even when it is zero. A zero somebody typed is an assumption. A zero nobody typed is an oversight, and afterwards the two look identical.

Right. Anyway, back to the arithmetic.

Measuring your own rate

You do not need a different model. You need one extra input, measured rather than assumed.

Take net burn from twelve months ago, call it b0, and last month's, b1. The monthly rate is the twelfth root of b1 over b0, minus one. Ours: 78,000 to 120,000 is a ratio of 1.538, twelfth root 1.0366, so 3.7 per cent.

I have shown that calculation to a number of founders and I do not have a tally, so I am not going to claim one, but the reaction is consistent enough that I stopped being surprised by it around the fifth time.

One of them put it better than I have managed to: "I know what we spend. I have never once thought about the shape of what we spend." The shape is the whole of it. At 3.7 per cent a month spending doubles in 19 months. At 4 per cent it doubles in 18. At 2 per cent it takes 35, which is why the difference between a guess of 2 and a real figure of 4 is not a rounding error but a year and a half of runway.

I had assumed, before I ran it on our own numbers, that a company notices burn growth while it is happening. It does not, and I did not, across 14 months of watching it happen. There is no month in which anybody decides to spend 4 per cent more. It arrives as 2 hires, 1 renewal at a higher tier, a tool somebody needed and a contractor who stayed 3 months longer than planned, and every one of those was approved by a different person on a different day for a perfectly good reason, which is exactly why no single meeting ever contains the decision to double the burn.

Strip one offs out of both endpoints. A single 40,000 dollar payment at either end distorts the ratio.

And if you have deliberately cut since, measure from the cut rather than through it, because 1 restructuring in the middle of a 12 month window makes the whole series meaningless. The formula cannot tell an intentional reduction from a trend. A 15 per cent cut in month 7 reads as negative growth for the whole 12 and hides everything that happened after it.

How I know any of this

Worth stating in 2 sentences. It decides how much weight the rest of this can carry.

This piece is arithmetic and not research, which is worth stating before the tables rather than in a footnote after them, because a table with 4 decimal places in it looks like evidence whether or not there is any evidence behind it, and I have been fooled by my own formatting before. I went looking for a version that did not need a growth term. I could not build one that survived 24 months of test data. I rebuilt the runway model 3 times before the compounding version stopped producing numbers I could not explain, which took about 6 hours. The model behind every table here is 1.8 million dollars of cash, a starting net burn of 120,000 dollars a month, and a growth term applied monthly, and those 3 inputs generate every runway figure in this piece.

That is a model rather than a study, and 3 inputs are not evidence about anybody else's company.

A model with 3 inputs cannot tell you what happens across a market, and I would not want anybody quoting these tables as though it could. I cannot tell you whether the pattern holds across 50 companies, because I have not found anybody who has published anything like it and I have looked twice. Where I say something is common I mean I have seen it several times and not that I counted it.

One finance lead put the committed number better than I have: "Everyone asks how much we burn. Nobody asks how much of it I could stop by Friday." That second question has a number attached, it takes about 40 minutes with a list of contracts, and almost nobody computes it.

The number nobody names

Gross and net are standard. The third figure I find more useful than either has no agreed name at all, which is odd for something that appears in every board pack I have seen. I call it committed burn: the part of next month's spending already contractually fixed. Payroll, rent, annual software you cannot exit, notice periods.

On those four companies it ran between 60 and 80 per cent of net burn. My suspicion is that the ratio rises with headcount and falls with how much you spend on contractors, but four data points cannot carry that and I am recording it in case somebody with a larger sample wants to check.

A company at 120,000 net burn with 82 per cent committed cannot cut to 40,000 next month however urgent the meeting was. It reaches about 98,000 immediately and the rest arrives over a notice period.

What the four months actually cost

I have been asked whether the gap mattered in the end, and the honest answer is that I do not know, because we raised and the company is fine and there is no counterfactual to check against.

What I can describe is the decision it changed. On eighteen months, the plan was to start conversations in the autumn. On fourteen and a half, with a process that runs four to six months from first meeting to money in the account, the plan should have been to start in August. We started the conversation in October, about 4 months later than we should have. The round closed in March with about seven weeks of cash left, which is not a disaster and is much closer to one than anybody in that building thought we were.

Nobody blamed me and I would have preferred somebody had. A number that is wrong in a direction that feels comfortable does not get argued with, and the absence of an argument is what let it stand for a year.

"The model was fine," the chief executive said when we went through it afterwards. "The timeline was the mistake." I have thought about that sentence a lot, because it is a reasonable reading and it puts the fault somewhere I would rather it were not. He may well be right about that. I think a model that produces a comfortable number and hides its own assumption is not fine, but I would say that, since the alternative is that I was simply slow to start a process.

One more from the same conversation, because it reframed the whole thing for me. "The formula is not the problem," she said. "The problem is that runway is the only number anybody looks at, and it is the one number that depends entirely on a guess about the future." I had been treating this as an arithmetic error for a year. She treats it as a reporting error, and I now think she is closer to right.

What is a good burn rate

I get asked constantly and I think the question is malformed, because burn on its own measures size rather than health.

The version I find useful is what a month of burn bought. For the 120,000 that left last month, what moved. If revenue rose 9,000, you spent 120,000 to buy 9,000 of recurring income, and whether that is good depends on how long it lasts. If nothing moved, you bought a month of optionality, which is sometimes the right purchase and should at least be a conscious one.

Stage benchmarks do exist and people quote them constantly. I do not trust most of them.

They are assembled from whoever happened to answer a survey, they lag by a year or more by the time anybody publishes them, and a company with a nine month enterprise sales cycle has no business being compared against one selling a 40 dollar subscription to consumers, which is exactly what happens when both get filed under the same funding stage.

Runway maths brushes up against revenue recognition, committed contracts, debt covenants and payment timing, and an article cannot see any of yours. If the numbers are close enough to matter, a finance person should be reading the ledger rather than a formula from a website.

The objection I have not beaten

Burn growth is a decision and can be stopped, so modelling it as compounding treats a choice as though it were weather.

That is true in principle. It is rarely true in a specific month. Cutting hard does work, and every company I have looked at grew burn every year before it did, none of them having planned to, and all four described it afterwards as a series of individually sensible decisions. It is a 6 week project with a morale cost nobody puts in the spreadsheet, and the spreadsheet is the only place the 6 weeks are free.

A second finance lead I put this to disagreed with the whole framing. "You are modelling a decision as if it were rainfall," he said. "If burn grows four per cent it is because somebody signed something." That is a fair objection and I have not fully answered it, which is why it appears here rather than in a footnote.

So I keep the growth term at 4 per cent and treat the flat number as the best case.

Someone I respect does the opposite, and argues the flat number is the plan while the growth number is a failure to execute it. We have had this argument three times and I am no longer sure which of us is describing the company and which is describing a wish. He also has better numbers than I do, over a longer run. He also has only ever run one company, which either makes his sample worse than mine or makes it the only one that matters, and I cannot decide which.