The model was the easy part

The first thing I built in this role was a cash-flow model with eleven tabs and a scenario switch I was genuinely, visibly proud of. The second thing I built was a single slide. The slide was harder, and it was the only one of the two that changed anything.

Eight months as an investor relations analyst across two early-stage portfolio companies, almost all of it remote, a fair amount of it between classes. Here is what I actually learned, which is mostly not modelling.

What a runway model is, honestly

Strip the tabs away and it answers one question: how many months until the money runs out, and what would have to be true for it to be more. That's it. Everything else is scaffolding for that sentence, and scaffolding has a way of becoming the project if you let it.

My eleven tabs were scaffolding. The version that survived to April has four.

Where the difficulty actually lives

Not in the maths. In the inputs. Nobody hands an early-stage company's numbers to you clean — you get an export, and a founder's categories, and the discovery that “contractors” meant two different things in Q1 and Q3 because the person doing the bookkeeping changed.

I spent more time establishing what a number meant than doing anything with it. I've come to think that isn't the boring part of the job. It might be the job.

A model is a machine for turning assumptions into consequences. Its accuracy is entirely borrowed from theirs.

What investor relations means at this size

At a large company, investor relations is a function with a team, a calendar and a script. At a company of fifteen people it is a person — usually the founder, occasionally the intern — answering one investor's specific question before Thursday.

That reframed the whole job for me. I wasn't producing reports; I was producing answers to questions somebody had actually asked. Much narrower target, much easier to hit. The eleven-tab model, I eventually realised, was me answering questions nobody had asked yet — which is a very comfortable way to feel busy.

The corollary is uncomfortable and I think correct: if you can't name the person who will read a piece of analysis and the decision it changes, you are doing it for yourself.

The sensitivity table that changed the conversation

For weeks I presented a runway figure and watched people argue about the figure. Then I put a scenario table into the deck — three columns, one variable moving in each — and the argument immediately became about which assumption we disagreed on.

That's a far better argument, and it's a much better meeting. A point estimate invites people to litigate your credibility. A range invites them to tell you where they'd put the input, which is the thing you wanted from them all along.

Macro shows up in a seed-stage spreadsheet

I did not expect to be writing about AI valuation froth and semiconductor lead times in materials for companies this small. But a hardware-adjacent business's inventory assumption is a macro call, whether or not anyone labels it one, and an early-stage company's fundraising timeline is a bet on a market that may not be in the same mood in nine months.

Naming those explicitly in the deck was uncomfortable — it looks like adding risk to your own story. It did the opposite. Investors who've priced those risks themselves relax visibly when they see you have too.

What I'd tell September-me

1. Build the smallest model that answers the question. Add a tab only when someone asks something it can't answer.

2. Put the assumptions on the slide, not in a footnote. If they're load-bearing, they're content.

3. A range beats a point, every time, in both directions.

4. Remote work is a writing job. Nobody sees you think — they only ever see the sentence you sent at 11pm.

Two companies, eight months, one deeply over-engineered first model that I keep a copy of as a reminder. The finance was the part I'd trained for and the part that took the least time. Translating it into something a person could disagree with productively took everything else.

The model was the easy part. I'm starting to suspect the model is always the easy part.