How Long Should a Proposal Be? (And How to Find Out for Your Own Work)

Every answer you will find to this is a number — three pages, ten pages, one screen — and every one of them is an average taken across industries whose buying processes have nothing in common. A software retainer, a building survey and a public-sector tender are not the same document, and no median over the three of them describes any of them.

So this piece does two things instead. It says which question length is actually a proxy for, and then it gives you a way to measure the right answer for your own work, from your own sends, in about ten of them.

Length is not the variable. Position is.

A twelve-page proposal whose decision sits on the first screen outperforms a four-page one where the decision is on the last page, and it is not close. The reader who has what they need in the first thirty seconds will read the remaining eleven pages if they want them and will still be able to act if they do not. The reader who has to hunt is being asked to invest before they know whether it is worth investing.

This is why “how long should it be” is nearly always the wrong first question and “what is on the first screen” is the right one — which is a decision in its own right, and the reason the ordering piece in this series comes before this one in practice even though it comes after it in the index.

Once the decision is at the top, length stops being a risk and becomes an inventory question: what else does this document have to contain, for whom, and can any of it be somewhere other than in the middle of the argument?

The two documents hiding inside your proposal

Almost every long proposal is two documents that have been stapled together and are now fighting each other.

• The argument. What you will do, what it costs, when it is finished, why you. Read by the person who decides, usually once, usually quickly, often on a phone, often forwarded from someone else.

• The evidence. Methodology, technical detail, case studies, CVs, assumptions, terms, the twelve-row table of deliverables. Read by the person who has to check the work, sometimes carefully, sometimes never.

Those two audiences want opposite things from length, which is why one document cannot satisfy both by being some compromise size. The fix is not to cut the evidence. It is to stop putting it between the reader and the decision — either below the argument in the same document, or in a second document you link to from it.

A collection is the tidy version of that: one proposal with an appendix, a rate card and a case study behind it, sent as one link rather than four attachments. The argument stays short without any of the evidence being lost.

How to actually find your number

Here is the measurement. It takes about ten sends, and it produces a curve that is true for your work rather than for a survey.

1. Record two things per send. The length of the document — word count, not pages, because page count changes with font and margin — and the duration of the first session that document gets from a real reader.

2. Only compare like with like. Same kind of engagement, same rough deal size, same kind of buyer. A proposal for a £2,000 piece of work and one for a £120,000 programme are two different populations and averaging them tells you nothing.

3. Plot length against first-session reading time. You are not looking for a correlation — there will be one, trivially, because a longer document takes longer to get through. You are looking for the point where the line goes flat.

4. The flat point is your ceiling. Past it, extra length is buying you no extra attention. Everything after that point in your document is being carried by readers who were going to read all of it anyway, and it is costing you nothing except with the readers who never got there.

This only works if the reading time is honest. A duration computed from the gap between opening a tab and closing it is close to meaningless — it counts lunch, it counts a background tab, it counts a browser left open overnight, and it only ever exaggerates. How to measure reading time honestly sets out why. In Quixli, reading time accrues only while the document is actually on screen, which is what makes the flat point mean anything.

Two things that will fool you

The measurement above is simple enough to trust and simple enough to misread, so here are its two failure modes before you meet them.

• The first visit is not always a person. A link pasted into a chat gets fetched by the software to build a preview card. If that lands in your data as a four-second session it will drag your averages down and it has nothing to do with your document’s length. Quixli filters automated fetches out of every count; if you are measuring somewhere else, check that whatever you are measuring in does too.

• A very short document produces a short session for reasons that are not about attention. There is a floor: a one-page quote cannot generate a six-minute read no matter how compelling it is. Below about a screen and a half, the measurement stops discriminating and you should stop reading it as a signal.

What to do while you are collecting ten sends

A working default, offered as a default and not as a finding: make the argument short enough that a reader could act on it without scrolling twice, and let the evidence be as long as it honestly needs to be, underneath.

That is not a page count, and deliberately so. It is a rule that survives being applied to a two-page quote and to a forty-page tender response, which no page count does.

And when you cut, cut the parts that are about you. The paragraph restating the client’s own brief back to them, the history of your company, the team photographs, the methodology diagram — these are the sections that feel obligatory and are read least. The reader already knows their brief. They are trying to find out what you will do about it.

If the answer to “how long” turns out to be “shorter, with the decision at the top”, how to structure a proposal so the part that matters is read first is the mechanics of doing it. If it turns out that nobody is opening it at all, that is a different diagnosis entirely. And what Quixli measures, limits included is what the method above is written against.