How to read a Venture Capital Term Sheet at a Structural Level

Read the term sheet as two lists, not one. Some clauses decide who gets what money and when. Others decide who can make or stop a decision. A reader who sorts every clause into one of those two piles has understood the document structurally, and a reader who reads it top to bottom has read a list of unrelated paragraphs.

A term sheet is read in a fixed order rather than from top to bottom, and each answer changes what every earlier answer meant. How any single clause works underneath is covered separately, as are the drafting, the amending and the arguing of a word of it. The whole skill here is sorting, and the document as handed over is not sorted.

What is a term sheet, and what is it not?

A familiar version of this happens at home. A household agrees to buy a flat. Long before anybody signs a sale deed, there is a single sheet of paper going back and forth: the price, what is included, when the keys change hands, how much is paid up front, and one line saying the seller will not show the flat to anybody else for a while. Nobody thinks that sheet is the sale. Almost none of it could be enforced if either side walked away. And yet every household in that position reads it far more carefully than the sale deed, because the sale deed is only going to write down whatever that sheet already decided.

A term sheetA short summary of a proposed investment, mostly not binding. is that sheet of paper for an investment into a company. It is short. A term sheet is written in deliberately plain words. The people signing it are meant to read it themselves rather than leave it to their advisers. And it is mostly not bindingActually enforceable, which most of a term sheet is not.. Unenforceability sounds like a reason to take the document lightly, and is in fact what makes a structural reading possible. The exercise is not parsing enforceable language. The reader is looking at a list of decisions somebody has proposed, and a list of decisions can be sorted.

Naming the things a term sheet is not helps just as much. A term sheet is not the agreement between the shareholders, a long document that comes later. Nor is it the subscription agreement under which money actually moves. The constitutional documents of the company are a third thing again, and several of these decisions eventually have to be written down there before they mean anything. And a term sheet is not a valuation produced by anybody independent, nor is it the money: a term sheet is a proposal, and a proposal that is signed is still a proposal that has to survive whatever the parties agreed to check afterwards. Every one of those later documents is written from this one. The reading matters for that reason alone, even though the paper barely binds anybody.

The structure of a term sheet as an object to be read, rather than as law to be drafted, is treated at length by Feld and Mendelson in Venture Deals, 2011. The order below is that framing turned into something that can be worked through.

Which two piles does every clause fall into?

Think about two people running a shop together. There are exactly two kinds of question they have to settle, and they feel completely different from the inside. The first kind is about money: how the day's takings are split, who is repaid first if they ever sell the shop, what happens to the split if one of them puts in more stock next year. The second kind is about decisions: who can sign a cheque, who can hire a cousin, whether either of them can bring in a third partner without asking the other. When the two of them write a partnership deed, those questions do not come out in two neat sections. The questions come out in whatever order the conversation happened.

A term sheet is exactly that document at a larger scale. An economic termA clause deciding who gets what money and when. decides who gets what money and when. A control termA clause deciding who can make or stop a decision. decides who can make or stop a decision. Every clause in the document goes into one of those two piles, some go into both, and the document itself will not do the sorting.

Why bother sorting at all, when the paragraphs could simply be read in order? Because the two piles fail differently, and on different days. A money clause matters on one day only: the day there is money to divide. A money clause can sit quietly for years and then decide, in a single afternoon, who walks away with what. A decision clause matters on every day between now and then. A decision clause governs the ordinary week: the hire that does not happen, the loan that is not taken, the budget that has to be approved by somebody who is not in the building. Sorting the clauses is how to work out which parts of this document are felt next month and which parts are felt once, at the end.

There is a second reason, and it is the one this reading order ends on. Once the clauses are in two piles, the piles argue with each other. A clause in the money pile can quietly change what a clause in the decision pile is worth, and the reverse happens too. None of that is visible while the clauses are still sitting in the order somebody typed them.

What is the reading order, pass by pass?

The order is five passes over the same short document rather than one careful walk from top to bottom. Each pass asks one question of every clause, and each pass can change what the previous pass appeared to say. Reading in passes is the only way round that: a term sheet is a small system, and a system cannot be read in one direction.

Notice how the order is built. The sequence is not the order the paragraphs appear in, it is not alphabetical, and it is not most important to least important. Importance depends on which future arrives. The order runs by what each clause controls: first the thing being bought, then the people already there, then the decisions, then the bad days, and last the clauses that are not where they look.

Why is the valuation not the first thing read?

Because it is not a fact that can be acted on. A valuation is one input into a division, and the other inputs are set by clauses somewhere else in the same document. Ask anybody outside this subject what a term sheet said and a single number comes back: the company was valued at so much. The valuation number is real, it was genuinely agreed, and on its own it says nothing about what anybody paid for anything.

Here is the arithmetic that matters, and it is arithmetic anybody can do. The pre-moneyThe agreed value of the company before the new money goes in. figure is the agreed value of the company before the new money arrives. The post-moneyThe pre-money plus the money raised. figure is that same value plus the money raised. The price per shareThe pre-money divided by the shares it is spread across. is the pre-money figure divided by the number of shares it is being spread across. Three sentences, and the third one is the only one anybody actually pays. A valuation with no share count attached is a headline, not a price, and a reader who stops at the headline has not finished the first pass.

The headline sets the same trap as a shop advertising a sack of rice for a fixed sum without saying how many kilograms are in the sack. The number on the board is true. The number is also useless until the weight is known, and the weight is written somewhere else, in smaller print, by somebody who knows the board gets read first.

What do passes one and two actually ask of the money pile?

The money pile is read in one direction: from what is being bought, out to what happens to the money on the day it is divided. Two of the questions cannot be answered until an earlier one has been, so they go in this order and the answer to each is written down before the next.

Notice how much of that list is not about today. Questions six, seven and eight are about days that may never come. Each is still a money question: it decides who gets what on the day that day does come. The money pile is not the pile of clauses about the price; it is the pile of clauses about who receives rupees, whenever those rupees appear. The mechanics underneath each of those clauses, how the arithmetic runs when a company is actually sold or a later round is actually priced lower, are worked separately in this subject area.

What do passes three and four ask of the decision pile?

The decision pile is read differently, and the difference is worth stating before the questions. The money pile works out amounts. The decision pile works out two separate things that readers routinely collapse into one: what somebody can stop, and what somebody can make happen. Stopping and making are not the same power, and they almost never sit in the same clause.

Think about a joint bank account in a household where both signatures are needed on anything above a certain sum. Either person can stop a purchase. Neither person can force one. Stop but not start is the whole shape of a consent list, and a reader who has felt it once at home never misreads it in a document again. A seat at a table is a different thing entirely: it is one voice among several, it can be outvoted on an ordinary matter, and it is also present for every conversation that never reaches a vote.

The invented private fund this subject area works from shows the two side by side. In Nilgiri Growth Partners Fund II, invented, as at the end of its Year 9 Quarter 2, seven reserved matters sit in the shareholders agreement at every one of its nine holdings. A majority of the five-seat board is held at only four of them. Same fund, same day, two entirely different kinds of power, and the arithmetic of how each one is used is worked separately in this subject area rather than here.

With that distinction in hand, the decision pile reads in this order. As before, each question has an answer that changes what the next one means.

How can a clause change the price without changing the price?

Before the arithmetic, sit with the domestic version of it. Six people agree that a cake on the table is worth Rs 600, and a seventh person is going to pay Rs 150 to join, which everybody understands as buying a fifth of what will then be on the table. Then somebody adds a line: before the new person joins, three extra slices should be cut and set aside for the people who work in the kitchen, out of the same cake. Nobody has changed the Rs 600. Nobody has changed the Rs 150. But the cake is now being divided into more slices before the money arrives, so every slice is worth less, and the six people who were already there are the only ones paying for the kitchen's three.

Now the worked instance, and every figure in it belongs to Aravalli Learning Systems Private Limited, invented, a company held by Nilgiri Venture Fund I, invented, at its Series A on Aravalli's own clock in Aravalli Year 3. Here is the headline as this record fixes it. The pre-money is Rs 1,20,00,00,000, the raise is Rs 30,00,00,000, and the post-money is therefore Rs 1,50,00,00,000. The pre-money is spread across the 1,00,00,000 shares in issue, so the price is Rs 120.00 a share, the Rs 30,00,00,000 buys 25,00,000 new shares, and the register afterwards totals 1,25,00,000 shares. The founders are at 56.0 per cent, the pool of employee shares at 8.0, the holders from the seed round at 16.0 between them, and the new investor at 20.0. The four holdings add to exactly 100.0.

Now add one clause, and it is a counterfactual throughout: suppose the option poolShares reserved for employees, which dilute whoever it comes out of. were to be enlarged by 5,00,000 shares before the round, out of the same agreed pre-money. The enlargement did not happen at Aravalli and is set out only to be read against what did. The counterfactual works through the four inputs given earlier. Input one is unchanged: the quoted number is still a value before the money. Input two is unchanged: the raise is still Rs 30,00,00,000. Input three has moved: the count the pre-money is spread across is now 1,05,00,000 shares rather than 1,00,00,000. Input four is where it lands: the 5,00,000 extra shares came out of the count that existed before the money, a count made up entirely of the people already there.

So the price per existing share is Rs 1,20,00,00,000 divided by 1,05,00,000, or Rs 114.29, down from Rs 120.00. At that price the Rs 30,00,00,000 buys 26,25,000 shares rather than 25,00,000, and the register afterwards totals 1,31,25,000 rather than 1,25,00,000. The pre-money did not move by one rupee, and the price fell by Rs 5.71 a share. Anybody who read only the headline saw nothing happen at all.

So who paid for that? Not the incoming investor. Its Rs 30,00,00,000 bought 26,25,000 shares out of 1,31,25,000, or exactly 20.0 per cent, the same 20.0 per cent it had in the round as this record fixes it. The pool went from 8.0 per cent to 11.4. And the two groups who were already in the register moved: the founders from 56.0 per cent to 53.3, and the seed holders from 16.0 to 15.2. The whole cost of that clause fell on the people who were already there, and the person the clause was negotiated with was untouched by it.

The arithmetic below needs one honest note first. The rounded column will look wrong otherwise. Rounded to one decimal, the four figures after the counterfactual read 53.3, 11.4, 15.2 and 20.0, adding to 99.9. The unrounded shares add to exactly 100.0, and nothing has been nudged to make the printed column balance.

There is one identity worth carrying away from this, and the mechanism behind it is worked in full separately. Issuing new shares moves every holder already in the register by the same multiplier: nobody is singled out, and nobody is spared. The pool clause changes that multiplier, and it changes the multiplier before the new money is counted. The change therefore lands on the existing holders alone. The full register of this company across its three rounds, and the arithmetic of what each round did to each holder, are covered separately.

Why is reading the valuation first the fault that costs the most?

What is read last?

Two things, in this order, and neither of them is a new clause. The first is the re-sort. The list built on the first four passes goes through once more, with a single question asked of every line: what does this clause actually do? Not where did it sit, not what is it called, not which pile did it look like it belonged to. What does it do?

The re-sort is where the reading pays for itself: the piles come out different from how they went in. Several clauses a reader files under price turn out to decide who can make or stop something, and at least one clause that looks like administrative housekeeping turns out to move more money than any paragraph in the document. The pool clause is the clearest instance of the second kind, and here are the others alongside it.

The second thing read last is the short paragraph in which the sheet states which of its own provisions are intended to bind. The binding paragraph is read differently from everything else on the sheet: not structurally, but word by word. For those provisions the words are the whole of the answer rather than a summary of a longer document to come. Everything else on the sheet is a proposal being described; this part is a decision already being taken. Which provisions a particular sheet puts in that category is fixed by that sheet, and no single version of it is the version.

Four words a reader will reach for have no place in a structural reading. Whether a clause is fair, unfair, ordinary, or the kind of thing one side or the other would want is not a structural question, and a reading order cannot answer it. A reading order gives what each clause decides and which other clause changes that answer.

How does somebody actually use this order in a working week?

Nobody reads a term sheet the way it has just been read here, slowly, from a standing start, with the arithmetic written out. In a working week people run the five passes quickly, in that order, and write down five answers. The value of having the order is that the answers come out in a shape somebody else can use.

Take somebody at an investment manager. Nilgiri Alternatives Advisors Private Limited, invented, runs six vehicles, one of which is Nilgiri Venture Fund I, invented, the fund carrying the position in Aravalli Learning Systems Private Limited worked through above. An analyst there who is handed a term sheet is not reading it to decide whether it is fair. The analyst is reading it to answer the three questions their own committee will ask, and the passes map onto those questions one for one. Pass one gives the price per share, the only figure that can be compared with anything at all. Passes two and five together answer who paid for this round, and that answer decides what the people running the company are looking at every morning. Passes three and four answer what happens next without the fund, and what cannot happen without the fund. The file note is those two sentences.

Take somebody inside the company instead, reading a sheet that has just been handed across a table. Their temptation is different and stronger: pass one produces a number they will repeat to their parents, their staff and their first hire, and it is very hard to keep reading past a number like that. The reading order is a discipline against exactly that pull. The order says the headline is input three of four, and that the clause which decides what the headline means is somewhere further down, written in language that sounds like an administrative note about employee shares.

And take somebody whose job is simply to write the summary that other people will read. The two piles are the summary: one sheet on what this decides about money and one sheet on what it decides about decisions beats any amount of prose that follows the document's own order. A reader who receives that summary can ask a sensible question. A reader who receives a paragraph-by-paragraph precis has been handed the same unsorted problem in a longer form.

What remains undone, even after all five passes?

Quite a lot, and being clear about it is part of the reading. Nothing has been negotiated. The order gives what is being proposed and which clause changes which other clause, and asking for something different is a separate skill, covered separately.

Nothing has been drafted either. Every decision on this sheet has to be written properly into longer documents afterwards, and the difference between a decision as summarised here and the same decision as it finally binds is where most of the work in a transaction actually sits. Nor does any of this establish whether the company is worth what the sheet says. A term sheet is not evidence about a company; it is evidence about what two parties were prepared to agree on a particular day. And none of it is certain to happen. Most of what has just been read is not binding, and that is exactly where the reading began.

What does remain, and it is worth having, is a document that can be held in two hands: one pile that decides rupees and one pile that decides who can say no. Two piles are the whole of what a structural reading is for.