---
title: "Chapter 9: The Vector"
book: "Collapse or Collapse"
author: "Vinay Pasricha"
url: https://vinaypasricha.com/books/collapse-or-collapse/chapter-9-the-vector
status: "On Kindle: https://www.amazon.com/dp/B0HNL53J5R. The full text is shared on vinaypasricha.com with the author's permission."
---

# Chapter 9: The Vector

Every company in this book is built from one thing, and this chapter is about how to build it.

A vector is the smallest unit that produces a result. It is one result, aimed at one market, carried by one chain of agents, and held by one pair of humans by default. Nothing smaller produces anything a customer would notice; an agent on its own is a skill, and a skill delivers nothing until it sits in a chain. Nothing larger is necessary; an Outcome is many vectors running the same playbook, and a company is many Outcomes. If you can build one vector well, you can build the company, because the company is that vector, copied and varied and held.

So I will build one, from nothing, the way a reader would build the first vector of her own collapsed company. I will take it from GoodSpace, but from the side of the company the last three chapters did not visit. Chapter 6 showed a vector that delivers mandates: a job description goes in, a placed candidate and an invoice come out. This chapter builds the vector that wins them: a description of the kind of client GoodSpace wants goes in, and a signed mandate with a job description comes out. It is the other half of the business, and it is the half most people assume cannot be run this way, because it is sales, and sales is supposed to be about people.

It is about people. That is exactly why it can be run this way.

## Begin with the result, at the right point

The first decision in building a vector is the one most companies get wrong, and it is not a technical decision. It is where to draw the result.

The obvious result for a winning vector is a signed mandate. A client agrees terms, signs, and sends a job description. It is a clean event, easy to count, and every sales team in the world is measured on something like it. And it is the wrong place to draw the line, because a signed mandate is not what GoodSpace’s customer feels, and it is not what GoodSpace is paid for.

Here is what happens to signed mandates in the real world. About half of them never reach a placed candidate. Some clients were never serious; they were shopping, signing with three firms to see what the market looked like. In some, the candidate who is selected does not join. In others another consultant fills the role first, or the client finds someone internally, or the position is put on hold and never taken off it. None of this is anyone’s fault in particular. It is the nature of hiring. But it means that a winning vector measured on signed mandates can hit every target it is given while the company earns half of what the targets imply, and the delivery vectors downstream fill up with mandates that will never close, and spend their mornings on them.

So the result of a winning vector cannot be drawn at the signature. It has to be drawn where the customer’s outcome is, and in recruitment that is a person who joins and stays, and an invoice that is paid. The winning vector does not produce that on its own; it produces half of it, and a delivery vector produces the other half. And so the doctrine draws the line around both of them.

At GoodSpace the design is one winning vector matched to one delivery vector, run as a pair of vectors and measured on one number: mandates closed and invoiced. The design target for each matched pair is twenty-five a month. Work backwards from that and every other number falls out. If about half of signed mandates reach fruition, twenty-five closed needs fifty won. Fifty won a month is the winning vector’s target; twenty-five closed is the delivery vector’s; and both are judged on the twenty-five, because the twenty-five is the only number the customer and the company both feel. A winning vector that brings in fifty mandates of which only ten close has not done its job, however many it signed; it has brought in shoppers. A delivery vector that closes fifteen of fifty good mandates has not done its job either. Neither can blame the other, because they share the result.

This is the first rule of building a vector, and the rest of the chapter rests on it. Design backwards from the result, and draw the result where the customer feels it. When the result sits between two vectors, size them together and measure them together. The old company drew a line between sales and delivery and put a department on each side of it, and each department hit its own number while the customer fell into the gap. The matched pair has no gap to fall into.

## The chain

With the result drawn, the chain is the sequence of steps that produces it, and each step is an agent. Here is the chain of a winning vector, built step by step.

**Find.** The first agent finds the companies that might need what GoodSpace sells: firms that are hiring, growing, funded, opening offices, posting roles. It works from the vector’s description of its ideal client and returns a list of companies and the people in them who hire. Today, at GoodSpace, the leads still come to each vector from marketing, which gives every vector its own leads so that no two vectors chase the same client. The design is that each vector finds its own, because finding leads is not a result anyone outside the company receives; it is the first skill in winning a mandate, and a skill belongs in the chain. The rule that no two vectors chase the same company does not need a department. It is one line in the master’s playbook, and the agents obey it.

**Qualify.** The second agent reads each company and decides whether it is worth approaching now: how many roles it is hiring for, how hard they are, whether it has used consultants before, whether it pays its bills. Most leads stop here, and they should.

**Research.** The third agent builds a short dossier on each qualified company: what it does, who leads it, what it has said in public about growth, what roles it has posted and for how long. The dossier is for the agents that follow and, later, for the human who will speak to the client.

**Approach.** The fourth agent writes the first message, in the client’s language and about the client’s problem, not GoodSpace’s product, and sends it through whichever channel the client is most likely to read. The fifth handles what comes back: answers questions, sends proof, follows up when there is silence, and knows when to stop.

**Book.** The sixth agent turns a reply into a meeting, and puts it in the diary of the agent that will hold it.

Then comes the conversation, and it is held by an agent.

**The first conversation.** Every new client has a real conversation before he signs, by voice or by video, and the conversation is held by an agent built for nothing else. Conversational agents are now good enough to do this well, and in one respect better than most salespeople: this one has read the dossier, knows every mandate GoodSpace has filled in the client’s industry, never forgets to ask a question, and never talks more than it listens. It tells the client plainly that it is an agent, and offers him a human whenever he wants one; most do not ask. What matters is what the conversation is for. A client who is about to hand a firm the hiring of someone who matters to him is deciding whether to trust it, and the firm is deciding whether the client is serious. Half of all signed mandates fail, and many fail because the client was only shopping; so the conversation agent is built, above all, to find out. It asks how many other firms are on the role, who decides, what happens if the role is filled internally, how soon the person must start, and what the last hire in the role cost them in time. It records the conversation, scores the client’s seriousness, and lays the record and the score in front of the human before anything is signed. The findability gate in the delivery vector protects the company from promises it cannot keep. The first conversation protects it from clients who are not really asking.

**Terms.** After the conversation, the seventh agent drafts the mandate: the advance, the fee, the exclusivity period, the payment terms, all from the playbook’s standard, with any exception marked. Here a skill agent stands at the gate: an agent that checks the terms against the company’s rules and the law, and flags anything outside them. It is not a legal department. It is a legal skill, placed at the one step where it is needed, and the gate after it stays shut until the check passes or a human with authority overrides it.

**The signature.** Here the chain reaches its human gate. The aligner reads the conversation agent’s record and its seriousness score, listens to the part of the call she needs to hear, and reads the mandate as it is read back. If the client is serious and the terms are right, she signs. If she is not sure, she can call the client herself before she does; that is her judgement, not a step in the chain. It is the one irreversible gate in the vector, because from that moment the firm has made a promise, and it is the one place in the winning chain where a human’s intention is required.

**Hand over.** The last agent takes the signed mandate and the job description and hands them to the matched delivery vector, whose first act is the findability report. The winning vector’s chain ends where the delivery vector’s chain begins, and because both are measured on the same twenty-five, the handover is not a wall but a seam.

Around the chain stand the two agents every vector has: the quartermaster, who answers for the whole of it to the human who asks, and the agent maker, who builds, tests and upgrades the others. That is the vector: nine line agents, one skill agent at a gate, a quartermaster and an agent maker, one human gate, and one result shared with the vector downstream.

## Gates follow the stakes

The chain has gates between every pair of agents, and almost all of them are reversible. A list of qualified companies can be sent back. A dossier can be redone. A first message that was not sent can be rewritten. These gates default open: the aligner samples what passed through them, and if she has not looked within the time the playbook sets, usually a day, they open on their own, and the chain does not wait for her.

One gate in the winning chain is different, and the doctrine treats it differently. The signature defaults closed and waits for a human however long it takes, because what follows it cannot be undone. Everything about the vector’s speed comes from the reversible gates, and everything about its safety comes from the irreversible one. The designer’s job is to know which is which, and to have as few of the second kind as the stakes allow. The human who holds an irreversible gate must also have the judgement it asks for: enough knowledge of the domain to know what her signature commits the firm to. That is a threshold, not a skill test at the door. A new aligner reaches it in the vector, beside her pair, and until she has, her irreversible gates are co-signed. A year ago the first conversation with a client would have been a second human gate, because no agent could hold it to standard. Now one can, and so the gate has moved, as gates do: the line between what the chain holds and what a human holds is drawn by what the machine can do to standard this year, and it has only ever moved in one direction.

## The pair

Two humans hold the vector, and the doctrine gives each of them one job and the other’s as a standby.

The aligner owns the result. She holds the daily review, samples the gates, listens to a sample of the agent’s conversations, and signs the mandates, and she answers for the fifty and, with the delivery vector’s aligner, for the twenty-five. The tuner builds and tunes the chain: she watches which first messages are answered and which are not, which qualified companies turn into calls and which turn into nothing, which signed mandates closed and which were shoppers, and she changes the agents so that the next month is better than the last. She also signs when the volume needs it, and because she does, she knows every client in the vector as well as the aligner does. When one is away, the other runs the vector without a day’s loss.

This is what the doctrine means by standby designed in, not improvised. In the old company, when the person who held a client went on leave, someone else was asked to cover, and the cover was a person who did not know the client, the history or the promises made. The client noticed. In a vector the standby is named before it is needed, and she already knows the chain, because she built it; and the clients, because she has spoken to them.

## No silent stalls

Henry Ford’s moving line had one property that every office in the century that followed failed to copy. When a station stopped, the line stopped, and everyone could see it. A stall could not hide.

The old office was built of silent stalls. A file sat on a desk for a week because its owner was busy, or on leave, or had decided to go and had not said so, and nobody downstream knew it was sitting there until a customer rang to ask. The stall was invisible because the work was invisible. A department could have twenty stalled files on any given morning and every report it sent upward would be true.

A vector cannot stall silently, because every agent in it reports its own flow, all the time. The find agent says how many companies it found today; the approach agent says how many messages it sent and how many were answered; the booking agent says how many conversations are in the diary. If any agent produces nothing for longer than the playbook allows, the quartermaster raises an alarm, the tile turns amber in the master’s control room, and the agent maker or the tuner is told before the aligner’s next review. A human gate that has waited too long is the same: a signature not given, a mandate signed and not handed over. The vector knows where everything is, and it says so. This is not a feature added to the design. It is the design: a vector that can stall silently is not a vector; it is a department with software.

## When a vector splits

A vector grows until its pair cannot hold its gates inside the hours they have committed to, and then it splits. The test is simple and is set in the playbook: when the signatures and the sampling of the gates no longer fit the daily contract of two people, the vector becomes two, the agent maker clones itself into the new one, and a new pair takes it.

There is a second reason to split, and it matters more as a company matures. A vector splits when its market divides. GoodSpace’s eight winning vectors today are identical. They talk to every kind of client, and the only thing that separates them is that each has its own leads. That is how a company should begin: copy the playbook, give each copy its own leads, and learn. But the vectors are learning, and what they are learning is that a manufacturer hiring a plant head and a software company hiring an engineering lead want different first messages, different proof, different conversations. As they learn it, they will specialise, one industry to a vector, each with a chain tuned to its market. A vector starts as a copy and becomes a specialist, and the company’s knowledge of its markets is the sum of its vectors’ specialisations.

## Count vectors, not employees

The old company measured its size in people. It reported headcount to its board, celebrated when it grew and worried when it shrank, and gave its managers status in proportion to the number of people beneath them. Headcount was the measure because people were the unit of production; a company did more by employing more.

In the collapsed company the unit of production is the vector, and so the measure of size is the number of vectors and the results each one carries. GoodSpace runs twenty vectors across its four Outcomes, each held by one or two people. That sentence describes the company better than any headcount could, because it says how many results the company is carrying at once, and with how much human intention behind each. Two people to a vector is the shape of the doctrine, and the default. A company that needs four people to a vector has not collapsed it. The playbook may write an exception, with its reason: a vector whose gates are few and slow may be held by one person with a named standby in the neighbouring vector, and a vector whose irreversible gates arrive faster than two people can sign may carry a third signer for a season. The exception is written, dated and read at the Monday test, and the doctrine tile counts any vector that is neither a pair nor a written exception. A vector held by one person doing both jobs, as some of GoodSpace’s still are, has no standby, and will find out on the day she is ill; it is a gap, and the company counts it as one.

A Keeper who wants the company to grow does not ask how many people to hire. He asks how many vectors the market will support, how many results each will carry, and how many pairs he can find whose intention he would pay for. Growth becomes the opening of vectors, each from a playbook, each with a pair; and the cost of growth becomes the cost of a pair and a chain, which is a fraction of the cost of a department, and is known in advance.

## The design and today

I have described the vector as it is designed. Here is the vector as it is today, because a doctrine that shows only its designs is asking to be believed rather than checked.

The design is fifty won and twenty-five closed for every matched pair of vectors. Today GoodSpace’s delivery vectors close eight to twelve mandates a month each, not twenty-five. Its winning vectors are not yet at fifty. And there are eight winning vectors and ten delivery vectors, which means the pairs are not yet matched one to one. The gap between the design and today is roughly a factor of two to three, and it is the most important number in the company.

Three things close it, and each is a change to a vector, not to a department. Delivery must get faster: more of each mandate’s candidates drawn from the pool of people already assessed, batches sent sooner, feedback turned into a sharper search the same day it arrives. The winning vectors must win better mandates, not only more of them: the conversation agent and the findability report filtering out the shoppers before they become half the failures. And as both improve, the share of signed mandates that reach a placed candidate will rise above half, so that fifty won produces more than twenty-five closed. Each of these is a tuning problem, owned by a tuner, visible every day on a tile, and none of them needs a meeting.

That is the vector: a result drawn where the customer feels it, a chain of agents with skills at the gates where they are needed, one human gate where intention matters, a pair who each can do the other’s job, an alarm on every stall, a rule for when to split, and a number that says honestly how far it is from its design. Everything else in this part of the book is about the people who hold vectors, how they are chosen and paid.

## Design a vector on one page

For the reader who wants to build one, here is the page. If you cannot fill every line, you are not ready to open the vector.

1. **The result.** What does the customer receive, and where exactly is the line drawn? If the result is shared with another vector, which one, and what is the shared number?
2. **The market.** Who is this vector for, and how does it avoid chasing the same customers as its neighbours?
3. **The target, worked backwards.** The result per month, and every number upstream of it, with the drop-off rate at each step.
4. **The chain.** Each agent, in order, with what it takes in and what it hands on.
5. **The skill agents.** Which checks the result needs (legal, compliance, finance, quality, verification), and at which step each one stands.
6. **The gates.** Which are reversible and default open, and after how long, and which are irreversible and wait for a human. For each human gate, write down what an agent would have to achieve to take it over.
7. **The pair.** Who aligns and who tunes, how many hours each commits to a day, and how the gates are shared.
8. **The standby.** Who runs the vector when either is away, and what she already knows.
9. **The alarms.** For each agent and each human gate, how long it may go without producing before the quartermaster raises a stall.
10. **The split rule.** At what volume, or at what division of the market, this vector becomes two.