You have something that sells. Somewhere out there are the companies that would buy it: a few ready this month, most not. Go-to-market is everything between those two facts and revenue arriving on its own. Deciding who is worth your effort. Noticing when one of them becomes worth it now. Reaching them with something worth reading, and keeping what is already moving from going quiet. Done once, that is a campaign. Done so it happens again next month without anybody assembling it by hand, it is go-to-market, and it is what the letters GTM stand for in the emails.
The word is also used more widely, for everything a company decides about how it sells: what to charge, how to package it, which channels, how to pay the people who sell, what happens after the signature. This page is about the part that arrives in your inbox, which is the part that has changed. And if you are 12 people with two of you selling, most of what follows is about businesses bigger than yours; the parts that are yours are the fourth job and the questions at the end.
Open postings, March 2026
3,000+
for a job title that did not exist in 2024.
One contact-data credit
5 cents
where a seat on an annual contract used to be.
Replies to a cold email
3 in 100
the average across 2025.
02Why you hear it so often
Why you have heard the word so often lately
Three things happened at once, over roughly two years.
The work got a new name. Sales operations, marketing operations and demand generation folded into one function, and the people who wire the tools together started calling themselves go-to-market engineers, a title that started appearing in job postings around 2024 and had more than 3,000 open postings by March 2026.
The tools got cheap. Contact data that came on an annual contract, per seat, now comes by the credit, from about 5 cents a credit and several credits to a verified contact, and the software that strings it together charges for what it runs rather than for who is logged in.
And the parts of the job that used to take a person a morning, the research on a company, the filling in of a record, the first draft of a message, the watching, can now be done by a model in seconds.
Put those together and the cost of sending 1,000 emails fell to nearly nothing. That is why the average reply to a cold email is now about 3 in 100, and why, since February 2024, the biggest mailbox providers reject bulk mail that is not authenticated or that too many people mark as spam. It is also why a lot of what is sold as go-to-market is a list and an email sequence with a new name on it. It is what happens to any field when the cheapest part of it becomes the easiest to sell.
The rest of this page is the whole of it, from the top down.
a person reached out55
a partner brought it25
it grew from an existing account15
they found you5
shares of last year's first conversations · one motion and a half · illustrative
Every proposal you receive uses three or four of these, and most of them are older jobs wearing new titles.
How it works
sales operations, the people who kept the CRM and the forecast honest
marketing operations, the people who ran the email tool and the website forms
demand generation, the people who bought the ads and counted the leads
growth, the software version of all three, measured on signups
revenue operations, the three folded into one function that owns the data across sales, marketing and account management
go-to-market engineering, one technical person who wires the data, the signals, the routing and the models together, which used to take three of the above
and sales development, the junior seat whose whole job is the first conversation, which is the seat most of the automation is aimed at
What changes
You can read a job title, a proposal or a product name and place it on the shape below, which is the only test that matters.
The data · and what it is measured againstattribution · revenue operations
1 · Decide who is worth effort at all
Not an industry code and a headcount. The companies that buy from you, come back, and do not cost more to serve than they pay have something in common, and it is almost never the thing on the filter. For a wholesaler it might be two or more sites with order intake still running on email. Written down as rules a person can apply, that description is an ideal customer profile, the ICP: the one sentence that says who this is for, which everything after it depends on.
Split into groups that buy differently and need different words, the profile becomes segmentation. Counted, every company that fits it is your total addressable market, a number you can defend because there is a list underneath it rather than an analyst's estimate. And the part of that list you have never spoken to, once your customers, your open quotes and your one-time conversations are taken out, is whitespace: the market you were not competing for, with names on it.
2 · Detect when one of them becomes worth effort now
A company that fits is not a company to call today. Something has to change first. Any observable change that has, in your own history, come before a purchase is a buying signal: a role posted that only exists when a company has your problem, a customer whose orders have slowed against their own rhythm, a public contract that comes up for renewal. The best of them are in your own records, free, and nobody sells that.
The part of a system that does the watching, continuously and for every company on the list, is signal detection, and the smallest version of it, one thing watched, one list, landing on the people who act on it, is a watchlist. Evidence bought from outside that a company is researching something like what you sell is intent data; it comes in three grades that are routinely sold as one thing: behaviour on your own site, which is precise; review-site intent tied to a named account, which is usable; and third-party content consumption, which is a guess with a logo on it. Filling in what a record is missing, the role, the size, the contact details, the systems they run, is enrichment. And everything worth knowing about one company in one place, cut down to the two minutes a person needs before a call, is account intelligence.
3 · Act on it with the context attached, without a person assembling it
The signal fired, the record is complete, the reason is known. Now somebody reaches them, and the words come from what changed rather than from a template with a name dropped in. The timed run of messages that carries it, which should stop the moment somebody replies or something else changes at that company, is a sequence; almost every one you have received was set up to send rather than to stop. In a market where the buyer has known the seller for 10 years, this job is a phone call, and the machine's part is knowing which call.
Contacting companies that have not contacted you, by email, phone or LinkedIn, is outbound. Running the ads and the content against that same short list of companies rather than against an audience guess, so the person who had the email also sees the ad, is account-based marketing, and showing the follow-up to people who already visited your site is retargeting. Deciding whose desk it lands on, with the reason attached, and what happens if nobody touches it by the deadline, is routing.
4 · Keep what is already moving from stalling
Finding companies is half the job. The other half is what happens to the quotes, bids, job orders or opportunities you are already chasing. The set of them, with a stage and a value on each, is your pipeline, and it is only as true as the last time somebody typed into it. What counts as stalling depends on the business: three weeks of silence on a wholesale quote, four months of it on a part that is being designed in over 18. This is the one job that applies to a company of any size, because everyone has something out that has gone quiet.
A deal that depends on one person at the customer, who may leave, go quiet or be overruled, is single-threaded; building the relationship on purpose with the three or four people who will decide it is multi-threading.
Underneath: the data, and what it is measured against
None of the four works alone. The customer file, the order history, the CRM if there is one, the mailbox and the website all have to say the same thing about the same company, and somebody has to keep them saying it. If there is no CRM, the list lands in an inbox, a shared sheet or the ERP; a CRM introduced before the four jobs exist is an empty box that asks salespeople to type, and if they were going to type they would have typed by now. Working out which of all this produced the work you won, so next month is tuned on what paid rather than on what was busy, is attribution, which is hard enough to have its own section below.
And there are parts of go-to-market nobody emails you about, because they cannot be sold as a list: what you charge, how you package it, what your salespeople are paid to do, and what happens after the signature. They are not on this page, and a proposal that skips them is not wrong to. It is just not the whole of it.
The function that owns the data across sales, marketing and account management at once is revenue operations.
ideal customer profile
One of a wholesaler's prospects, from the day it fits to the day somebody calls. Figures `illustrative`.
How it works
the profile says two or more sites, order intake by email, replaced the ERP in the last three years
the company matches, so it is one of the 1,400 on the list, and since nobody has ever spoken to it, it sits in the whitespace
in March it registers a second site and posts a job ad for a purchasing lead, a role that in your own history has come about four months before a supplier change
the record is enriched: the sites, the buyer's name, what they run
the signal, the fit and the history are scored against the other 59 that fired this month and it ranks fourth
on Thursday it lands as a task on the desk of the person who covers that territory, with the reason in one sentence
she calls, and the call, the reply and what was said go into the record without her typing
in June the quote goes quiet, the pipeline flags it at 21 days, and the second person at the customer, the one who signs, gets a call too
What changes
Nobody assembled any of that. The list, the reason, the timing and the follow-up arrived, and a person spent the week on the conversations.
A lead is a person who did something: filled in a form, replied, downloaded a thing. It is the unit most marketing reporting counts, and the wrong unit for everything above, because companies buy and people click. A lead from a company that fits the profile is job two firing. One from a company that does not is noise with a name on it, and most inbound is the second kind. The word survives because it is easy to count.
No. The profile describes the company: what it has to look like for the purchase to make sense. A persona describes the person inside it: the role that feels the problem, the role that signs, and what each of them is measured on. You need both, and the profile comes first, because a perfect persona at a company that will never buy is a well-described stranger.
Marketing is one of the ways to do job three, and some of job one. Go-to-market is all four jobs and the floor, whichever department does them. In a wholesaler with no marketing department the four jobs still exist; they are done by inside sales, the ERP and one person's memory. Naming the whole thing is what lets you see which of the four nobody in the building owns.
04The motion
The motion
How the four jobs are arranged for a given business has a name too. A motion is the route a company relies on for most of its revenue: who makes the first move, who does the convincing, and how much of it a person has to do. Four are in common use, and most companies run one and a half. Two more that the lists leave out, and that most of the companies on this page run: the relationship, where the buyer has known the seller for 10 years and the conversation is the whole job; and inbound, where people search for you and the job is answering inside the hour.
[object Object]
A person finds, qualifies and closes. The right shape when the purchase is large, technical or bound up in a relationship: a component manufacturer selling into somebody else's design, where an engineer decides 18 months before the purchase order, or a mechanical subcontractor whose work comes through a general contractor who has known them for 10 years. The four jobs put that person in the right conversation earlier; they do not replace it.
[object Object]
The product does the first selling. Somebody signs up, uses it, and the account grows from inside. Only possible when the thing can be tried without a person, so it is mostly software; the four jobs then read the usage rather than the news, and the buying signal is a trial that added six seats.
[object Object]
Somebody else sells it for you: a distributor, a reseller, an integrator, a marketplace. A fastener maker selling through wholesalers, a software company selling through the firm that installs it. The four jobs run on two lists at once, the partners and the customers behind them, and the hard question is which of a partner's customers are worth the partner's effort.
[object Object]
The list is short and named, and every channel, the email, the ad, the call, the content, points at the same 200 companies and nobody else. It is less a separate motion than a way of running the others when the market is small enough to name. A staffing firm with 200 old clients is account-based whether or not anybody there uses the phrase.
Naming yours matters because a tool built for one motion, sold to a business running another, is the most common way money is wasted in this field. A sequencer built for 5,000 sends a week has nothing to offer a subcontractor with 40 general contractors.
Most companies run one and a half motions and buy tools built for a third.
How it works
take last year's revenue and mark where each account's first conversation came from: a person reached out, the customer found you, a partner brought it, or it grew from an existing account
the largest share is your motion, and the second largest is the half
check the tools and the services you pay for against that: a sequencer sized for 5,000 sends a week on a sales-led motion with 40 accounts, a product-usage tool on a business with no product to try, a partner portal with no partners
then check the four jobs against the motion: on channel, job one has to be done twice, once for partners and once for their customers; on product-led, job two reads the usage and not the news; on account-based, job three has to say the same thing on every channel to the same 200 companies
What changes
You stop paying for the motion somebody else runs.
Almost everything written about go-to-market is written about software, where the product is tried online, the buyer is found on LinkedIn and the signal is a visit to the pricing page. Most businesses are not that. The four jobs do not change. What changes is where the data lives, what counts as a signal, and what you are allowed to do once you know.
a wholesaler
the order file
a component manufacturer
the product programme
a mechanical subcontractor
the permit register
a staffing firm
the careers page
a software company
the product itself
a wholesaler
the order file
a component manufacturer
the product programme
a mechanical subcontractor
the permit register
a staffing firm
the careers page
a software company
the product itself
A wholesaler.
A customer ordered every three weeks for six years. The last order was in April; it is now September. The person who owned them left in March and the account went to somebody carrying 240 others. Nothing noticed, because every system watches orders and none watches for the absence of one. That is job two, and the signal is your own order history, not anything bought. Job one ranks by margin rather than spend. Job three is a call by inside sales on Thursday, not a sequence, routed to whoever used to own them.
A component manufacturer.
By the time the request for quotation arrives, the drawing is frozen, the model already has somebody's part number in it, and you will be told you lost on price. The decision was made 18 months earlier, when an engineer picked a part and stopped looking. The signal is a new product programme, a line expansion, a new plant, a rule with a date that forces a redesign. The four jobs exist to put your engineer in the conversation while the specification is still being written.
A mechanical subcontractor.
The invitation arrives, the estimator prices it, and it was never coming to you: the shortlist existed before the documents did. Bid cost is the one line in the commercial budget that nobody measures and everybody pays. Job one is a score against the shape of what you win, from your own bid history, so a no-bid has a reason attached. Job two is a watch on the early records, the developer's announcement, the plat, the project registration, months before a bid list exists. Job three is almost never outbound. Your relationship with a general contractor is not something a machine should email.
A staffing firm.
You placed for them in 2021. Their careers page has four openings on it today, one of them open for 30 days, which means their own team cannot fill that one, and nobody in your building knows because nobody's job is to look. Job two is a watch on every company you have ever invoiced. Job three is routing to the recruiter who owned the relationship, with what you placed attached, so an opening is not a lead. It is your opening at your client.
A software company that moved upmarket.
The board asked for bigger deals and you hired two account executives. Your product knows something nobody has asked it: the accounts that renewed, expanded and referred have a shape, sitting in the usage data and the won-deal notes. Job one is derived from that rather than from an industry code. The motion becomes sales-led for the accounts that matter and stays product-led for the rest, and the two lists must not be worked the same way.
Most companies are not missing all five. They are missing one, and paying for it in the other four.
How it works
no profile: the list is bought on an industry code and a headcount, and 9 of every 10 companies on it were never going to buy
no detection: the team calls whoever is next on the list, on their turn, and the good week and the busy week look the same
no action layer: the signal arrives and a person spends 40 minutes finding out who to call and why before the call, so it happens on Monday, or not at all
nothing keeping it moving: the quote goes quiet in week three and nobody notices until the forecast is wrong
no floor: three customer numbers for one customer, a mailbox that knows more than the CRM, and a report every Monday that somebody rebuilds by hand
What changes
You can tell which one you are missing from which of those five sentences you recognised, and that is the cheapest diagnosis in this field.
Five industries, 45 moments on the industries page, one shape underneath all of them.
How it works
something changed at a company: an order that did not arrive, a programme announced, a plat filed, a role open 30 days, six seats added
somebody in your business should have known, and would have acted
nothing was watching, because the change did not happen inside your system and nobody's job was to look outside it
the four jobs are the answer to that shape: decide which companies are worth watching, watch them, hand what fires to the person who would act, and keep the ones already in motion from going quiet
what differs by industry is where the watching happens, in the ERP, on a careers page, in a permit register, in the product
and what you are allowed to do next, which changes between Europe and the US more than anything else on this page
What changes
You stop reading about software companies and recognise your own week in the same four jobs.
Written September 2026. This section dates faster than the rest of the page.
Data and enrichment.
Until recently you bought contact data from one provider, per seat, on an annual contract, and got what that provider had, which for phone numbers and email addresses was often under half of what you looked up. Two things changed. Coverage: the software now asks several providers in turn for the same record, stops at the first that answers, and charges only for the hit. That is a waterfall, and it takes a match rate one provider gets to a third or a half of records up past 80%. And price: the credit replaced the seat. Clay charges by the credit from about 5 cents with no limit on users; Apollo publishes per-user prices from $49 a month with a credit allowance; ZoomInfo still sells on a quote, at a median of about $33,500 a year by one benchmark. In Europe, Cognism and Dealfront sell the same thing built to the GDPR and the national do-not-call registers, at similar prices. The data is no longer the scarce part of any of this.
The system of record.
A CRM is the system that records who you spoke to, what happened and what is open. Salesforce and HubSpot hold most of the market. What matters more than which CRM is whether there is one. A staffing firm lives in Bullhorn or a similar applicant tracking system. A wholesaler or a manufacturer lives in the ERP, Exact, SAP Business One, Prophet 21, Business Central, with a CRM module nobody fills in. A contractor lives in Procore, the bid portals and Outlook. Go-to-market work that assumes a CRM assumes a software company. Everywhere else the first job is reading what the business already records, and the CRM, if one is ever needed, comes last.
Sequencers and outbound.
Outreach and Salesloft at the enterprise end, Apollo, Lemlist, Instantly and Smartlead at the self-serve end, all run sequences, and the difference between them matters less than what starts one. Two years ago the trigger was the list: everyone on it, day one. The move since is to the signal: a sequence starts because something changed at that company and stops when something else does. Two forces pushed it. A message written from a specific change at the company replies at several times the 3 in 100 average. And since February 2024, Google and Yahoo reject bulk mail from domains that are not authenticated, offer no one-click unsubscribe, or whose spam complaints pass 0.3%, and Microsoft has followed. So sending now means separate domains, authenticated, warmed before any volume, and watched on placement rather than open rate. That work is deliverability, the part everyone assumes and almost nobody checks. And in Germany, § 7 UWG requires prior consent for advertising email even to a business address, with a narrow existing-customer exception, so job three there is the phone and the relationship, whatever a sequencer's pricing page says.
Intent and signal providers.
Bombora, G2 and 6sense sell evidence that a company is researching something. Most of the category publishes no price; where figures surface they start in the low tens of thousands of dollars a year, and what is bought is modelled, about the company rather than the person, and days to weeks late. The category that grew instead is the signal platform: Common Room, Unify and Pocus watch your own product, your website, your community and the public sources at once and put the result on a person's desk, from about $1,500 a month. Website de-anonymisation, RB2B and Warmly, names the visitor in the US and only the company in Europe, where naming the person has no lawful basis without consent. And the parallel dialer, Nooks and Orum, grew fastest when email got squeezed, which says where outbound moved: back to the phone. The signal that predicts a purchase for a wholesaler is still in its own order file, free, and nobody sells that.
The go-to-market engineering stack.
Go-to-market engineering means all of the above wired together: data in, enrichment through a waterfall, a signal watched, a score, a route into the CRM or the inbox, a sequence if one is wanted, a report at the end, built by one technical person in Clay or a similar tool rather than by a team of four. It is a real change in who can build this and what it costs. It is also, in most of the cases where it goes wrong, the whole of job three built before anybody did job one.
Two years ago
Now
per seat, on an annual contract
Now
per credit, for what runs
one provider, whatever it had
several providers, asked in turn until one answers
the list starts the sequence
the signal starts it, and stops it
the open rate
placement, on separate warmed domains
One company, through the whole stack, at published prices where a price is published. Figures `illustrative`.
How it works
a company enters the list from the market map, which cost nothing beyond the mapping
the record is enriched through a waterfall: three providers asked in turn, one answers, about 5 to 15 cents for the hit
a signal fires: a job ad read from a board every night, or an order that did not arrive, read from your own ERP, both free
it is scored against the profile and the history, which is a rule and not a purchase
a task lands in the CRM, or a row in a shared sheet, or an email to the owner, an action that costs under a cent to run
if outbound is switched on, a message goes from a separate sending domain, authenticated and warmed
the reply is read and routed
the whole run is written back, and at quarter end the report asks which of these companies became invoices
What changes
The data and the software for one company cost less than a coffee. The expensive part was never the stack. It was knowing which companies to put through it.
No. Job four needs one; the other three can run on what you already record. A wholesaler's order history says more about who is worth a call than any CRM would, and a staffing firm's placement history is already in the applicant tracking system. Where there is no CRM, the list lands in an inbox, a shared sheet or the ERP. A CRM introduced before the four jobs exist is an empty box that asks salespeople to type, and if they were going to type they would have typed by now.
Your company's email domain carries a reputation with the mailbox providers, built from every message it has ever sent. Cold volume from it puts that reputation, and every invoice and every reply your business sends, at risk. So outbound goes from separate domains bought for the purpose, authenticated with the three records the providers check, and sent at low volume for a few weeks first, so the providers see normal behaviour before they see 500 a day. Skipping the warming is the most common reason a sequence "does not work": the emails were never delivered, and the open rate said nothing, because it only measures the ones that were.
clay.com/pricing and clay.com/blog/data-waterfalls · apollo.io/insights/apollo-vs-zoominfo · support.google.com, bulk sender requirements · overloop.com on intent data and on § 7 UWG · g2.com, Common Room pricing · salesmotion.io on Unify · warmly.ai on RB2B pricing and EU limits · marketbetter.ai on Nooks pricing · syncgtm.com on Dealfront and Cognism
07AI and agents
AI and agents, the last two years
What got fast. Four parts of the job used to take a person a morning and now take a model a few seconds: research on a company and the person there; filling in what a record is missing; the first draft of a message written from what changed; and watching, which is the one that matters most, because a model does not get bored reading 4,000 careers pages every night. Every sequencer and every data tool now drafts, summarises and scores.
What did not change. Who is worth your effort, and what is worth saying to them. A model can write 400 openings in a minute; none of them is better than the judgement about who should receive one. Where this has gone wrong in the last two years it has gone wrong the same way every time: volume that got cheaper, aimed by a list that was never good, in a voice that was never the company's.
What an agent is. An agent is a model given a goal, a set of tools it may use, and permission to take several steps on its own toward the goal without a person approving each one: read this account, look up what changed, decide whether it matters, write the brief, put it on the right desk. That last clause is the whole difference from a script. An agent decides which step comes next; a script was told. Which is why the most important document in any agent you are sold is the written list of what it will not decide on its own, and who it tells when it stops. Sending anything in your name. Changing a price. Contacting a customer with an open complaint. Marking an account as lost. If nobody can show you that list on paper, the agent has no edges, and an agent with no edges is a liability with a login.
AI SDR. The letters stand for sales development representative, the junior salesperson whose job is the first conversation, and an AI SDR is an agent sold to replace that seat: pick the people, write to them, handle the replies, book the meeting. The category has dozens of products, and most of what is sold under the name is a sequence with a language model on top: job three, automated, with jobs one and two left as they were. Where the list was good it works about as well as a sequence ever did. Where the list was not, it fails faster and in your name. The reported numbers from inside the category, high churn at the first break clause, claimed customers that had left, results that needed a person to correct them, are what a market looks like when the cheapest part of the job is the easiest to sell.
Will not decide on its ownOn paper
send anything in your name
change a price or a term
contact a customer with an open complaint
mark an account as lost
When it stopstells the account owner, within the hour
four examples, not a standard
88 touches before one purchase·4 channels·10 people·9 months·last-click credits the yellow ones
refusal list
The most important page in any agent you are sold is the one that says what it will not do.
How it works
what it may do on its own: read the account, look up what changed, fill in the record, rank it, write the brief, put it on a desk
what it may not decide: send anything in your name, change a price or a term, contact a customer with an open complaint, mark an account as lost, touch anyone on the do-not-contact list
what it does when it reaches one of those: stops, writes down why, and tells a named person within a stated time
what a person checks, and how often: a sample of the briefs each week, every message before it goes while outbound is new, and the list itself every quarter
and where that document lives: on paper, agreed before the agent runs, and changed only in writing
What changes
An agent with edges is a tool. An agent without them is a login with opinions.
Three ways: per seat, per action, and per conversation. Salesforce sells its agents at $2 a conversation or by the action at about 10 cents each, and a sequence that takes 20 or more actions costs more by the action than by the conversation. The stand-alone products mostly price by the month, from a few hundred to a few thousand dollars. The honest comparison is not with a salary. It is with a sequencer and a data subscription, which is what most of them are, and the question to ask is what the agent decides that those two did not.
It replaces the part of a salesperson's week that was never selling: the research, the record-keeping, the finding out who to call. In most businesses that is half the week or more. The part it does not replace is the conversation, and in a market where the buyer has known the seller for 10 years the conversation is the whole job. The honest effect is fewer hours per opportunity, not fewer people. Where a company cuts the people, it usually finds it has cut the judgement the agent was relying on.
Three things, reliably. It invents when it is not given enough: a fact about a company it could not find becomes a plausible one. It flatters the template: given 400 companies, it writes 400 openings that sound personal and are not, because the input was the same. And it cannot tell a signal that matters for you from one that is merely true; that judgement comes from your own history, and if nobody has written it down the model has nothing to reason from. All three are fixed the same way: give it less to guess, give it your own data, and give it a list of what it may not decide.
The rules do not care whether a machine sent it. They care who received it, where, and on what basis. In the US a business email can go cold with an opt-out. In the Netherlands, mostly the same for a business address. In Belgium, not without an opt-in. Calls are stricter again: about a dozen US states need every party's consent before a call is recorded. And the first message a person receives should say where their details came from and how to stop. What a machine changes is the volume at which you can be wrong, which is why the written list matters more for an agent than for a person.
techcrunch.com, 24 March 2025, on 11x · salesforce.com/agentforce/pricing · gtmepulse.com, GTM job market 2026
08Attribution
Attribution, and why it is hard
Attribution is the answer to one question: of everything done to win this work, what did the winning? If you buy ads or content, this is the section you have been misled about, mostly by reports designed to flatter the tool that produced them. If you do not, one paragraph here is yours and the rest can wait: at quarter end, ask the question backwards from the invoices. Of the jobs you won, which were on the list, and what put them there. That can be run from your own order book next week, with no vendor.
Why last-click is wrong.
Most reporting credits the last thing that happened before the form was filled in or the order placed: the ad that was clicked, the email that got the reply. A B2B purchase in 2026 takes around 9 months, dozens of touches across several channels, and 10 people at the buying company, and last-click sees exactly one of those touches and one of those people. It credits whatever sat nearest the end, search ads and the salesperson's final email, and ignores whatever sat at the start, which is the reason they were on the list at all. Multi-touch attribution, which spreads the credit across the touches by a rule, is the same guess with more arithmetic, and Google retired four of its own multi-touch models in November 2023.
Why the account is the only honest level.
People click; companies buy. Asking which of 10 people's clicks won is asking the wrong unit. The honest question is which companies were on the list, which of them did something, which became work you won, and what those had in common. That can be answered from your own records without any browser tracking, because the list is yours and the order book is yours.
Measured to what you won, not to what happened.
Every company on the list carries its history: how it got there, what fired, what was sent, who spoke to it, what happened. At quarter end the question is asked backwards from the invoices: of the 40 jobs won, which were on the list, and what put them there. Signals that came before wins are kept. Signals that fired 200 times before nothing are switched off. Anyone who reports to you in opens and clicks is measuring the tool's activity, not your revenue, and the two are only accidentally related.
A quarter measured from the invoices back, rather than from the clicks forward. Figures `illustrative`.
How it works
start from the work won this quarter, 40 jobs
for each, was the company on the list, and since when
what put it there: the profile, a signal, a referral, an existing relationship
what was the first thing that moved: a call, a message, an ad, a visit, a partner
who at the company was involved, and how many of them you had spoken to
then the other direction: of every signal that fired this quarter, how many preceded work won, and which fired 200 times before nothing
keep what predicted, switch off what did not, and change the profile if the winners did not match it
What changes
The system gets tuned on what paid rather than on what was busy, and nobody in the building has to believe a dashboard that credits the last email.
A rule for sharing the credit across the touches before a purchase: equal shares, more to the first and the last, more to the recent. It was an improvement on last-click, but it depended on following one person across the web, which the browsers have made harder every year and which never followed the other 9 people at the company at all. Google removed four of those models from its own analytics in November 2023. What is left is a model that guesses from the data it has, and the account-level count above, which needs no tracking because the list and the order book are yours.
Run them against the account list rather than an audience, and the question becomes answerable: which companies on the list saw the ad, which of those did anything, and which became work won. That is the same backwards count with one more column. What it cannot tell you is which single click won, because none did. What it can tell you is whether accounts that saw the ads moved faster or more often than accounts that did not, which is the only thing the ad budget was for.
Whoever they are: an agency, a tool, a new hire, a friend with a spreadsheet. The answers tell you which of the five they are selling, and whether they know it.
1Which of the four jobs is this, and what happens to the other three? Anything honest can answer in one sentence. "All of them" is an answer too, and it should come with a reason why one firm is good at four different things.
2Where does the list come from, and what would make a company fall off it? A list is only as good as the rule that removes companies from it. A filter on industry code and headcount is job three with job one left out.
3What is the signal, and how do you know it predicts anything for a business like mine? The right answer names the source, and admits which signals cannot be gathered lawfully in a market you sell into. That list is shorter than people expect and longer than zero.
4What goes out in my name, who decides, and how does it stop? Every sequence and every agent should be able to show you, on paper, what it will not do on its own and who it tells when it halts. If nothing can be shown, nothing is stopping it.
5Where does it live when you are gone? In your accounts under your logins, or in theirs. Written down so somebody who was not there can follow it, or in one person's head. The answer decides what you own at the end.
6What will you measure, and is it clicks or work I won? Opens, clicks and meetings booked are the tool's activity. Accounts on the list that became invoices are your revenue. Anyone who can only report the first is not measuring the second.
7What would make you tell me not to do this? The most useful answer you can get is sometimes that there is nothing here worth building. A seller who cannot describe that situation has never been in it.