Decide who is worth a conversation
Not a persona workshop. A scored set of bets you can kill. Each bet gets a list, a message, a channel, a conversion number. The ones that do not convert stop consuming inventory.
A volume target without an ICP is a working-capital problem.
You still have to build, ship, demo, and support. If you do not know who buys, that spend goes against guessed buyers. You miss the number because the right people were never found, or you spend against the wrong ones.
A salesperson cannot fix that. Their job starts after someone has already been named.
What you are buying
The work is the unglamorous half of revenue: who is worth a conversation, a list you can actually work, a phone number and an email on each name, outreach that does not need a human on the other end of the line, and a database clean enough to learn from. Monday's closer uses Friday's result instead of rediscovering it.
A salesperson's unit of work is a meeting. A GTM engineer's unit of work is a system that produces the right meetings, every week, without hiring in lockstep with the quota.
From a finance seat: one is a variable cost that scales linearly with conversations. The other is a fixed cost that raises the yield of every closer, every rental slot, and every demo unit you put on a truck.
A salesperson
Closes the deal once someone is already in conversation. Harvests what is in front of them. The right last mile.
A GTM engineer
Inventories the market. Tests who will take a robot. Makes the next closer inherit a universe instead of a blank search.
Why not just hire another closer
You do not have a coverage problem you can hire your way out of. You have a finding problem. The people who will actually take a robot do not share an industry code. There is no list to buy.
A second AE doubles the conversations. It does not double the number of correct conversations. Hardware makes the miss more expensive than software. A bad robot meeting wastes a unit, freight, a technician, and a week of calendar. Until someone is scoring who is allowed to buy, and who will reject the category on sight, every AE you add multiplies that waste.
Five things. All cheaper than another quota.
Build the lists. Names, emails, mobile numbers.
Target accounts first, then the human who can sign. Waterfall enrichment until there is a verified email and a number that connects. The output is not a TAM slide. It is reachable records a closer can dial.
Run email and LinkedIn so the closer only talks to the warm ones
Asynchronous motion does the first touches. Phone does the ones that matter. Route the motion and the product before anyone books an hour. You pay for human time after a machine has already decided the account is worth it.
Keep the database clean and usable
Existing customers, rentals, inbound from the videos: one table, a next action, one record per person. Cleanliness is an operating rule, not a one-time scrub. A dirty file is a tax that grows with every AE.
Gather as you go, so the next quarter is cheaper
Every email, LinkedIn touch, connect, and "we don't do robots" writes back: which bet, which product, who said no. The ICP becomes a conversion table. Applied AI only works if the history is trustworthy. CAC should fall as the file gets smarter.
What shows up in the model
| Salesperson | GTM engineer | |
|---|---|---|
| What you buy | Conversations and closes | A repeatable way to produce the right conversations |
| Cost behavior | Linear. More pipeline ≈ more heads | Fixed. One system, many closers |
| Time to first useful output | Ramp plus a territory that may be wrong | Lists and a scored ICP in weeks, then compounding |
| Risk if ICP is unknown | Burns demo units and travel on the wrong buyer types | Tests a few buyer types, kills the ones that do not convert, puts inventory behind the one that does |
| What the company owns | A book that lives with one person | A list and a database the whole team uses |
| What shows up in the model | Another fully loaded AE against an unproven CAC | Lower cost per qualified conversation; fewer stranded units; CAC that should fall |
- Cost per qualified conversation If you do not know the buyer, this number has no floor. A GTM engineer's job is to put a floor under it.
- Units committed vs. units placed Building ahead of a guessed ICP is inventory risk. Building against a list that is already answering the phone is a production schedule.
- Leverage on the next AE The second closer should inherit a working universe, not start from LinkedIn search.
- Whether the file gets smarter Every no is a data point if someone writes it down in a place you can trust. If they do not, next year you pay again to relearn the same rejections.