Running a campaign

What a campaign is

A campaign is one piece of outbound work: a list of companies you want to reach, the people at those companies, the messages sent to them, and whatever came back. You start it with three things — a name, a brief in your own words, and a language. Everything else is built up underneath it afterwards, mostly by agents, always with you holding the last switch.

You find it in the Revenue section. The left-hand list is your campaigns; clicking one opens it. A campaign that has nobody written to yet is not an empty campaign — the list will tell you so, saying something like “37 companies, none written to yet” rather than “nothing queued”.

The whole journey, in one picture

Step 1 — Start the campaign

In the Revenue section, press + beside Campaigns, or click the New campaign row at the bottom of the list. You are asked for three things.

Name. Whatever you will recognise it by. FUFLD · Sweden is the sort of thing that works.

Brief. Who this campaign is looking for, and what you are offering them, in ordinary sentences. This is not decoration — it is the actual instruction the discovery agent works from when it goes looking for companies. A vague brief produces a vague list. Say the country, the kind of business, the size, and anything that should rule a company out.

Language. A fixed list of eight: English (UK), English (US), Swedish, Vietnamese, Norwegian, Danish, German, French. It is a dropdown rather than a text box on purpose, and it matters more than it looks.

A campaign starts as a draft with nobody in it. That is normal and expected.

Step 2 — Companies appear

Open the campaign and you land on the Companies tab. This is the list the owner sees and wonders where it came from. It came from an agent.

Asking for more

Top right of the Companies tab is Find more companies. Pressing it opens a chat with your discovery agent — by default she is called Artemis — and shows you the exact sentence to send her, which names the campaign and its id so she files what she finds in the right place. It looks like this:

Find more companies for the FUFLD · Sweden campaign (campaign id abc123).

You have to actually send that message. The button opens the conversation; it does not submit anything on your behalf, and it never searches by itself. That is deliberate: finding companies is a job with a cost and a judgement in it, so it happens in a conversation you can steer, interrupt, and argue with.

If the button is missing and you are told “Nobody on this instance finds companies yet”, there is no discovery agent on your instance. Add one from the agent templates and come back. See Who does what.

What the stages mean

Each company carries a stage. The strip at the top of the campaign counts them; the buttons on the Companies tab filter by them.

The first three always show, even at zero, because “0 qualified” is a fact worth seeing. The last three appear only when something is actually stuck in them, tinted amber so a stall never hides inside a total.

What the score is

Themis scores each company out of 100 across eight things: brand strength, room to grow on Amazon in the US, logistics fit, how deep the product range is, fit with the US market, how reachable the decision-maker is, signs the business is growing, and fit with what you are offering. The score and its tier show under the stage chip in the table.

Two habits of this agent are worth knowing, because they look like faults and are not. It leaves a component blank rather than scoring it low when the evidence is missing — a company nobody could find revenue figures for is not the same as a company with bad revenue figures. And when the blanks could still lift a company out of the bottom tier, it parks it at needs more research instead of disqualifying it. Missing evidence never quietly throws a company away.

Step 3 — People and addresses

A company is not someone you can write to. Ariadne, the contact finder, takes a qualified company and looks for a named decision-maker and an address that reaches them. She works the cheap evidence first — public pages, free registries, the pattern other addresses at that domain follow — and only asks to spend money on a paid lookup if you approve the budget.

Every address she finds is graded by how strong the evidence behind it is. In plain terms:

There is one more sharp edge here. The rule above — probable needs a reviewer, unknown and catch-all are never written to — is enforced strictly when a campaign’s contacts are exported to an outside sending tool. Inside Neutron, the approval screen shows you the grade as a check mark beside each message, but it does not disable the Approve button. The judgement is yours, and the screen gives you what you need to make it. Look at the check before you click.

Step 4 — What must exist before anything sends

This is the part worth being blunt about. A qualified company with a named person and a probable address still cannot be written to until five separate things are true.

A sending domain that is warmed and not paused. Mail from a brand-new domain goes to spam. Neutron makes a domain earn its volume over seven stages before it will send at full rate, and a background watchdog pauses it if the mail it sends starts bouncing. See Email for what the states mean.

A sending address on that domain, switched on. An address like [email protected] that belongs to a specific person or agent. Created on their profile, not on the campaign.

A sequence. The schedule of messages: what goes out on day 0, what follows on day 3, what follows on day 7. This is the gap. The place to store one exists; the way to create one does not — no screen, no menu, nothing in the interface. Until that is built, no campaign will queue a message on its own.

Approved claims in the campaign’s language. A claim is one factual statement about your business that you have written down, backed with a source, and approved with a review date. The Claims tab on the campaign shows the ones available in that campaign’s language, and lets you approve or retire them. Approving one asks you for the date its review falls due; past that date it stops being usable until someone renews it.

Your approval, on every single message. Not a setting. Not a batch. Nothing in the system — no agent, no schedule, no worker — can move a message from waiting to approved. Only an administrator clicking Approve can, and the code refuses every other route.

Step 5 — Watching it go

The Sends tab

Every message the campaign has queued, one row per person, in words rather than codes.

A held message is the one that matters, and it is why this screen exists in this shape. Every check the system runs is re-run at the moment of sending, not just when the message was queued — so a message approved yesterday can still be stopped this morning if the ground moved under it. The reasons, in plain words:

• on the do-not-contact list — this address has been suppressed, permanently and across every campaign.
• not approved yet — nobody has clicked Approve.
• sending domain is paused — the watchdog stopped the domain, or you did.
• domain hit its daily cap — the domain is still warming up and has sent all it may today. It will go tomorrow.
• no active sender on this campaign — the address it would send as has been switched off.
• no address for this person — there is nobody to write to.

Held is not a dead end. Fix the cause and the message goes back into the queue.

Where you approve things

Two places, same effect. The Sends tab has an Approve button on each waiting row. The approvals inbox — the notifications panel — gathers everything waiting across all campaigns, and shows three checks against each one: whether there is a real name attached to a usable address, whether the campaign has approved in-date evidence to cite, and whether that evidence names its source. The checks inform you; they do not decide for you.

The Mail tab

The same activity arranged as conversations rather than rows: one thread per person, newest turn open, older ones collapsed. Each thread offers three things — reply as whichever address the thread speaks as, hand off to open the full conversation, and suppress to stop all mail to that address permanently and everywhere.

Replies stop the follow-ups

When someone answers, every message still queued for that person on that campaign is cancelled immediately — waiting, approved and held alike. Nobody gets a “just following up” after they have already written back. Messages already sent are left alone, since they cannot be unsent.

Separately, anything that reads as an unsubscribe — the word on its own, “opt out”, “stop”, “remove me”, or mail to the campaign’s unsubscribe address — suppresses that address permanently without anyone looking at it. Every other reading of a reply is left to a person, because guessing wrong either buries a live prospect or invents a deal that was never offered.

Step 6 — Replies become deals

When a reply is a good one, marking it as interested opens a deal on the Deals tab: who it is, what company, what it is worth, and what commission it carries. You never create one by hand.

The deal moves up a fixed ladder, one rung at a time, forwards only:

Handed off → Meeting → Proposal → Won → Commission due → Commission paid

The Deals tab offers exactly one button — the next rung. Moving to Won asks what the deal is worth, and you may leave it blank if you do not know yet. Moving to Commission due insists on a figure, because a commission with no number cannot be reconciled against anything later. If the store refuses a move, the reason is shown next to the button rather than the click silently failing.

Step 7 — The Learning card

The Learning tab appears on a campaign only when there is an experiment running behind it, so an empty tab never invites you to look for a verdict nobody was measuring.

An experiment splits the people on a campaign into groups — arms — each getting a slightly different subject line or wording, plus one group called the holdout that gets the ordinary message unchanged. The holdout is the control: it tells you what would have happened anyway. “Beats the holdout” means a variation did better than doing nothing different.

The card will refuse to give you a verdict in two situations, and says which.

Not enough people. Below 20 people in either the winning arm or the holdout, there is no verdict — only the numbers. Twenty is not a formality. Below it, the difference between a 4% and an 8% reply rate is indistinguishable from luck, and a verdict there would be worse than no verdict, because you would act on it.

Too close to call. Above twenty, the card compares the plausible range of each group’s true reply rate. It only declares a winner when those ranges do not touch at all — when the winning arm is better than the holdout even taking the least flattering reading of the winner and the most flattering reading of the control. Otherwise it tells you the difference could still be chance.

The card notes at the bottom that it recomputes on every load with no stopping rule, so an early verdict is provisional. The honest reading is the one from a closed experiment.

Where things actually stand

Where to go next