Outreach scripts

A script is a message your team sends to leads. Instead of every setter writing their own, or one message living in somebody's notes, the workspace keeps its scripts in one library, so a lead always starts from what is known to work and the numbers show which wording earns the most booked calls.

Open the library from Settings › Sales, on the Outreach scripts card, with Open scripts. A manager also gets an Add one link straight from a lead that has no script yet. The card says how many scripts are in use, and shows Waiting for your approval in amber to a manager when something needs a decision.

Scripts belong to an offer. The page opens on one offer, with a switcher across the top for the others and for Company-wide. Each offer has a sequence, in the order a lead meets it:

• First message. Sent to every new lead.
• Follow up 1, 2, 3 and so on. Follow up 1 is sent when the first message gets no reply, follow up 2 when follow up 1 gets none, and so on. Add as many as you use.
• After a no. Sent once, to someone who said no to this offer, to offer them a different one. You choose which offer it moves them to. It is there once the offer has a first message and another offer to move people to.

An offer with no first message or follow up of its own borrows the Company-wide ones, so a new offer is not left without a message while its own library is being written. The message after a no is never borrowed.

Write {first_name} or {company} where the lead's details go. The setter's panel fills them in from what the app knows about the lead, and a lead's name is left out gracefully when there is none, so "Hey {first_name}," becomes "Hey there,". The structure, the offer and the price are kept exactly as you write them.

Everyone in the workspace can read the library and add a script. Changing what is sent needs an owner, admin or manager. See Who approves what.

The system chooses, never the setter. The setter panel on a lead (see Working a lead) hands over the message that lead is due, and there is no menu of scripts to pick from. That is on purpose: if setters chose, the numbers would measure the setter's taste instead of the script.

The system decides two things for each lead:

• Which message is due. A lead nobody has messaged gets the first message. A lead that did not reply gets the next follow up. A lead that said no gets the message after a no, if the offer has one, otherwise nothing. A lead that is interested, booked, past the workspace's give up limit, or already offered another offer gets nothing: a person takes it from there, and the panel says so plainly.
• Which version of it. Most leads get the version that books best, which the library calls Most leads get. A share get a version being tested, labelled Being tested. The same lead gets the same version every time the page is opened, so nobody can refresh until they see one they like.

How big that share is, 30 percent to start with, is one of the numbers a manager can change. With no tested version, every lead gets the best one. With no best version yet, leads are split between those being tested. And when the best version has stopped booking and nothing has replaced it, half the leads go to the tested ones, so a replacement is found twice as fast.

Scripts are ranked by calls booked, because a booked call is what outreach is for. Replies and yes answers are counted and shown, but a reply is not a win: someone who said no replied, and it is kept apart from silence.

The library says it in plain words. Each live version shows how it is doing, such as Books 1 call per 18 sends, or Judged after 12 more while it is still new, with the supporting numbers underneath (sent, percent who replied, percent who said yes, how many were changed before sending).

What happens, and when:

• A send counts as judged once it has an outcome recorded, or once it is three days old, which is long enough for a reply to have had its chance.
• A new version is compared with the best one only after both have 40 judged sends each. A version that books clearly more takes over as the one most leads get, and the old one is stopped. A version that books clearly less is stopped. Clearly means the system is at least 95 percent sure, so two bookings from 20 sends does not decide anything.
• A version that books no calls after 60 judged sends is stopped.
• A version still level with the best one after three times the comparison number (120 judged sends to start with) is stopped too, and the best one stays.
• The best version never disappears on its own. If it books nothing after 60 judged sends it keeps being sent, because the team has to send something, and the library notes that bookings have dropped off while a different approach is written.

Judging runs by itself in the background after each send and each recorded outcome, and again when anyone opens the library, so the page never shows a verdict the numbers have already reached. It only ever chooses between scripts a manager has already approved.

The three numbers

Under How versions are judged, at the foot of the library, a manager can change the three numbers in plain sentences. Everyone else reads them as sentences.

A stopped version is not lost. It sits under Stopped, with why it was stopped and its numbers, and Use again brings it back to be tested, or as the main version if that place is empty.

Nothing changes what setters send without a person agreeing to it.

• A manager, admin or owner adds a script and it is live at once: the main version if that place had none, otherwise a version under test. Add a version to test is on every message in the sequence.
• Anyone else can add a script too, and it waits. They are told a manager approves it before it is sent, and they see it under Waiting for approval.
• Everything the AI suggests waits for a manager, without exception, whether a manager asked for it or the system did.

What a manager sees at the top of the library is Waiting for you, with each item written out in full. A script from your team has Approve and Decline. An AI suggestion has Approve and start testing and Decline. The words of either can be edited before you approve. A completely different approach covers the whole sequence, so it is approved or declined as one.

Approved words are frozen. To change a script that is live, add the new wording as a new version and let the numbers decide. That keeps each version's results honest, because they are always the results of the same words.

A manager can also Stop sending any live version at any time. Its results stay.

Under Ask the AI a manager has two buttons:

• Suggest a better version changes one thing in a message, to see whether that books more. It needs a live first message to start from, and it says what it changed and what it is testing.
• Try a completely different approach researches your buyers first, then writes a new sequence on a different premise, with why the old approach stopped working. Under Point the research you can say who you are selling to and anything you want found out.

Both land under Waiting for you, and are written in the background: a version takes under a minute, and a different approach a couple of minutes because it reads up first. A different approach also tells you if it is close to the old one reworded.

The system asks by itself in one case. When a version is stopped for not booking, or the main version stops booking, and nothing else is being tested, it asks for a completely different approach. It still waits for your approval.

Research runs only then: when an approach is stopped, or when a manager asks. It never runs per message. A brief younger than two weeks is reused instead of paid for again, and what it found is kept on the library as Your buyers, with what hurts, what makes now the moment, how they say no, and where it came from.

The library also shows Why people said no: the reasons your setters logged on the lead page, as percentages. The AI reads these when it writes the next suggestion.

If a suggestion could not be written, the library says so in red, states that nothing in your scripts changed, and offers Try again or Dismiss.