The Outbound Signal Engine: Trigger-Based Outreach That Gets Replies

Trigger-based outreach: write when something just happened

A Detroit homebuilder kept asking recent buyers what finally made them move. Couple after couple gave the same answer. They had just come home from a friend's funeral.

So the builder moved its newspaper ads from the real estate pages to the obituaries. Traffic rose 37 percent and ad costs fell 70 percent, Bob Moesta reports in Demand-Side Sales 101. He calls the close rate on those leads unbelievable.

The houses did not change. The timing did.

Trigger-based outreach is the same move for cold email. You write to a company only when a recent, dated event gives it a reason to talk this week.

We hold every message to one check, the Because Test. Finish the sentence "I am writing because" with a fact from the last few weeks. No fact, no send.

We draw one line first. The builder put ads where grieving people looked. We never write to a person about their private life.

Company events are fair. Funerals are not.

Why timing beats a better list

A list tells you who might buy. A signal tells you when.

Moesta describes a buying timeline. First comes a nagging thought that something is wrong. Then comes passive looking, which can last for years.

Passive lookers notice things but plan nothing. An event pushes them into active looking, where they spend time and money on a fix. A deadline then pushes them to decide.

One case in the book is a friend who bought a mattress on a Saturday shopping trip and called it an impulse buy. How long had he slept badly? Two years.

So what does a cold list do? It mails passive lookers at a random moment.

Most of them are not ignoring you. They are not shopping yet.

Lists still matter. In Traction, Gino Wickman tells how his sales training firm counted every company that fit its target: 525 presidents and CEOs. Over time it won more than half of them as clients.

Keep that list. It is the universe you watch, and it tells you how many companies you could ever win.

The signal decides which name on it gets a message this week. A list with no timing is a phone book.

What counts as a buying signal?

A buying signal is a dated event that suggests a company just moved from passive to active looking. It earns a message only if it passes three checks.

You are allowed to collect it. It is still fresh. It earns a specific first line.

Some of the best signals are public records. A private company that raises money under a common exemption files a Form D, a short notice to the SEC with the offering size and the date of first sale.

Others you see by reading: a job post for the role your service supports, a second location opening, or three recent reviews that name the failure you fix.

The best ones are already yours. A trial that stalls at the same step. A contact who backed you at a past customer, now announcing a new job.

Two hard lines. We never buy lists of people, and we never point software at a site whose terms ban scraping.

Code can watch public records and your own data. A person reads everything else. If you cannot write the first line, the signal is noise.

Who deserves a message? Expected profit, not company size

Marketing engineering began as a way to make exactly this call. Gary Lilien and Arvind Rangaswamy put the term on a textbook first published 1998, as the history of the field shows. They meant decisions made with data, a model and a manager's judgment.

Their textbook has a worked example every outbound team should copy. Expected profit is the chance a customer buys, times how much they buy, times the margin. Contact only the customers whose expected profit beats the cost of reaching them.

In the example, reaching each customer cost $3.50. Contacting the four with the best expected profit made $9.29. Contacting all ten made $2.20.

And the three biggest buyers? Contacting them lost $5.02. Big did not mean likely.

A marketing engineer, as we define the role in what a marketing engineer is, builds the system that makes this call every day. A fresh signal raises the chance someone buys. So we score each account before any draft is written.

Fit with your best customers, 0 to 3. Signal strength, 0 to 3. Freshness, 0 to 2.

Draft only at 6 or more. Move the cutoff once replies show which scores pay.

The Because Test: finish the sentence or do not send

The Because Test is our rule that every outbound message opens with a true, dated reason to write now. It takes ten seconds to run.

Write "I am writing because" and end it with a fact. If the honest ending is "you are in our target market," the message dies.

If the ending is "your second office opens next month," it may live. That fact becomes the first line. The ask is one small next step.

Why so strict? The first message has one job. It earns a reply, not a sale.

Every draft also carries the source link in the log. If the approver cannot click through to the fact, the fact does not exist.

No because, no send.

Build it as a chain of small jobs, not one big agent

One AI agent told to find prospects, invent offers and email them will fail in a known way. John Berryman and Albert Ziegler walk through that failure in Prompt Engineering for LLMs.

They gave an all-purpose agent web search and email tools. It ran a lazy search. Then it wrote form letters with the placeholder [your_name] still in them.

Their fix was a workflow. Each task is small, with a clear input and a clear output, and code moves the work between tasks.

We build the engine the same way, as seven jobs: watch, check, research, score, draft, approve, log. The chart below shows what goes in and out of each.

Their book asks a question we put on every job: can this task run without an AI? If yes, use plain code. It is cheaper and more reliable.

Watching a filings feed is code. Checking freshness is code. Writing the first line is where the AI earns its place.

Write each job's brief the way a job spec for an AI agent shows. Keep the briefs, the scores and the banned phrases in your growth repo.

Why a person approves every send

Lilien and Rangaswamy drew the line on automation long before AI agents. Automate a decision when a wrong call is cheap, like a product recommendation. Keep a person in it when a wrong call is costly.

A cold email is the costly kind. It puts your name in front of a stranger. It cannot be unsent.

Chains make it worse. Chip Huyen's AI Engineering does the math: a step that is right 95 percent of the time, chained ten times, ends right about 60 percent of the time.

So the AI researches and drafts. A named person approves, edits or kills every message.

Each kill gets a reason. The reasons go into a banned-language file. The drafting job reads it every time.

Start with a cap of 20 approved messages a day. Read every reply yourself for two weeks. Raise the cap only when complaints stay flat and replies stay good.

The agent drafts. A person sends.

Where outbound engines break, and the fix

How to burn a sending domain in four easy steps:

1) Buy a list of 5,000 people who never heard of you.

2) Tell an agent to "personalize" by praising each profile.

3) Send 500 a day from your main company domain.

4) Count messages sent and call it pipeline.

That is not outreach. That is spam with a damn dashboard.

Polite replies, but nobody books. The signal is real, but the ask is too big. Ask one small question about the event instead of offering a demo.

Every draft sounds the same. Research returns thin facts, so the AI fills the gap with flattery. Make research return the dated fact and its link, or nothing.

One signal fires all day. It is too broad, like "posted any job." Narrow it to jobs for the problem you fix.

How do you know a signal works? Count qualified replies

Messages sent is an activity count. It goes up every time you do more, whether anyone cares or not. That is a treadmill, not a pipeline.

A qualified reply is a human answer from an account that fits, showing interest or asking a real question. Count it once per person. Out-of-office notes, unsubscribes and "wrong person" do not count.

Track it by signal. If new-location signals bring replies and review signals bring none after a fair run, stop watching reviews.

The engine should get narrower before it gets bigger.

Alistair Croll and Benjamin Yoskovitz make a related point in Lean Analytics. How easily you get meetings early is a preview of what selling will cost. They also say to write down each stage of the sales funnel and its conversion rate after the first few sales.

We add one honest test. Hold back one in five accounts that show a signal, and send them nothing for a month.

Does the signal cause the meetings, or just predict them? If the accounts you wrote to book no more meetings than the ones you held back, the signal is not doing the work.

Work backward from revenue with book-a-call math to see how many qualified replies you need. Route every reply in minutes, as speed to lead describes. Grade the engine the way we measure a marketing engineer: on what buyers did.

Which rules apply before the first send?

Outbound email is regulated, and the rules differ by country. The dated box below holds the details we checked. It is not legal advice.

In the United States, the FTC's CAN-SPAM guide says the law covers all commercial email, with no exception for business-to-business messages. Honest headers, a real postal address and a working opt-out are the floor.

Inbox providers add their own rules. Gmail's sender guidelines require email authentication for every sender and set a ceiling on spam complaints.

Boring? Yes. Optional? No.

Canada and the UK go further for some contacts. Have a lawyer review the program once before launch, and again before you add a country or a channel.

Claims inside the email follow the same truth rules as any ad. The FTC risk drivers in marketing claims lists the ones that cause most trouble.

Build your Because Test engine with AI this week

The prompt below builds the engine in your preferred AI. It asks about your customer, your signals and your daily limit. Then it writes the signal table, the score, sample emails, a checklist and the ledger.

It never invents a fact about a prospect. Where it needs one, it leaves a marked gap.

This week, three steps. Today, count the companies that fit. Tomorrow, pick two signals and write the Because line each one earns.

By Friday, draft ten messages, approve them yourself, and start the signal ledger. Next, use the customer truth system to learn which pains your buyers name this week.

The homebuilder did not find better buyers. It found the moment. A list is the map, and the signal is the green light.