The Follow-Up Rules We Turned Into Software

Yesterday we described the four situations a tracked email ends up in and the right move for each. That advice is only useful if you remember to apply it, every day, across dozens of threads. So we wrote it down as three explicit rules and put them in software. This post lists the exact rules, how we know when to stop suggesting, and where they fall short.

Table of contents

We are publishing the rules in full for a simple reason: a suggestion you cannot explain is a suggestion you cannot trust. If the extension tells you to follow up on a thread, you should be able to say exactly why it did, and disagree with it when you know better.

Why rules before AI

It would have been fashionable to start with a language model and let it decide when you should follow up. We did the opposite, on purpose. The decision of whether and when to follow up depends on a handful of facts the tracking pixel already gives us: was the email opened, how many times, how recently, and has anything happened since. Those facts fit in three lines of logic. Putting a model in front of them would make the behaviour slower, more expensive, and harder to explain, without making it more correct.

Rules have a second advantage: they are predictable. Every user gets the same suggestion in the same situation, and when the suggestion is wrong, we can say precisely which rule fired and fix it. Where a model genuinely helps is the part rules cannot do, which is writing the message. That is a separate layer, and it is tomorrow's post.

Rule 1: not opened after 3 days

If a tracked email has not been opened three days after sending, suggest a follow-up with a different subject line.

The reasoning is the one from yesterday's post: nobody has read the offer, so the offer is not the problem. The subject line and the sender name are the only things the recipient has judged. A reply in the same thread asks them to judge the same subject again. A new subject gets a second, independent chance.

Why three days and not two or five? Because most emails that will be opened are opened well inside that window. After three days, the odds that the first message is still going to be read on its own drop sharply, and a new message costs the recipient nothing. Before three days, you are mostly interrupting someone who has not got to their inbox yet.

The rule deliberately does not look at what the email said. It will fire on a quote, an introduction, a job description, or a thank-you note alike. Deciding that a thank-you note does not need chasing is your job, and it takes one click, covered below.

Rule 2: opened, no reply after 2 days

If the email was opened, and two days later no reply has been detected, suggest a short "any questions?" style follow-up in the same thread.

This is the "wrong moment" situation. They saw it, they meant to come back to it, and they did not. The follow-up is not a new pitch; it is a low-effort way back into the thread, and the suggested text reflects that: a couple of lines, one question, no restatement of the original. Two days is enough of a window for someone who meant to reply to do so on their own before the suggestion appears.

Two days is a compromise between two failure modes. Nudging the day after an open reads as impatience. Waiting a week means the context is gone and the nudge has to re-explain everything, which defeats the point of a nudge. If you want the longer argument for the window, it is in when to follow up after an email is opened.

Rule 3: three or more opens in 48 hours

If a recipient opens the same email three or more times within 48 hours, flag the thread as a hot prospect and suggest calling rather than emailing.

This is the only rule that suggests you do something other than send an email, and that is the point. Three opens in two days is the signature of an email being discussed: forwarded, opened on a second device, pulled up in a meeting. We explained the mechanics in why an email gets opened multiple times. While that discussion is happening, another email joins the queue. A call joins the conversation.

The threshold is intentionally low. Three is enough to be unusual for a single reader and low enough to catch the moment early, while the conversation is still going on. It will occasionally fire on someone who simply reread your message three times alone; in that case, a friendly call still does no harm, and you are free to dismiss it.

One caveat applies to any open-based rule: rereading your own sent message can register as an open, with the consequences described in your own opens are polluting your data. Keep that in mind before calling someone on the strength of three opens.

Detecting a reply without touching your inbox

Every one of these rules depends on knowing whether the thread is still waiting for an answer. The obvious way to know is to read your inbox. We do not do that, and we did not want to: the extension has never had access to your incoming mail, and a follow-up feature is not a good enough reason to start asking for it.

So we use the only signal we already have. When you send a new tracked email to the same recipient whose subject is "Re:" followed by the same subject line, the thread is considered active. Somebody replied, you replied back, and the suggestion for that thread stops. It is a crude proxy for a reply, but it is one that comes entirely from the sending side, which is where the extension already lives.

Because it is a proxy, there are two buttons next to every suggestion:

• Already replied, for when the recipient answered by phone, in person, or in a thread the pixel never saw. One click, the thread goes quiet.
• Ignore this thread, for the thank-you note, the internal memo, and anything else you never intended to chase.

Both are per thread, not global. Ignoring one conversation does not change how the rules treat the next one.

The honest limit

These are suggestions, never certainties, and it is worth being precise about why. The rules know what the pixel knows and nothing else. They do not know that the recipient replied yesterday, if that reply is sitting unread in your inbox and you have not yet answered it with a "Re:" of your own. In that case the day-2 or day-3 suggestion will still fire, and it will be wrong. The recipient is not ignoring you; you are behind on your inbox.

The same goes for a reply that came as a phone call, a message on another channel, or a meeting. The rules will keep counting days as if nothing happened until you press "Already replied". And the 48-hour rule can be triggered by a single enthusiastic reader on three devices as easily as by a committee.

We could have hidden these gaps behind a confident interface. We chose instead to show the reason next to every suggestion ("opened once, no reply for 2 days") so you can check it against what you know in two seconds. The rules are meant to remove the work of remembering, not the work of judging.

Where these rules run now

All three rules now run inside the mailcheck Chrome extension, in a closed beta with a small group of subscribers. They produce a daily list of threads to act on, each with its reason, and quick ready-to-use follow-up texts. There is no AI involved in any of that: it is the logic above, applied to the opens you already have.

What the rules cannot do is write a good follow-up for your email, in your voice, with the context of what you originally sent. That is the second layer, and it is the part that needed something more than rules. Tomorrow's post explains the rest.

Conclusion

Three rules: not opened after three days, follow up with a new subject; opened but silent for two days, send a short nudge; three opens in 48 hours, call. One proxy for a reply, based on your own "Re:" going out, plus two buttons to correct it. And one honest limit: the rules can only see what the pixel sees, so they will sometimes nag you about a reply already waiting in your inbox. We think a transparent rule you can overrule beats a black box you have to trust.

Frequently asked questions

Can I change the thresholds, for example follow up after 5 days instead of 3?

Not in the closed beta. The three thresholds are fixed and identical for every account, which is what makes the suggestions predictable. Per-thread control is done with the "Ignore this thread" and "Already replied" buttons.

Does the extension read my inbox to know whether someone replied?

No. Reply detection is based only on emails you send: a new tracked email to the same recipient with "Re:" and the same subject marks the thread as active. The extension has no access to incoming mail.

What happens if I reply from my phone or from another mail client?

The extension only sees emails sent from the compose window it is installed in, so a reply sent elsewhere is invisible to the rules. Press "Already replied" on the thread and the suggestions stop.

Will the rules suggest a follow-up on every email I track?

Only on emails that match one of the three conditions. An email that was opened and answered within two days never triggers anything. For messages you never want chased, use "Ignore this thread".

Why not let an AI decide when to follow up?

Because the "when" depends on four or five facts that fit in a rule, and a rule can be explained, audited and corrected. We use AI for the part rules are bad at: drafting the message itself. That is the subject of the next post.

All articles