A prompt that finds the threads you promised to follow up on and never did, then writes and sends a real sequence in your voice — scheduled by day, stopped the moment they answer, and logged so you always know which deal is actually moving.
Almost nobody loses a deal because they decided not to follow up. They lose it because the follow-up needed a quiet afternoon, and the afternoon never came.
Writing the third email in a thread is small work and large friction: it needs the thread re-read, the last question recalled, and a sentence that does not repeat the first two. It survives on the bottom of a list that never gets to the bottom. The deal does not stall because you decided against it; it stalls because the write-up cost more than the fifteen seconds it takes to decide not to.
By the time you notice a thread has gone quiet, three weeks have passed and the context has gone with it. Which ones already got two nudges? Which one is a customer versus a prospect? Which one did you promise something specific to? Reconstructing that from memory across two hundred threads is what turns follow-up into archaeology, and archaeology is what gets skipped.
The same fourth-touch email, sent identically to a warm champion and a stranger who never replied, reads as automation because it is. Volume does not fix tone — it amplifies it, and the threads that would have answered a human get dropped instead. Most sequences fail on the message, not the timing, and no scheduling tool fixes the message.
Copy it once, paste it into Zaira, and it takes over the chase: find the stalled threads, check each one for a reply, draft the next touch in the right tone for that stage, send it, and record what went out so the next run knows where it left off.
Three stages, one goal. The agent runs the sequence on a schedule, so the only thing left for you is reading what came back — because it stops the moment a human replies.
Because the sequence is scheduled rather than remembered, the person who went quiet in week two gets their third touch in week three whether or not week two was a good one. Consistency across a quarter is what most people actually fail at, not the writing.
Every run re-checks the threads it has touched, and any reply pulls that thread out of the sequence immediately. Nothing gets nudged while a real conversation is in progress — which is the single thing that makes follow-up automation feel safe to switch on.
Threads that never answered after two or more touches get one clean goodbye and then stop. The pipeline report stops counting them as open, and the list you actually look at is the list that is still alive.
The practical questions people ask before putting an agent in charge of their outbound.
Yes — Step 4 sends the drafts it wrote, because that is the whole point of a follow-up sequence, and this prompt deliberately does not ask you to approve each one. Read it before you run it, because Step 3's judgment about which threads are worth chasing is the part that matters. The safety comes from the structure: a stop-list in your context, a hard cap of touches, and the fact that any reply ends the sequence. If you want approval first, run Step 3 alone for a week and send from Gmail yourself.
Four touches over about three weeks is the default most people settle on, and the reason is not etiquette so much as signal: after the fourth unanswered email you have learned what you can, and continuing mostly trains people to ignore your domain. The prompt encodes your cap in Step 2 and then closes the loop rather than trailing off, so the number that matters is the one you actually set rather than the one you feel like on day nineteen.
Those tools start from a list you bought and treat every contact as a fresh row. This starts from conversations you were already in, so every message is written against the actual history — what they asked, what you already sent, what went quiet. That is why it can be as conversational as it wants to be, and why it works on warm and mid-funnel threads that a sequence tool would only ever treat as a lead ID.
Write the tone per stage as a job, not an adjective — 'touch 2 is a short check-in with one new piece of information' rather than 'friendly'. The agent is capable of plenty of register, and vague adjectives make it pick the same voice every time, which is what makes automation recognisable to a recipient. Two or three concrete sentences about what each touch should accomplish is more useful than a paragraph about your personality.
This matters, and it is worth being precise about. Zaira works through Gmail's own official MCP server — the permissioned connection offered to approved AI partners, not a scraper or a browser automation rig. There is no shadow account and no spoofed client; sending happens inside the access your mailbox already has, and you can revoke it from your Google account at any time. Sending volume also stays inside the cadence a person could plausibly write, which is the behavior that keeps an account healthy.
It reports them, and stops touching those threads. Step 5 surfaces what came back and what it suggests you do about it, then removes the thread from the sequence so nothing automated follows a human conversation. Sending a considered reply to a specific objection is the part worth doing yourself, and the sequence is at its most useful precisely because it gets that thread out of the queue and onto your desk.