Three hundred invitations, thirty-four replies, eleven requests for a sample, nine parcels, four videos. That sequence comes from a two-person team selling home goods across Indonesia and Thailand, and the ratio is not unusual in creator outreach. What is unusual is that they knew where the other twenty-three replies stopped — most teams can tell you how many invitations went out and how many videos came back, and nothing at all about the middle.
Creator outreach gets described as a sequence of actions. In practice it behaves like a sequence of leaks, and each leak has a different cause. Sending more invitations to fix a leak in the sampling stage just spends money to find the same leak again.
This is written for the seller running ten to thirty SKUs with one or two people carrying the whole programme, in one or two Southeast Asian markets where the reply arrives in Bahasa or Thai and usually at night. At that size you cannot run the chain from memory, and you cannot afford a tool that covers one stage and pretends the rest does not exist.
The list decides the reply rate before you write a word
The first leak sits upstream of everything and only shows up two weeks later as a bad reply rate. A list built from a public creator directory contains two kinds of people that look identical in a spreadsheet: those who have sold something in your category, and those who have never sold anything at all. Both have a follower count. Only one of them answers.
Two sources consistently beat a generic directory. The first is your own history — creators who have already posted for you, tagged in a way that still means something six months later rather than a colour-coding system only one person understands. The second is the competitor shelf: the creators visibly promoting the product that outsells yours. That second source is why a shortlist built from rival shops converts better than one built from a search box — you are copying proven demand instead of guessing at it.
What software cannot do here is decide whether your category belongs with creators at all, or which kind. A seller of kitchen tools and a seller of industrial components face different maths, and no filter makes the second one work. That judgement happens before the list, and it is the one part of creator outreach nobody can automate for you.
Vetting before samples is cheaper than any recovery step
Every programme has a moment where a parcel gets sent to someone who was never going to post. That parcel costs freight, costs stock, and costs the two follow-ups that come after it. The invoice is small per unit and enormous per quarter, which is why the vetting question is worth more attention than the invitation copy everyone rewrites.
The useful test is not follower count. It is whether the creator’s last ten posts look like your product could sit in them without an apology — same lighting, same category adjacency, same audience expectation. A creator at five thousand followers who posts twice a week in your exact category beats one at two hundred thousand who posts unboxings of unrelated gadgets, and the gap is not small. A team I work with stopped sending to anyone above fifty thousand followers in one market and their sample-to-video rate went up, not down.
Software carries the numbers into one row so a person can make that call quickly. It does not make the call. Any vendor promising “AI-selected creators” is selling you the first half and hiding the second.
The invitation window decides more than the wording
Teams rewrite invitation copy constantly and leave the window at whatever the default was. That order is backwards. A creator deciding about next month’s content is not reading your message on the day it lands; she reads it when her calendar opens, and if the invitation has already expired the best copy in the world meets a dead link.
The practical rule: set the window to reach past the month you expect her to film, not the month you expect her to answer. It feels excessive until the first time a creator says yes in week six to an invitation sent in week one — which happens often enough to be the difference between a list that fills a quarter and a list you rebuild every six weeks.
Volume has a second-order cost nobody quotes in advance. Bulk invitation at scale degrades quality not because the message gets worse but because the filtering gets skipped when the list is large. A programme that can send four hundred invitations in an afternoon will send them to people it would have rejected if it had forty to send. DAMI caps a group at a hundred creators and spaces the sends a few seconds apart, which is not a throughput limit so much as a speed bump that keeps the list being read.
The ones who did not reply are the actual workload
In the creator outreach sequence at the top, thirty-four of three hundred replied. The twenty-six who replied and then stopped, plus the eleven who asked for a sample, plus the nine who got one — that is where the week goes. A programme with no second touch is running at roughly half its list size, because most replies in Southeast Asian markets land after the follow-up rather than after the first message.
| Stage | Count | Where the rest went | Who can close it |
|---|---|---|---|
| Invitations sent | 300 | — | — |
| Replies | 34 | 266 silent: wrong category, expired window, or a message that read like a broadcast | The list and the window, before sending |
| Sample requests | 11 | 23 stopped after replying, mostly because nobody followed up inside a week | The follow-up loop |
| Parcels shipped | 9 | 2 requests never shipped, held behind a per-request approval decision | An approval rule set by tier |
| Videos posted | 4 | 5 went quiet after delivery, usually a question nobody answered | The sample-state messages |
Three channels matter and they do different jobs. The in-platform message is where the conversation lives and where the record belongs. Email reaches the creator who checks it once a day and the one who has notifications off. WhatsApp gets read in minutes, which is why it wins on speed and why it can also get a personal number restricted if the volume grows past what a person should be sending. Using all three without one place to see them is how the same creator gets messaged twice by two people on the same team — keeping creator communication readable across all three is the part that decides whether the channels reduce the work or just multiply it.
The judgement that stays human is reading the reply. “Send me the details” and “maybe later” look similar in a chat list and mean opposite things — one is a sample request, the other is a polite no that will cost you a parcel if you read it wrong. That call needs someone who has seen a hundred of them.
Samples: four states, and the silence between them
Shipped, delivered, opened, posted. Four states, and a creator who goes quiet at any of them either asks a question at two in the morning her time or quietly moves the product down her list. The messages that prevent both are short and entirely predictable, which is exactly why they are the first thing to automate and the first thing to get skipped when the week is busy.
Approval is the other half. Approving every request protects the relationship and gives away stock; approving none protects stock and loses the creators who were about to say yes. The workable middle is a rule set by tier and by history rather than a decision made per request at eleven at night — the per-request version does not survive a hundred requests a month. DAMI applies that rule before the request reaches a person, so what lands in the queue is the exceptions rather than the routine.
Creator outreach does not survive on enthusiasm. It survives on a small number of messages arriving on time, and the parcels are where timing is most visible to the creator.
After the video: the leak that shows up next quarter
A campaign that ends with four videos and no record of why the other nine parcels produced nothing is a campaign you will run again from scratch in three months. The useful output of a first round is not the videos — it is a tagged list of who posted, who went quiet after delivery, and who never had a chance because the parcel arrived after their shoot week.
That record is what makes the second round cheaper. It is also what most teams never build, because the campaign is over and everyone has moved to the next thing. The sellers who compound are the ones who treat the post-mortem as part of the campaign rather than as an optional extra.
This is the whole chain, and it is why a map of where software helps and where it does not is worth reading before choosing anything: creator outreach is only one of five areas, and three of its leaks are judgement problems rather than tooling problems. DAMI covers the mechanical stretch — the invitation window, the follow-up loop, the sample states, the tagged list — and it deliberately does not pretend to cover the two stages that need a person looking at videos.
FAQ
How many invitations should a two-person team send a week?
Enough that every reply gets answered within the same working day. For most teams that lands between eighty and two hundred, and the number is set by the reply side rather than the send side. Sending four hundred a week while replies sit for two days is not a bigger programme, it is a slower one.
What is a realistic reply rate for cold creator outreach?
Between eight and fifteen percent on a list built from category-relevant creators, and under five percent on a list built from a public directory with no filtering. If you are below five, the problem is the list. If you are between five and eight, the problem is usually the invitation window. Rarely is it the wording, which is the part everyone rewrites first.
When does creator outreach stop being a one-person job?
When a single missed follow-up costs more than an hour of someone’s week — in practice somewhere around a hundred active creators. Before that, adding a second person mostly adds a second interpretation of the same list. Creating an account is the cheapest way to find out where that line actually sits for your volume, and the DAMI outreach workspace is where the invitation, the parcel and the tagged list live in one place instead of three.
