There is a version of outbound where you describe what you could do, and there is a version where you have already done a piece of it and are handing it over. The second one is not a clever variation on the first. It is a different transaction, and it changes what the merchant is deciding.
In the first version they are evaluating a claim, which requires trust they do not have. In the second they are evaluating an object, which requires only that they look at it. Everything in this article follows from that distinction, including the parts where artifact-first outreach gets expensive and stupid if you are not careful.
The value equation, and where cold outreach breaks it
The standard framing of an offer is that perceived value has to exceed the sum of price, effort and risk. Most people building offers spend all their energy on the first term — bigger promise, better proof, more features. In cold outreach that is exactly the wrong place to push, because a stranger discounts your promise to near zero no matter how large you make it.
The tractable terms in cold outreach are effort and risk. Effort is what the merchant has to do before they see anything: fill in a form, book a call, sit through a discovery. Risk is the probability that the whole thing is a waste of their afternoon. An artifact drives both toward zero. There is nothing to do but open it, and the risk is bounded because the thing already exists and cost them nothing.
This is also why artifact-first works better in ecommerce than in most B2B categories. A store is a public object. You can inspect the catalog, the stack, the traffic history and the local profile without asking permission. In categories where the relevant information is behind a login, you cannot build anything meaningful before the first conversation. Here you can.
What counts as an artifact
An artifact is a finished thing addressed to one store, which would have been billable if they had asked for it. Not a template with their logo dropped in. Not a fourteen-page automated report where twelve pages are boilerplate. Something a competent person would recognize as work.
Concretely, for stores: a list of the thirty keywords currently sitting in positions four through twelve with the page each one lands on and what is wrong with it. A rewritten set of product titles and meta descriptions for their twenty highest-traffic SKUs. A teardown of their checkout on mobile with the four friction points timestamped. A one-page reading of their organic trend showing the month the decline started and the three most likely causes. A Google Business audit for the stores that have a matched profile: category accuracy, photo coverage, the unanswered reviews, the claim status.
The common property is that each one is generated from data you already have plus fifteen to thirty minutes of judgement. Not two hours. If the artifact takes two hours, the model does not survive contact with a three-hundred-store list, and you will quietly stop making them around store number nine.
It has to contain something they cannot see themselves
The failure mode of free audits is that they tell the merchant things they already know. “Your site is slow on mobile.” “You should collect more email addresses.” “Your product pages could have more reviews.” The merchant has heard all of it from the last six agencies, and your artifact joins the pile.
What earns a reply is a fact they do not have access to. Cross-store comparison is the cheapest source of this: how their product count, traffic, keyword footprint or review velocity sits against the distribution of comparable stores on the same platform in the same country. A merchant knows their own numbers. Almost none of them know where those numbers sit relative to two hundred peers, because nobody has ever been able to tell them.
The other source is history. A trend series of roughly a hundred and seventy monthly buckets means you can say “this started in a specific month” rather than “traffic is down”. Attaching a decline to a date is what turns a vague worry into an investigable problem, and merchants respond to investigable problems because they can act on them.
Unit economics: what an artifact is allowed to cost
Do this arithmetic before you build anything. Take your average first deal value and your realistic conversion from artifact sent to client signed. If a deal is worth €4,000 and one in forty artifacts converts, each artifact can absorb about €100 of cost before the channel stops making sense — and you want a margin, so target a fraction of that.
At fifteen minutes of your time plus a few API calls, an artifact costs you maybe €20 to €30 fully loaded. That works. At two hours of a senior person’s time it costs €150 and the channel is underwater unless your deal size is much larger. This is why the generation step has to be mostly automated and the judgement step has to be short and human.
The practical build: a script or an agent pulls the store record, runs the analysis, and produces a draft. You spend fifteen minutes correcting the draft, deleting the parts that are obviously wrong, and adding the one observation the automation could not make. That last fifteen minutes is what makes it an artifact rather than a report. Skip it and you have built a slightly more elaborate spam machine.
Name the competitor
“We work with brands similar to yours” is a dead sentence. It has been used by everyone, it commits to nothing, and the merchant reads it as an admission that you have no case study worth naming.
Naming a specific competitor changes the temperature entirely. “Your three closest competitors on organic in France are these, here is where each of them is beating you and on which queries” is uncomfortable to read and impossible to ignore. This is the one place in outbound where discomfort is the goal — not manufactured urgency, but the genuine and correct feeling that someone else is capturing demand you thought was yours.
Two constraints on this. The comparison must be real and checkable, because a merchant who knows their market will catch an invented competitor in four seconds. And keep it factual rather than gloating — you are showing them a gap, not telling them they are bad at their job. The line between “here is a problem” and “here is why you are incompetent” is thinner in writing than you think.
Do not put the artifact in email one
This is counterintuitive and it is the mistake almost everyone makes. Having built something good, the instinct is to lead with it. But an unsolicited attachment or link in a first message from a stranger is a deliverability liability and a trust liability at once, and the merchant has no reason yet to spend attention on it.
Sequence it instead. Email one states the observation and asks whether it is a problem worth looking at — a question they can answer in four words. Email two, if there is no answer, mentions in passing that the analysis is already done. Email three hands it over. By the third message the merchant has seen your name three times, has an observation they could not immediately dismiss, and knows the thing exists. Now the artifact is being opened by someone who has already engaged with it mentally.
This also protects the economics. You do not build the artifact for all three hundred stores. You build it for the eighty who reacted at all to the observation. The sequence is a funnel and the expensive step belongs near the bottom of it, not the top.
Risk reversal after the artifact lands
Once the artifact has been received and acknowledged, the conversation moves to price, and price conversations are where risk reappears. The merchant has now conceded that the problem is real. What they are unsure about is whether you can fix it and whether they will get their money back.
Handle the two standard objections structurally rather than rhetorically. Against “it is expensive”, split the payment or reduce the scope to a first phase with a defined deliverable — you are not discounting, you are lowering the size of the first commitment. Against “I am not sure it will work”, offer a bounded trial, a performance component, or a defined guarantee. And when you cannot tell which objection you are facing, ask directly whether the hesitation is about interest or about budget. The two have completely different answers and guessing wastes the call.
One heuristic worth adopting: never end a call without the next one scheduled. “I will think about it and get back to you” is not an outcome, it is an ending dressed as a continuation.
The failure mode: giving so much you cannot charge
Artifact-first has a real downside and it deserves a paragraph rather than a footnote. If the artifact is complete enough to act on, some merchants will thank you, implement it themselves, and never reply again. That is not a betrayal, it is a design error on your part.
The fix is to make the artifact diagnostic rather than prescriptive. Show what is wrong, quantify it, and be specific about the gap — but stop short of the implementation. A list of thirty near-miss keywords with the pages they land on is diagnostic; the rewritten pages are the engagement. A checkout teardown identifying four friction points is diagnostic; the fixed checkout is the engagement. The merchant should finish reading convinced the problem is real and clear that solving it is a project.
And qualify before you build. If the store cannot plausibly afford your engagement — and the floor test from your filter stack should already have caught this — then an artifact is thirty minutes donated to someone who was never going to be a client. The artifact is the most expensive thing in your funnel. Spend it on the stores that passed the filter, and on nobody else.