The Amazon Job

There is no clean database of French DNVBs. There is a business register with every company in the country in it, and there is the open web. So the first pass is a wide net: everything in the right revenue band and the right activity codes out of the register, plus a web sweep for brands that look the part — own storefront, own product, sold direct.

Wide is deliberate. Anything filtered out here is invisible for the rest of the campaign, and a brand missing from the first pull can never be recovered by a smarter step six. Precision comes later and it is cheap later. Coverage is only available now.

The register hands over an entity: a name, a number, an address. It does not hand over the storefront, and half the time the legal name has nothing to do with the brand on the packaging. So each candidate gets resolved — find the real website, confirm it is live, confirm the company behind it is the one the register named.

Then the question that decides everything downstream: is there an Amazon presence, yes or no. Not inferred from the brand's marketing, proven — the listings, the seller behind them. This single fact is what splits the campaign in two a few steps later, so it is worth being slow about.

The campaign is not aimed at French e-commerce. It is aimed at brands built the DNVB way: Shopify for the store, Klaviyo for the retention, Meta for the acquisition. That trio is a proxy for a whole way of operating — one that leans on paid social for nearly every order and feels it in the margin every month.

A stack detection pass reads the site and confirms it. A brand on a custom stack, or one that was marketplace-first from day one, is not a bad company. It is just not this campaign, and it drops out here rather than absorbing research budget for three more steps.

Back to the register, this time for the details that decide fit and reachability: legal identity, headcount, filing history, and the list of company directors.

Headcount is the real filter. One to fifteen people means the founder still runs growth and still reads their own inbox. Past that there is a marketing team, the founder delegates the channel, and a founder-to-founder email lands with someone who never asked for it.

The director list matters for a different reason: it is the only trustworthy answer to who actually runs this company. LinkedIn will cheerfully offer three candidates. The register names one.

Everything so far was qualification. This is segmentation, and it is where the recipe earns its keep. Profile A already sells on Amazon and does it badly — thin listings, PPC that burns more than it brings back. Profile B is not there at all: a strong Shopify brand paying Meta for every single order.

Same size, same stack, completely different conversation. A gets shown the money already on the table and left there. B gets shown a channel they don't have. Send the wrong one and the email isn't merely ignored, it is wrong in a way the founder notices.

Ambiguous cases don't get a coin flip. They go back for a second and a third look, and if they are still unclear they are dropped. An unclear segment can only produce an unclear email.

An explicit rejection gate. Everything off-target — wrong size, wrong model, wrong country, a retailer wearing a brand's website — drops out here, before it costs anything downstream.

The reason this is its own step and not a filter buried in the last one: the rejections are the part worth reading. If a whole category is being cut for the same reason every time, the sourcing query was wrong, not the companies. That's a fix at step one, and you only see it if the discards are kept somewhere you can look at them.

A final pass, run per profile, asking whether the fit actually holds. A brand that crossed €300k once on the back of one good Black Friday is not a €300k brand. A company that looks like a DNVB but makes its living from wholesale is not the target, however good the website is.

This is the most expensive check per company, and it runs on the smallest list. That is not an accident, it is the whole reason for the order of the steps: run it first and it costs fifty times more to answer the same question.

Only now does a human being get attached to a company. LinkedIn search turns up the candidates, an AI judge picks the profile whose title and tenure genuinely mean founder, and the register's list of directors is the referee — when the two disagree, the register wins.

No heads of growth, no marketing managers, no info@ addresses. The offer is a founder-level decision about a sales channel, so a founder-level inbox is the only place it makes sense. A campaign that quietly loosens this rule to hit volume is a different campaign, with worse numbers.