Every row of a mailing list is a person
Every row of a mailing list is a person who gave you their address to receive your news. The statistics that follow each send go further still. They name who opened, who clicked and who stopped reading you, which makes them the most revealing material in your whole toolset.
What an AI needs in order to be useful sits one floor above that. Which campaign did better than the others. How this audience grew since the start of the year. Which subject line got opened and which one fell flat. None of those answers requires knowing a single address, and the Mailchimp connector is built on that observation, like the other sources in the connector catalogue.
A campaign's totals, never a subscriber
On an email marketing source, the boundary runs between what can be counted and what can be identified. Here is where it sits, as it is written in the connector.
Individual activity is excluded for a reason that has to do with consent rather than with technique. The people on your list gave their address to receive your news, and what an AI would do with it is not part of that agreement. A count of opens answers the management questions without naming anyone.
The connector cannot reach a subscriber
Mailchimp exposes dozens of paths that lead to the individual: the members of a list, per-recipient send activity, the detail of who opened, the detail of who clicked, named unsubscribes, abuse reports, the customers of a store. None of those paths has a matching method in the connector, so none of them can be called by accident.
A second barrier sits lower still, at the wire. Before the connection opens, the client compares the path it is about to call against a short list of allowed paths, which holds campaigns, reports and audiences and nothing deeper. It also compares it against a list of refused words. A path naming members, per-email activity, open or click detail, unsubscribes or abuse reports is refused before the call.
The redundancy is deliberate. The allowed-path list would be enough on its own, and the refusal list exists so that a coding mistake becomes loud instead of going unnoticed. The same mechanism closes a less obvious door: the allowed-path list stops at the campaign itself and never descends into its sub-resources. The HTML content of the email you sent is therefore not readable, whereas its title and its subject line do come in.
One department reads your campaigns, and it has to be named
Mailchimp is a flat account. The access key sees the whole account, and there is no read permission set on a campaign. Several audiences inside one account do not designate several departments either, since they run in parallel and none of them belongs to one department rather than another.
The connector therefore applies a single filing decision to everything it reads, and that decision starts closed. Four situations leave your campaigns waiting for an administrator.
- No marketing department has been named, which is the case on the first pass
- The name given is not a valid department identifier
- The department named does not actually exist in your organisation
- Automatic filing has been switched off by an administrator
The questions we get asked about this connector
A CRM contact and a mailing-list subscriber are not the same person in terms of what they agreed to. The HubSpot connector openly reads a contact's business identity and says so on its page. This one reads no subscriber, and what does come in is reached only by the department named, following your Kastel's reading rules.
Will a connected AI be able to pull out my subscriber list?
No, and not because a setting forbids it. No method in the connector reads the members of an audience, and the path that leads there is refused before the connection opens. The file never comes into your Kastel, so there is nothing in it to extract.
What about opens, will it know who opened what?
It will know how many people opened and how many clicked, as totals and as rates. The paths that give the named detail of an open, a click or an unsubscribe are among those the client refuses to call. Your Kastel can say a campaign was opened a certain number of times, never by whom.
A subscriber unsubscribes. What becomes of them in my Kastel?
They were never in it. Their unsubscribe shows up as a count that moves on the next pass, and as nothing else, since their address and their name were never read. Erasing them is therefore not a question that arises. It arises for a campaign or an audience you delete in Mailchimp, and in that case the record is marked as gone at the source and kept until an administrator erases it.
Why do a campaign's figures change several days after the send?
Because opens keep arriving long after the emails have gone out, and Mailchimp attaches no modification date to a report. There is therefore no way to ask what changed. The connector rereads every campaign, report and audience on each pass, and rewrites a record when its figures have moved.
What this connector does not do
It does no marketing. It sends no campaign, builds no segment, changes no audience and writes nothing into your account. The connection has one verb at its disposal, reading, so the question of writing never arises.
The price of that choice is real and we will not hide it. A connected AI will not be able to tell you which prospect to call back because they opened your last campaign three times. That answer requires named activity, which never comes in. If that is what you are after, this connector is not the tool.
A campaign's title and subject line are free-text fields, so their content is whatever your team writes there. A subject line naming a client, or holding something sensitive, comes in with the campaign.
A campaign, report or audience deleted in Mailchimp is not erased from your Kastel. It now carries a gone-at-the-source note, it stays readable by the same people, and only an erasure an administrator decides removes it. If it reappears, the flag clears on its own.
See what Kastel builds with the rest of your tools.
A campaign makes sense next to the deals in flight, the client exchanges and the company's decisions. The product page shows how your Kastel assembles those sources and what each connected AI can reach in them.
See the product
