Privacy policy
Last updated: [DATE].
[COMPANY NAME], [ADDRESS], registered under [REGISTRATION NUMBER] ("we"), runs this LinkedIn outreach service. For your own account data we are the controller. For the data about your leads and their messages, you are the controller and we process it on your instructions, under the data processing terms in our terms of service.
What we store
- Your account: email address, a hashed password (never the password itself), company name, your timezone, when you accepted our terms, and the time of your last login.
- Your business profile: the answers you give in the onboarding and on the Business profile screen (your company, your offer and prices, results and client types, your ideal customer and their problems, objections with your answers, your tone and wording, your call to action and booking link, and what must never be said), and the context file our server makes from them. Our operator can read it in our operator console. It is kept while your workspace exists. Do not put personal data about other people or anything secret in it.
- Your LinkedIn connection: the identifiers our connection provider gives us for your account, your LinkedIn public profile name and handle, whether the account has Premium, its connection status and your sending limits. We never store your LinkedIn password or session cookie. If you reconnect with a different LinkedIn profile, we keep the safety state of the earlier profile (its LinkedIn member identifier and handle, a pause after a restriction with its reason, and the start of its warm-up period) so that the limits follow the profile. It is deleted once it no longer changes a limit, and with your workspace at the latest.
- Your leads: what you import or add (LinkedIn profile URL, name, company and title), and what our connection provider returns when the service looks up their profile: their LinkedIn member identifier and handle, their name if you left it empty, and how closely you are connected to them on LinkedIn. Also each lead's status in your campaigns.
- Campaigns and conversations: your message templates, which steps went to which lead and when, and the messages exchanged with your leads, including their replies and your drafts. Also the notes you write on a conversation (never sent) and the outcome you set for it, with your reason. A message that was sent automatically carries a mark that says so, and a response prepared from your reply rules carries the kind of reply it answers.
- Reply rules: per campaign, the mode you chose (off, suggest a response, or send automatically), whether you let the AI model sort replies, the responses you wrote for each kind of reply, and how a response is picked for each kind (the first, in turn, or by the AI model). They are kept with the campaign.
- Reply decisions: for each reply of a lead in a campaign that has reply rules, one record: the kind of reply, a confidence number, whether the AI model was used, which of your responses was picked (its number and a code that shows whether you changed it since), when it was due, and what happened to it (nothing prepared, kept as a draft, sent, failed or cancelled, with a short note of the reason). No message text. Deleted 180 days after the decision was closed. The inbox shows you this record in words: above a lead's reply the kind it was sorted as and whether our rules or the AI model sorted it, above a draft how the response was picked and, where it was meant to be sent automatically, why it was not, and above every message that was sent automatically the label "Sent automatically".
- Assistant chat: what you type to the assistant and what it answers is stored in your workspace, per user, at most 500 entries per workspace (the oldest are deleted first), each for 90 days. The answers you gave in a set-up that is not finished, including step texts the AI model reworded and you have not confirmed yet, are stored with it; they are no longer used after 30 minutes without a message and are deleted after 90 days at the latest. You can delete your chat at any time with "Clear chat" on the Assistant screen. That deletes your entries and an unfinished set-up at once; campaigns and responses that were already stored stay as they are. The chat is kept on our server only, not in your browser.
- AI usage counts: how many requests your workspace made to the AI model per day and per kind of request (sorting, rewording, draft): a count, never text. Kept for 60 days.
- Sorting cache: when the AI model sorted a reply, we keep the kind and the number it answered, filed under a scrambled code made from the masked text, for 30 days. The code is made with a secret key of our server and cannot be turned back into text. We do not store what was sent to the AI model. Of its answers we keep only this kind and number, a draft you asked for, and reworded step texts that were shown to you.
- Messages that are not with your leads: our connection provider tells us about every new message in your connected LinkedIn inbox. When a message is not with one of your leads, we remove its text and sender as soon as we have processed the notification, and keep only a technical note (that a message arrived and why it was ignored). That is normally within seconds for a message you receive, and within about two minutes for a message you send yourself on LinkedIn (we wait a minute to see whether it is one of ours). A notification for a LinkedIn account we do not know yet waits up to about five minutes.
- New connections: our connection provider also tells us when someone becomes a connection of your LinkedIn account, also when that person is not one of your leads. We remove the name once the notification is processed; the LinkedIn member identifier and handle stay in the notification until it is deleted after 7 days. To see which invitations were accepted, the service also reads your list of connections at the provider and stores nothing about people who are not your leads.
- Notifications from our connection provider: the text of a message is removed from them once they are processed. They are deleted 7 days after we received them. A notification that cannot be processed is deleted with its content after 30 days.
- Activity records: a record of each LinkedIn action (its type, its time and the LinkedIn profile it was sent from, no content), used for the limits and kept for 60 days.
- Login cookie: one cookie that keeps you signed in for up to 30 days. No tracking cookies.
- Login attempts: to limit attempts to log in, sign up or reset a password, we count them per IP address (for a login also per email address with IP address) in memory, for up to one hour. This is not written to the database.
- Technical logs: record identifiers only, never message text, names or email addresses, kept by our hosting provider for [LOG RETENTION]. Alerts to our own team carry the kind of event, a workspace identifier and a short note of the reason, never names or message text.
Why we use it
- To run the service you signed up for: your account, your LinkedIn connection, your leads, campaigns, reply rules, conversations, the assistant chat and your business profile (performance of our contract with you; for lead data, on your instructions).
- To keep your LinkedIn account and the service safe: activity records, reply decisions, limits, AI usage counts and technical logs (our legitimate interest in account safety and security).
- To use the AI model, only in the three cases described below and only where you asked for it or chose it.
Where it is stored and who processes it
- Hosting and database: a hosting provider based in the United States, with our servers and database in its Amsterdam region (Netherlands).
- LinkedIn connection provider: a company in France, hosted in France. You log in to LinkedIn on its page, and it keeps an encrypted LinkedIn session token for your account until you disconnect. If the provider cannot be reached at that moment, we keep retrying until the connection is removed there. It sends your LinkedIn actions and receives your LinkedIn messages. It routes this traffic through proxy providers located in Lithuania, Israel, the United States and Singapore. The proxy is usually located in a country near you, but can be elsewhere.
- AI provider: one provider, based in the United States, only in the cases described under "What goes to the AI provider" below.
- Alerts to our own team: an alert service, if we use one. It receives the kind of event, a workspace identifier and a short note of the reason.
- Names: we give customers the current list of these providers by name on request to info@clawbuildr.com.
- Transfers outside the EU: where a provider above is located outside the EU, or reaches the data from there, the transfer relies on [TRANSFER BASIS].
Our pages load no fonts or other files from third parties. We do not sell personal data and do not use it for advertising.
What goes to the AI provider
The AI model is off unless we switch the model on for the service. Reply rules work without it: replies are then sorted on our own server, by word lists in English, Dutch and German. The AI model runs at an outside provider, listed under "Where it is stored and who processes it". It is the same provider for every use below: a reply draft, sorting replies, picking between your responses, and rewording step texts in the assistant. We use no other AI provider. When the model is on, text goes to the AI provider in three cases and in no other:
- Draft a reply: only when you click "Draft a reply" in a conversation. Sent: the last 20 sent or received messages of that conversation and your business context (below).
- Sorting replies: only for a campaign where you chose it, by letting the AI model sort the replies our own rules are unsure about, or by letting it pick between several responses of one kind (you set that per kind of reply). Sent: what the lead wrote since your last sent message, your prepared responses (those of one kind when it only picks), and your business context (below). The AI model answers with a kind and a number only.
- Rewording step texts: only when you choose, in the assistant, "Create and let the AI model reword the texts" for a new campaign. Sent: your description of the audience and the offer, the tone and the language, the step texts and your business context (below). Nothing else you type in the assistant chat is sent to the AI model. The reworded texts are shown to you beside the originals, and you choose which of the two the campaign gets. Nothing is stored in a campaign before you choose.
Your business context. Each of these requests also carries the context file our server makes from your business profile, at most 6,000 characters, and only the one of your own workspace. Before it is sent, email addresses and phone numbers in it are replaced with placeholders, and the last names of people are removed: your own name goes as your first name only, except where it is part of your company name, which is sent as you wrote it, and another person's name keeps its first name where a greeting, a title, a sign-off, a role such as "our CEO" or a phrase such as "talk to" shows that it is a name. Everything else in it is sent as you wrote it: your company name, your offer, prices, results, client types, industries, regions, your website and booking links, your tone and wording. This is pattern matching too and can miss a last name or a phone number (see the list below). The Business profile screen in the app shows the file as "What the AI is given". Do not put personal data about other people or anything secret in your business profile.
Apart from your business context, the text passes through automated masking on our server before it is sent. It replaces with placeholders: the name and company of the lead and your own name and company as we hold them, your booking link, email addresses, phone numbers, links, handles, long numbers and codes that look like identifiers, names after a greeting, after a title, in a sign-off or after a phrase such as "my name is", and company names followed by a company type such as B.V., GmbH, Ltd or Inc. We send no workspace, lead, account, conversation or campaign identifier and no profile link. When a draft or a reworded text comes back, we put the original values back on our own server.
The masking is pattern matching and can miss a name or a company. When it does, that name or company is sent as it was written, so personal data can still reach the provider. The kinds of case it misses:
- a single name as the first word of a sentence;
- a name or a company typed without a capital letter and without a greeting, a title, a sign-off or a company type next to it;
- in German, a name or a company in the middle of a sentence without a greeting, a title, a sign-off or a company type next to it, because German writes every noun with a capital letter;
- a street address: a house number is not masked, and neither is a street name at the start of a sentence or typed without a capital letter;
- your own name while no LinkedIn account is connected;
- a spelling it has no pattern for;
- in your business context: a last name with none of the signs named above, names listed together after one greeting or one phrase such as "talk to" (joined by "and" or "or"), a name typed without a capital letter, a phone number inside a link, unless it is a WhatsApp link or a "phone=" value, and a phone number written like an amount (with dots or spaces between groups of three digits), as a date or next to a currency sign.
Whatever else a message says about a person (their role, their plans, their situation) is not masked.
What the AI model can and cannot do:
- Under reply rules, the text a lead receives is always one of your own responses, with its variables (such as the first name) filled in. The AI model only chooses among them.
- A reply whose kind the AI model chose is never answered without your review: your response is kept as a draft for you to approve.
- A reply sorted as a question is never answered without your review either, whoever sorted it and in every mode: your response is kept as a draft for you to approve.
- When our own rules recognise the kind and you let the AI model pick between several responses of that kind, the response it picked can be sent as an automatic reply without your review (see "Automatic replies").
- AI drafting never sends a message by itself: a draft goes out only when you click "Approve and send".
Automatic replies
Automatic replies exist. They are off unless we switch them on for the service and you switch them on per campaign. Reply rules are off for a campaign until you choose "Suggest" or "Send automatically" in the campaign's "Responses to replies"; while they are off, replies are not sorted and no response is prepared. On "Suggest", a matching response is kept as a draft for you to review. "Send automatically" cannot be chosen while we have not switched automatic replies on; a campaign that was set to it earlier then keeps its matching responses as drafts. When you choose it, the app asks you to confirm.
When both are on and our own rules (not the AI model) recognise the kind of a lead's reply, and that kind is not a question, the response you wrote for that kind is sent from your LinkedIn account without your review, 4 to 12 minutes after the reply at the earliest, within your working hours and daily limits, and at most twice per conversation. Automatic replies only ever send text you wrote. While an automatic reply waits, the inbox shows about when it goes out, and you can cancel it there. Once it is sent, the message carries the label "Sent automatically" in the inbox.
If you let the AI model pick between several responses of that kind, the lead's reply goes to the AI provider (masked as described above) and the response the AI model picked is the one sent, without your review. The kind itself is still the one our own rules recognised.
- A question is never answered automatically, also when the campaign sends automatically. Your response to it is always a draft for you to approve.
- A lead who asks not to be contacted again never gets a response from reply rules, neither sent nor drafted.
- A reply that also shows a sign of a decline or a negation is not answered automatically with a response of another kind; your response is kept as a draft.
- A response that could not be sent within 18 hours of the lead's reply is kept as a draft and not sent later.
- When you answer in the app while an automatic reply is still waiting, it is cancelled. An answer you type on LinkedIn itself cancels it only once LinkedIn has told us about your message.
How long we keep it
We keep your data while your workspace exists, except where a shorter period is stated above. The shorter periods are: reply decisions 180 days after they were closed, the assistant chat 90 days, AI usage counts and activity records 60 days, the sorting cache 30 days, provider notifications 7 days after we received them (30 days when they could not be processed).
We delete on request. Write to info@clawbuildr.com from the email address of your account. Within [DELETION PERIOD] we remove your LinkedIn connection at our provider and delete your workspace with your account, business profile, LinkedIn connection record, leads, campaigns, reply rules, conversations, reply decisions, assistant chat, AI usage counts, sorting cache, activity records and the provider notifications that belong to it. One exception: if the same LinkedIn profile is also connected in another workspace, the activity records of the last 8 days stay, without any link to your workspace, so that profile's daily and weekly limits keep counting; they are deleted after 8 days. Copies in database backups are overwritten within [BACKUP RETENTION]. You can delete a single lead yourself in the app (Leads), and disconnect LinkedIn at any time.
Your rights
You can ask us for an export at any time: "Export data" in the app opens an email to us with the request, and we send the file to the email address of your account within 30 days. The file holds your leads, campaigns with their steps, and conversations with their messages, your notes and the outcome you set, plus the assistant chat of your workspace (who wrote each entry, its text and its time) and your business profile with the context file made from it, in a machine-readable format (JSON). The Business profile screen also downloads the context file alone, as context.md. The file does not hold your reply rules and responses, the reply decisions, the AI usage counts or the sorting cache. For a copy of those or of anything else we hold about you, such as your account details or the record of which step went to which lead and when, or to correct or delete your data, restrict or object to its processing, write to info@clawbuildr.com. If your leads contact you about their data, you answer them as controller; we help you do that. You can complain to your data protection authority.
Contact
info@clawbuildr.com