A B2B services team asked us to build cold outreach that runs from a prompt, refuses drafts its safety checks cannot approve, and stays inside every country's spam law.
Tell us what you're working on. We'll walk through what a similar build looks like for you.
Every business selling to other businesses needs a steady flow of new conversations. The tools for creating those conversations keep landing in one of two ditches. The email either reads like a chatbot wrote it, full of claims a human would never make, or the email is polished but breaks a country's anti-spam law and quietly poisons the sender reputation for months.
Personalisation at scale pulls teams toward AI that writes freely. Compliance pulls teams toward templates that never take a risk. Most tools pick one side and hope the other side does not notice. The client wanted a system that treats truth and compliance as pre-conditions, not options, so we built one for them.
The result is an outreach platform that plans campaigns from a plain-English prompt, refuses to send anything its guardrails cannot approve, respects every country's rules in code, and hands live replies to the sales team's CRM. Below is what changes when a business runs on it, and every safety layer that keeps it out of both ditches.
"Cold email lives or dies on two things. What it says being true. How it is sent being legal. Most tools nail neither."
The rule the platform runs on
This is not one more sales dashboard. It changes who is on the critical path for new conversations, whether compliance is a stack of PDFs or a switch in the admin, and whether the CRM fills with leads a human actually needs to answer.
Four steps on the team's side. Everything else is automated. Your team learns the flow once, then never has to think about it again.
Type a one-line brief. The platform finds real prospects that match, skips the ones that fail a basic sanity check, and re-tries them later. Nobody hand-builds a target list.
Every draft is written and passed through internal safety checks before it can be queued. Anything the checks cannot approve is set aside for the team to review. Nobody hand-writes each email.
Approved drafts leave the mailbox only inside the recipient's local working hours, on their working days, and only if every country and safety check passes. Nobody watches the outbox.
Real replies go to the sales CRM as leads. Bounces, out-of-office notes are filed silently. One-click unsubscribes block the whole domain permanently, and queued follow-ups auto-cancel the moment a lead engages.
Every business owner we speak to has already tried stitching ChatGPT to a mail-merge tool. The result reads better than the old templates for the first week and then starts making things up. The difference is not the model. It is the guardrails, the country layer, and the safety gates around it.
A generic tool lets the AI write what it wants and hopes the reader will not notice. This one runs every draft through internal guardrails before it can be queued. Anything the guardrails cannot approve is quarantined for the team, not sent. The output the mailbox actually sees has already survived the checks.
A general tool has no idea a business email to a UK address needs a real registered postal address, or that a message to Germany without prior consent is a legal problem. This one reads the recipient's country, checks the anti-spam law, respects the working week and public holidays, and adds the right postal address to the footer before the send goes near an outbox.
A checklist a human runs before sending is a checklist a human forgets on a bad Monday. Every email here passes an automated set of pre-send checks: country switched on, mailbox has budget for the day, local working hours, no public holiday, address not on the suppression list, mailbox not in bounce-shutoff. Any check fails, the email waits.
One bad list, one dirty import, and a mailbox can be blacklisted for months. The sender watches the bounce rate on a rolling window and shuts sending off automatically the moment the rate crosses a threshold the team sets. The mailbox stays clean, the team gets the reason, and nothing has to be noticed by a human first.
A brand-new mailbox that sends 50 cold emails on day one gets marked as suspicious the same afternoon. The platform ramps a fresh mailbox up gradually, spread across the recipient countries proportionally, and only reaches full daily volume once the reputation is warm. The team does not have to pace it by hand.
A generic tool fires and forgets. This one reads every reply, classifies it as a bounce, an out-of-office reply, an unsubscribe, a negative or an interested one, and only wakes a human for the ones that matter. The real conversation lands in the sales CRM as a lead, and the queued follow-ups for that lead cancel so nobody gets nagged after they engaged.
Most sales teams pay every month for a sales-engagement tool: HubSpot, Outreach, Apollo, Instantly, Salesloft, or one of the newer sequencers. The bill renews on autopilot, and it grows every time a seat is added or the team crosses a plan tier. A custom-built platform ends the recurring bill. The subscription stops being needed. The sales tools your team pays for shrink to the ones they actually use.
General AI writes whatever completes the sentence best. That is the whole reason it hallucinates. The platform never lets an AI draft go to an outbox on that pattern alone. Every draft passes through internal guardrails that decide whether it is safe to send. Only drafts that survive reach the queue.
Anti-spam law is not one law. It is a different law in every country the mailbox sends to. In our system every country's rules are encoded once and read by the whole pipeline, so a compliance change is a config change, not a training slide the sales team is going to forget by Wednesday.
An unsubscribe link that opens a landing page and asks a prospect to log in is the same as no unsubscribe. This one is one click, permanent, and applied to the whole domain, not just the address. The suppression list is checked at query time before any send, so an unsubscribed prospect can never be contacted again from any campaign.
A brand-new mailbox that sends 50 cold emails on day one gets marked as suspicious the same afternoon. The platform ramps a fresh mailbox up gradually and spreads the volume proportionally across the active countries. The team never has to pace it by hand.
One dirty list can put a mailbox on a blacklist for months. The platform watches the rolling bounce rate on the mailbox and shuts sending off automatically the moment the rate crosses a threshold the team sets. The reason is logged, the team can look at it, and sending only comes back on when a human turns it back on.
A great cold email that lands at 3am local looks the same as spam. Every send reads the recipient's country, computes their local time, and only leaves the mailbox during their working hours on their working days. Public holidays in the recipient's country block the send too.
Most cold-email tools fire and forget. This one reads every reply, sorts it into one of five kinds, and only wakes a human for the ones that matter. Real conversations land in the sales CRM as leads. The queued follow-ups for that lead cancel the same moment.
The platform under the friendly admin is a set of pieces. Each piece has one job, does it well, and hands off cleanly to the next. What breaks in most cold-email tools is the seam between the pieces, so the seams are where the build focused the most care.
Turns the plain-English prompt into a shape the rest of the platform can work with. Refuses countries that are switched off before anyone builds a target list.
Where every draft is produced and every safety check runs against it. Drafts that cannot pass the checks are set aside for the team, not queued.
Each country's law, working week, holidays, greeting and postal address written once and read by the whole platform. Adding a country is one entry in the settings.
Instant stop, daily send caps per country, warm-up ramp for fresh mailboxes, auto-shutoff on bounce spikes, fake-address filter, working-hour window. Any check fails, the send waits.
Reads every incoming message, sorts bounces, out-of-office replies, unsubscribes and real conversations. Only real ones wake a human, and they land in your CRM.
Decides what work is due, holds a lock so nothing runs twice, and keeps every heavy job under a memory ceiling. Every other piece does its slice when the scheduler wakes it.
The build is not a black box. Every job the platform does is visible in the admin, editable by the team, and safe to open at any time. Below is what the team logs into.
The team writes prompts. The system does the rest. Each step is small and clear, and any layer can be changed without rewriting the ones on either side. Your data, your mailbox and your CRM stay in your perimeter.
The platform runs on tools the business already owns. Nothing about it asks a team to abandon a mailbox, migrate a CRM, or pay for a proprietary data cloud. Below are the inputs, layers and destinations.
Your team sees "type a prompt, walk away, check the CRM later". Underneath, six small steps make the output land right.
The current platform runs cold email end to end. A real build starts with the same shape, then adds the channels, safety settings and destinations the business actually uses.
The current platform. One-prompt campaigns, 13 country regimes handled in code, live replies into the CRM. Nothing that fails the guardrails ever reaches a mailbox.
Bounces, out-of-office notes, unsubscribes and real conversations are routed automatically. Only real conversations wake a human.
The mailbox reputation is protected in code, not by a manual watcher. Sending shuts off automatically when the rolling bounce rate crosses the threshold.
Same guardrail-gated drafter, applied to LinkedIn messages, so a team can reach the same prospect on the channel where they respond faster.
In markets where WhatsApp Business is the primary channel, the same pipeline can drop a message through the Business API. Same drafter, same guardrails, different transport.
Today one owner runs a campaign end to end. Next version lets a small team co-own campaigns, split ownership by geography or target audience, and see each other's queue without stepping on sends.
You Cannot Sell a 4,600-Acre Ranch the Same Way You List a Suburban Home
355 Million Women Managing Their Periods with Zero Reliable Guidance
Your Model Is a Commodity. Your Data Is the Moat.
Selling Dumbphones to a Generation Addicted to Smartphones
We built this for A B2B Services Team. We can build it for you, same rigour, your domain.
No spam. Just a conversation about your project.
Hope that was helpful. Reach out anytime.