Solutions / Growth
A researched list of prospects every week.
An agent that finds businesses matching your ideal customer, researches each one, and drafts a first message worth reading for you to approve.
The loop
How it works
From your rules to a researched weekly list
Seven steps, from the first event to the result. Choose a step to read what happens in plain words, how it is built, and what keeps it right.
Step 1 of 7
The weekly search starts
In plain words
Every week, on the day you choose, the agent starts looking for new businesses that match the customer you described to us: the trade, the size, the area.
Safeguard The do-not-contact list and your existing customers are loaded before the search starts. Nobody on either list can enter the run.
Under the hood
The agent is built to work from an ideal-customer profile written with you as rules it can check, not a loose description: sectors in and out, a size range, the towns you cover, and the signals that make a business worth contacting now, such as a new branch or a job advert. It also carries your exclusions: existing customers, competitors, and a do-not-contact list that only grows. Each run is a batch with its own id, so a restart picks up where it stopped.
Works with
- Your CRM
It searches public sources
In plain words
It looks where anyone could look: business directories and the companies' own websites. It does not buy contact lists or guess at private email addresses.
Safeguard Only what a business has published about itself is used. The agent has no access to bought lists and never guesses a personal email address.
Under the hood
The agent searches public business listings and reads company websites with tools that can fetch a page and do nothing else. Each prospect is stored with the page it came from and the date it was read. Results are de-duplicated by website and trading name, then checked against your CRM, so a business already in your pipeline is not found twice. Contact details are kept only where the business itself publishes them for enquiries.
Works with
- Public business listings
- Company websites
- Your CRM
It checks each one fits
In plain words
Each business is checked against your rules one by one. A firm outside your area, or too small, is dropped with the reason written next to it.
Safeguard A missing fact counts against a prospect, never for it. The agent may not assume a business fits because it probably does.
Under the hood
The language model tests each prospect against the profile rule by rule and returns a structured verdict: pass, fail or unknown for each rule, with a quote from the source page as evidence. The verdict is checked against a schema. A prospect goes forward only if it passes every required rule; an unknown on a required rule drops it rather than letting it through on a hunch. Dropped prospects are kept with their reasons, so you can see what the rules exclude.
It researches and drafts
In plain words
For each business that fits you get a short brief: what they do, which role to approach and why now. Then a first message about them, not a template with their name in it.
Safeguard Every message carries the opt-out line. A draft without it, or with a claim that has no source, cannot reach the approval queue.
Under the hood
The brief and the draft are written from the pages collected in the search, and nothing else. Each claim in the message, such as a new branch or a service they offer, is linked to the page it came from; a claim with no source is removed before you see the draft. The message says who you are and why you are writing, carries no false urgency, and ends with a plain line telling the reader how to ask not to be contacted again.
You approve each message
In plain words
You go through the week's list: approve a message, edit it, or reject the prospect. Nothing is sent to anyone you have not approved yourself.
Safeguard The send step runs only for a message carrying your approval. There is no bulk approve, and the agent cannot approve on your behalf.
Under the hood
Approval is per message, not per batch, and it is a permanent rule, not a setting for the first weeks. The approval screen shows the draft next to the brief and the source pages, so a claim can be checked in one click. What you reject, and why, is logged, and those reasons tune the profile rules for the next run. Whether a business should be contacted at all is your decision: the agent only proposes.
You ask for a rewrite: goes back to step 04, It researches and drafts.
It sends within limits
In plain words
Approved messages go out from your own email address a few at a time through the working day, not in one burst.
Safeguard One approved message per prospect. Any follow-up is a new draft that waits for your approval like the first.
Under the hood
Messages are sent from your mailbox under a daily cap you set and spaced through working hours. Just before each send, the address is checked again against the do-not-contact list, because someone may have opted out since you approved. Each send carries an idempotency key (a unique id for that message), so a retry after a dropped connection cannot send it twice. A bounced address is recorded and never tried again.
Works with
Replies land in your CRM
In plain words
Your CRM holds the week's researched list, what was sent and who replied. Interested replies are flagged for you to answer yourself.
Safeguard An opt-out is final. The business goes on the do-not-contact list as soon as the reply is read, and is left out of every future run.
Under the hood
Each prospect is written to your CRM with its brief, sources, message and approval. Replies are read from your inbox and attached to the right prospect by the email thread they belong to, then sorted as interested, not now, or asking not to be contacted. A reply it cannot sort with confidence pauses all contact with that business until you have read it. The agent does not answer replies: a conversation with a real prospect is yours.
Works with
- Your CRM
Opt-out: never again: goes back to step 01, The weekly search starts.
What it knows
Your ideal-customer rules, your exclusions and do-not-contact list, and what businesses publish about themselves in public listings and on their own websites. Every fact it uses is stored with its source page and the date it was read.
What it may do on its own
Search public sources, check prospects against your rules, write the briefs and drafts, and sort replies. It cannot send anything: every message stays a draft until you approve it, and that does not loosen after the first weeks.
What waits for your yes
Every message, one at a time, and any follow-up. Changes to the ideal-customer rules, the daily sending cap and the opt-out wording. Every answer to a prospect who replies is written by you.
How we test it
Before the first live run, a sample week is built and checked with you: the dropped prospects and their reasons as well as the kept ones, and every claim in every draft traced to its source. Test entries planted on the do-not-contact list confirm that nothing on it can be drafted or sent.
What we measure
Once it is live, these are tracked and reviewed with you.
- Prospects found and kept, with the reasons for drops
- Drafts you approved, edited or rejected
- Replies by kind: interested, not now, opt-out
What you get
- An ideal-customer profile set with you
- A weekly researched prospect list
- Draft outreach per prospect, with an opt-out
- Reply tracking in your CRM
Works with
- Your CRM
- Public business listings
- Company websites
Who it is for
B2B firms and installers who know who they want but have no time to look.
Built to order. We scope it with you on a call, price it before any work starts, and run it in draft mode until you trust it.
Related
- Websites wired to agents
A website that answers every enquiry it brings in.
Custom web development with the agents built in: the site is built to load fast and be found, and it passes every form, chat and booking to an agent that follows up.
Built to order - GEO content agent
Be easy to find when customers ask AI assistants.
Generative engine optimisation: an agent that finds the questions your customers ask ChatGPT, Gemini and Google, and drafts and maintains the pages that answer them.
Built to order - Google reviews agent
A Google review request after every job, at the right moment.
An agent that asks every customer for a review once the job is done, follows up once, drafts your replies and tells you when a customer answers with a complaint.
Built to order