Agency AI Cold Outreach Engine
Start getting 20 qualified messages per day for your agency with this simple AI cold out reach engine.
Instructor
David Tainsh
Course details
Lectures
:
4
Level
:
Beginner
- Description
- Curriculum
- FAQ
- Reviews
Free course · For SMMA, marketing and AI agency owners
By the end you have twenty researched messages sent — not twenty drafted.
Six prompts that turn a vague list into an ideal client you can actually find, twenty personalised openers across LinkedIn and email, and the four-message sequence that follows them. Built for new and growing agencies whose pipeline depends on strangers.
- A narrow list, not a big oneBuilt on a trigger you can see from outside, so the opener writes itself.
- Research by hand, not by toolFour minutes a prospect, on public information, with nothing to buy.
- Fourteen days, not one messageThe sequence is written once and covers the whole batch.
Who this is for
New and growing agencies with an empty pipeline.
Whether you are looking for your first client or your next one, the problem is the same: finding somebody who will pay for the work. If you have sent cold messages and had almost nothing come back, it was written for you.
- 01SMMA owners
Selling social and paid to local and e-commerce businesses. A narrow list built on a visible trigger beats a big one built on a job title.
- 02Marketing and digital agencies
Chasing retained work. The sequence is what turns a cold list into conversations, rather than the opener nobody answered.
- 03AI and automation agencies
Selling something the prospect has not budgeted for. You need a trigger that proves they are already looking, and this finds them.
- 04Starting out, or going out on your own
No case studies and no list. The prompts are built to write a message that works with no claim in it at all.
- 05Anyone who rewrote and sent again
You have already tried better copy. The fault is upstream of the words, and this fixes the part you have not touched.
You do not need a case study, a list, a sending tool or a data subscription. You need twenty businesses you would genuinely like to work with.
Why it fails
It was never the copy. It was these four.
Every agency owner’s first instinct after a silent campaign is to rewrite. It rarely moves the number, because none of the four faults below live in the wording.
- ×The list is vague
A job title is not a reason to care. Two thousand names sharing nothing but seniority produce silence, however well you write to them.
- ×The detail is decoration
Their follower count. Their founding year. Found in a second, recognised as a template in less — and worse than sending nothing personal at all.
- ×The claim has no source
A result with no date and no named client. They discount it, and then they discount everything sitting beside it.
- ×It stops at message one
Most replies arrive later than the opener. Nearly all outreach ends before they do, then gets written off as not working.
Fix them in that order. Fixing them out of order is how you end up with a beautifully written message going to the wrong four hundred people.
The trap
Volume is not the same as outreach.
The default advice is to send more. It is the answer you will reach for on your own, it feels like effort, and it is almost never the fix — because sending more of something that is not working simply gets it recognised faster.
The blast
- A list bought or scraped on job title alone
- First line merged from the company name
- A claim nobody could check
- Two asks in the same paragraph
- Sent once, then judged
The researched batch
- Twenty businesses sharing one visible trigger
- A detail that took judgement to find, and a source
- No claim you could not defend on a call
- One ask, almost free to accept
- Four messages across fourteen days
The through-line
AI writes the messages. The research has to be yours.
The prompts can define, draft and diagnose. What they cannot do is look at a business for you — and everything that goes wrong with generated outreach goes wrong on the wrong side of that line.
- 1You do the four minutes
Six checks on public pages, by hand. The detail has to have needed judgement, or it is decoration and the reader knows it.
- 2The prompts write and flag
Every opener is built on a detail you supplied — and the ones built on nothing get marked as weak rather than dressed up.
- 3Nothing gets invented
No claim you have not evidenced, no personalisation manufactured for a prospect with no trigger. Dropping a prospect is a legitimate outcome.
If the openers come back generic, the fix is almost never a better prompt. It is that four minutes of research was thinner than it felt at the time.
The framework
Every message you send is made of the same six parts.
Openers and follow-ups both. Learn the six and you stop guessing at what to change when a campaign goes flat, because the diagnosis is always which one is missing.
- 1Who you write to
The business, the trigger that makes them look, and the words they use for it. A list that fits everybody lands with nobody.
- 2How you sound
Register, sentence length, and the things you would never say. Stops twenty messages reading like twenty outputs of one machine.
- 3What you noticed
One specific, checkable thing, found in under four minutes. If it took thirty seconds, it is decoration.
- 4What you offer
One idea, plainly, with no claim you cannot evidence. One sourced result beats three impressive ones.
- 5What happens next
One ask, scoped, almost free to say yes to. A calendar link plus ‘or just reply’ is two asks, and two asks is none.
- 6What you send after
Four messages over fourteen days, each standing on its own. This is where most of the replies actually come from.
Part three is the only one that cannot be systematised, which is precisely why it is the one that works.
What you walk away with
A batch sent this week — not a course completed.
The measure of this course is whether twenty real businesses receive a message from you. Everything below exists by the end, on a list you chose, and is ready to send rather than ready to start.
- An ideal client profile narrow enough to build a list from today
- A working voice profile, honest about where it is thin
- A four-minute research checklist that needs no paid tool
- Twenty personalised openers — ten LinkedIn, ten email
- The weak ones flagged, so you know where your research was thin
- A four-message sequence with the dates already on it
- A diagnosis that names one broken stage and refuses to say ‘send more’
- The sample size you need before any change means anything
The most common first reply is a question rather than a yes. That means it was read, and the course covers exactly what to send back — which is less than most people think.
What’s included
Every lesson comes with the prompts that build it.
The classes are short. The work happens in the prompt pack that comes with them — a numbered SOP, the six prompts in order, and a note on what a good output actually looks like so you can tell when to run one again.
- SOPA step table in the pack
Seven numbered steps. Three run once, four run every batch. Work down it and a batch goes out.
- AICopy-paste prompts, in plain English
No codes and no variables to install. Square brackets mark the few things only you can supply.
- QAWhat good looks like
A concrete standard for every prompt — including when to push back on Claude for being too polite about your research.
- DEMOA worked agency and three prospects
Apex AI and three South London gyms, already researched. One strong trigger, one moderate, and one worth dropping.
How the course runs
Three short lessons, then the whole thing run live.
Lesson two produces the two profiles everything else reads from. Lesson three writes from them. The demo runs all six prompts end to end so you know what each one should return before you spend an evening on your own list.
Lesson one
Why cold outreach fails
The four faults, the six parts, and the difference between a detail that took judgement and one that took a second.
Lesson two
Define the client
Prompts 01 to 03. The ideal client profile, a working voice profile, and the four-minute research checklist.
Your evening
Research twenty prospects
Four minutes each, on public pages. Expect to drop some — that is the discipline, not a failure.
Lesson three
Write it, send it, fix it
Prompts 04 to 06. Twenty openers, the four-message sequence, and the prompt that names the one broken stage.
Live demo
All six prompts, end to end
The full chain run on Apex AI and three real prospects, including the output that comes back flat and the instruction that fixes it.
The live demo
Including the moment it comes back generic.
Most demonstrations show the run that worked. This one keeps the output that did not, because the moment a prompt returns something flat is where nearly everybody concludes the system does not work — and it is worth seeing what to do about it before it happens to you.
The agency
Apex AI
- Booking automation for local businesses
- One referenceable result, anonymous
- Nothing else it can evidence — and says so
The prospects
Three South London gyms
- One with a job advert naming the problem
- One with reviews saying it, unprompted
- One with nothing usable at all
What you see
The full chain
- A profile pushed back on until it narrows
- An opener flagged rather than manufactured
- The sequence built, message by message
Is it for you
Agency owners who will send a batch, not just build one.
A good fit
- You run an agency, or are building one, and need to start conversations cold
- You will spend four minutes per prospect doing the research yourself
- You want one system you can repeat on every batch from here
- You will run the full fourteen days before judging anything
- You would rather send twenty good messages than four hundred fast ones
Not a fit
- You want a tool that finds and messages people for you
- You want the AI to supply the observations as well as the words
- You are looking for volume rather than replies
- You would send a claim you could not evidence on a call
- You intend to automate this before it works by hand
What it costs
Nothing, and nothing attached to it.
Free
No card. Lifetime access, including anything added later.
4 lessons · 6 prompts · prompt pack · demo pack · live demo
No upsell inside the lessons. There is a paid course this leads to, and it is mentioned once, at the end, after you already have a batch ready to send.
Quick answers
The questions people ask before starting.
Do I need a case study first?
No, and the prompts are written for that. Prompt 01 asks what you can honestly evidence and accepts ‘nothing yet’ as an answer — the strongest opener in the worked demo carries no claim at all.
Does this work for an SMMA specifically?
Yes. The framework is channel-agnostic on purpose, so the messages read as a diagnosis rather than as a pitch for whatever you happen to sell. The demo targets local service businesses, which is where most SMMA lists sit.
Which AI does this use?
Claude. The prompts are plain English with square brackets marking the few things only you can supply — nothing to install, no account to connect, no template to configure first.
Do I need a sending tool or a data subscription?
No. Everything here runs on public information and whatever you already send email and LinkedIn messages with. Enrichment and sequencing tools only make a working system bigger, and this course is about making it work first.
How long does the research actually take?
Four minutes a prospect once you have the checklist, so an evening for twenty. Your first few will take longer because you are still learning where to look, and some prospects should be dropped rather than finished.
Isn’t four messages a bit much?
It reads as pursuit only when every message repeats the same ask. The third one gives something and asks for nothing, which is what makes the fourth land as reasonable rather than as the fourth time you have asked.
What if the openers come back generic?
Then the research was thin, and prompt 04 will have flagged which ones. That is the most useful output in the course — a flag is a research problem, not a writing problem, and the fix is four more minutes rather than a better prompt.
What exactly do I get?
Four lessons — three teaching lessons across 45 slides plus a live demo of the full chain — six plain-English prompts, the prompt pack and SOP in PDF and Word, a worked agency with three researched prospects, and lifetime access including anything added later.
Stop rewriting the message. Fix the list underneath it.
A narrow list, four minutes a prospect, twenty openers and the four messages that follow them — sent this week rather than drafted.
A Digital Plan · adigitalplan.com
Agency AI Cold Outreach Engine
-
1
Lesson 01 — Why cold outreach failsYou wrote them carefully, you sent them, and almost nothing came back. So you rewrote them and sent them again. Rewriting almost never moves the number, because the fault is upstream of the words: the list was vague, the personalisation was decoration, the claim had no source, and the sending stopped after message one.
This lesson makes the case for fixing those four in that order. You will learn the six parts every message is built from, the difference between a detail that took judgement to find and one that took a second, and how to write proof when you have one result — or none at all.In this lesson
• See why the list carries more leverage than any rewrite ever will
• Learn the six parts, and which one to check when a campaign goes flat
• Tell real personalisation from decoration, in your own sent folder
• Write a claim you can defend on a call, or none at all
You finish with: The framework and the argument behind it. Nothing to run yet.
-
2
Lesson 02 — Define the client
Everything after this lesson reads from two profiles, and they are built once rather than per batch. If the ideal client profile is broad, every opener you write afterwards inherits that — and no amount of good writing rescues a message aimed at somebody with no reason to care.
You will run the first three prompts, meet Apex AI and the South London gym target, and see what twelve minutes of real research actually returns: one strong prospect, one moderate, and one worth dropping.In this lesson
• Run prompt 01 and push back until the profile is narrow enough to build a list from
• Separate triggers you can see from triggers that are real but invisible
• Run prompt 02 with a message you actually sent, not website copy
• Run prompt 03 to get the six-check research routine
You finish with: Two profiles, a research checklist, and twenty prospects of your own to research.
-
3
Lesson 03 — Write it, send it, fix it
The writing half. Twenty openers built on the research you did after lesson two, with the weak ones flagged so you know where your four minutes were thin rather than where your writing was. Then the sequence — four messages over fourteen days, written once and reused across the whole batch.
The lesson closes on the part nobody does: reading your own numbers honestly. Prompt 06 is forbidden from telling you to send more, has to name a single broken stage, and tells you how many more sends you need before any change means anything.In this lesson
• Run prompt 04 and read the flagged openers before the good ones
• Learn what changes between a LinkedIn note and a cold email, and what does not
• Run prompt 05 to build the four-message sequence, with the third asking for nothing
• Run prompt 06 against real numbers to find the one stage that is failing
You finish with: Twenty openers, a sequence with dates on it, and a batch ready to send.
-
4
Lesson 04 — Live demo — all six prompts, end to end
No new teaching. This is the whole chain run start to finish on screen, using Apex AI and the three South London gyms from the demo pack — so you can watch what each prompt actually returns before you spend an evening on your own list.
It includes one output that comes back generic, and the follow-up instruction that fixes it. That is deliberate. The moment a prompt returns something flat is where most people conclude the system does not work, and it is worth seeing what to do about it before it happens to you.In this lesson
• Watch prompt 01 get pushed back on until the profile narrows
• See prompt 04 refuse to manufacture a detail for a prospect with no trigger
• See a weak output, and the instruction that corrects it rather than restarting
• Follow the full sequence build, including the message that asks for nothing
You finish with: A clear picture of what good looks like at every step, before you run your own.
Is it really free?
Yes. No card is needed and you get lifetime access, including anything added later. There is a paid course this leads to, mentioned once at the end, after you already have a batch ready to send.
What will I have sent by the end?
Twenty researched messages, ten on LinkedIn and ten by email, built on an ideal client profile narrow enough to build a list from. You also get a four-message follow-up sequence over fourteen days with the dates already on it.
Do I need a sending tool, a data subscription or a bought list?
No. Everything runs on public information and whatever you already use to send email and LinkedIn messages. You need twenty businesses you would genuinely like to work with.
Do I need a case study first?
No. Prompt 01 asks what you can honestly evidence and accepts nothing yet as an answer, and the strongest opener in the worked demo carries no claim at all.
How long does the research take?
About four minutes a prospect once you have the checklist, so an evening for twenty. Your first few will take longer, and some prospects should be dropped rather than finished.
Does it send the messages for me?
No. This is not a tool that finds and messages people for you. You do the research by hand, Claude writes the openers from the detail you supply, and you send them.
What if the openers come back generic?
Prompt 04 flags the weak ones rather than dressing them up. A flag means the research was thin, so the fix is a few more minutes on that prospect rather than a better prompt.
What exactly do I get?
Four lessons, three teaching lessons plus a live demo of all six prompts end to end, the prompt pack and SOP in PDF and Word, a worked agency with three researched prospects, and lifetime access including anything added later.
Please, login to leave a review