0.0
I've added 47 leads to my CRM this
3.0
month, and I haven't sent a single cold
5.379
email or
6.28
scrolled a single LinkedIn feed.
8.019
And AI Loop generates them while I work
10.359
on other things.
11.56
Today I'm building it with you in 18
13.859
minutes.
16.49
Here's the picture.
17.62
Make.com runs every six hours.
20.039
It pulls fresh posts from a list of
22.239
public sources.
23.179
For me, that's a curated set of subreddits
25.899
and x-accounts where my ideal customers complain
28.6
about problems I solve.
30.12
Each post gets sent to a lead scout,
32.659
a system prompt that knows my ideal customer
35.259
profile cold.
36.38
It returns either qualified with reasons or skip.
40.34
Qualified leads get a Notion entry and a
42.859
Slack ping.
43.719
Skips disappear.
44.88
I never see them.
46.039
18 minutes to build.
47.479
No cold email.
50.32
Step 1, the lead scout.
52.659
This is the brain.
54.219
We're not building a custom GPT this time.
56.7
We're building the system prompt that comes.
58.6
We're building the system prompt that the custom
58.78
GPTs are made of and deploying it inside
61.38
Make.com.
62.46
Same logic.
63.399
Different context.
64.64
Here's the prompt structure.
66.28
Identity.
66.879
You are a B2B lead qualifier for a
69.319
fractional CMO who serves SaaS companies between 5
72.28
and
72.519
50 employees, with revenue between 2 and 20
75.439
million.
76.099
Substitute your own ICP.
77.739
Rules.
78.48
Read the post.
79.299
Decide qualified or not.
80.739
Qualified means matches the company size, mentions a
83.9
problem related to growth or marketing,
85.64
and the person posting has decision-making authority.
88.5
Output format, respond in JSON.
90.7
If qualified, return name.
94.019
Step 2, the trigger.
96.219
Open Make.com, click Create New Scenario.
100.12
The trigger module depends on your source.
102.62
For X, use the X module watch posts
105.519
filtered by keywords or by a list of
107.659
accounts.
108.26
For Reddit, use the Reddit module.
110.519
Watch new posts in subreddit.
112.439
For RSS, use the RSS module.
115.719
Watch RSS feed items, best for niche industry
118.92
blogs.
119.799
I'm using Reddit because three subreddits cover my
122.64
entire ICP.
124.019
Configure the trigger, subreddit name, limit 10 new
127.099
posts per run.
128.3
Schedule the scenario to run every 6 hours.
131.08
The schedule is on the bottom left of
132.96
the canvas.
133.699
Run.
134.28
Once drop-down, change to .
137.8
Step 3, the qualifier.
140.159
Add an OpenAI module after the schedule.
142.439
Choose Create a Completion.
145.12
Model, GPT-40 Mini.
147.46
The system prompt is exactly what we wrote
150.139
in step 1.
151.06
Paste it into the system message field.
153.319
The user message, just the post content from
156.099
the trigger, mapped from the Reddit module's output.
159.039
Now the temperature setting.
160.68
Drop this to .1.
162.28
Low temperature means consistent output, which we need
165.599
because we're parsing JSON downstream.
168.039
High temperature means creative output, which here would
171.06
mean inconsistent JSON downstream.
174.16
The
176.42
.
179.879
.
181.98
.
182.62
.
182.8
.
182.96
.
183.24
.
183.28
.
184.08
.
184.18
.
184.5
.
184.56
.
186.28
.
191.2
.
191.78
.
191.819
.
191.84
.
191.879
.
192.02
.
192.039
.
192.06
and .
192.099
.
192.12
.
192.139
.
192.159
.
192.18
and .
192.219
.
192.24
.
192.259
.
192.28
.
192.3
.
192.319
and .
192.36
.
192.379
.
192.4
and .
192.439
and .
192.479
.
192.5
and .
192.539
.
192.56
and .
192.599
and .
192.639
.
192.659
and .
192.699
.
192.719
and .
192.759
and .
192.8
e .
194.74
and .
194.9
and e , 1 .
195.199
.
196.84
Let's talk about how to change the typical
197.52
look of a level EXCEPTION than the STORIES.
199.98
Add
203.72
a filter .
205.46
condition JSON response
206.28
.
206.42
.
206.46
.
206.54
.
206.58
.
206.599
JSON string field.
207.919
The first time you run this, paste a
210.219
sample qualified response into
211.879
the sample field so Make.com knows the
214.46
structure.
215.24
After parsing, you'll have the qualified
217.3
lead's name, company, problem, and suggested message available
221.419
as separate variables.
224.46
Step 5.
225.8
The Actions.
226.879
We're back to familiar territory from the last
229.599
video.
230.3
On the same
231.139
branch, add Notion.
232.719
Create a database item.
234.52
Pick your CRM database.
236.52
Map the parsed JSON
237.919
fields, name to title, company to company column,
241.439
problem to notes, suggested message
243.639
to a draft outreach column.
245.479
Set status to new inbound.
247.699
Save.
248.56
Add a slack.
249.719
Send a message
250.68
module.
251.4
Channel.
252.219
Leads.
252.96
Message.
253.8
New inbound lead.
255.28
Name from company.
256.66
Problem.
257.54
Problem.
258.339
Draft message ready in Notion.
260.319
Map the parsed JSON fields.
261.12
Name to title.
261.12
Company to company column.
261.12
Problem to notes.
261.12
Save.
262.759
Now, when a qualified lead is detected anywhere
265.439
in your monitored sources, you can
267.139
save the data.
268.3
This is where the loop closes.
270.62
When I review a lead in Notion, I
273.06
open the
273.519
draft message, edit it for 30 seconds, send
276.62
it from Gmail.
277.54
The recipient replies.
279.54
That
280.1
reply hits my Gmail inbox, and the email
282.839
triage system from the last video catches it,
285.42
classifies
286.12
it as a lead, updates the same Notion
288.74
entry with the reply, and pings me again.
291.04
So the workflow goes.
293.139
AI finds lead.
294.68
AI drafts message.
296.48
I send.
297.439
Recipient replies.
298.86
AI catches
300.04
reply.
300.72
AI updates CRM.
302.779
I see notification.
304.379
Closed loop.
305.519
I touch this entire pipeline
307.56
for 30 seconds.
310.42
Three mistakes to avoid.
312.319
1.
313.0
Using the heavy GPT-4 model for the
315.699
qualifier.
316.3
The mini model is plenty for binary qualification
319.28
at one-tenth the cost.
320.839
2.
321.019
Using the heavy GPT-4 model for the
321.019
qualifier.
321.019
The mini model is plenty for binary qualification
321.019
at one-tenth the cost.
321.019
2.
321.379
Letting the scenario run without a kill switch
323.639
on cost.
324.399
Add an open AI usage cap inside
326.639
make settings.
327.62
Mine set at $20 a month.
329.639
3.
330.22
Overbroad sources.
331.459
I started with 12 subreddits and got noise.
334.18
3 high signal sources beat
335.839
12 mediocre ones every time.
337.819
Curate the inputs, not the filters.
339.939
Not the filters.
340.959
Not the
341.439
filters.
343.199
30 day results.
344.879
47 qualified leads added to CRM.
348.319
6 replied to my outreach.
350.36
2 BDs.
351.0
2 BDs.
351.019
3 BDs.
351.779
1 back in my control.
353.0
1 BDs.
355.339
2 BDs.
357.56
2 BDs.
359.16
3 BDs.
360.42
Here we go again and a load up.
361.42
2 BDs.
361.839
2 BDs.
361.939
3 BDs.
363.339
1 BDs.
367.539
4 BDs.
369.22
4 BDs.
369.48
3 BDs.
369.959
4 BDs.
370.36
6 BDs.
370.639
10 BDs.
371.04
12 BDs.
372.519
15 BDs.
373.24
15 BDs.
378.5
12 BDs.
380.1
Scout System prompt and the Make.com blueprint
382.74
are in the description.
384.56
Modify the prompt for your ICP, swap in
387.56
your sources, you have the same loop in
389.74
20 minutes.
390.56
Subscribe because next video I'm comparing Notion AI,
394.579
ChatGPT, and Clawed side by side
397.439
after 30 days using each as my daily
399.959
driver.
400.639
The verdict is not what I expected.
402.86
See you there.
