How Far Can You Automate Sales Outreach Without Removing the Human?
I tested how much of the outreach process ChatGPT could handle while keeping the final decisions and sending in human hands.
I wanted to test something.
Not whether ChatGPT could write an email. We already know it can do that.
Thanks for reading Behind the Service Call! Subscribe for free to receive new posts and support my work.
I wanted to know how much of the sales outreach process could be handed off without giving up the parts that still need human judgment.
So I created a hypothetical scenario.
Imagine a local HVAC company wants to sell more commercial maintenance agreements.
Instead of relying only on ads or its existing customer list, the company wants to find organizations that already have relationships with local business owners, property managers, facility managers, and commercial property owners.
The question was:
How much of that outreach process could ChatGPT handle before a person needed to step back in?
So I tested it.
Step one: find the right organizations
I started with the goal and had ChatGPT research potential outreach targets.
That could include organizations like:
chambers of commerce
property management associations
commercial real estate groups
facility management organizations
economic development groups
local business newsletters
business networking organizations
industry associations
But I didn’t just want a list of names.
For each organization, I wanted the research to explain:
why it was relevant
who it served
why its audience might care about the offer
what kind of outreach would make sense
its website
a contact person, if one could be found
a direct contact page or email address when available
That matters because a long list of organizations isn’t very useful by itself.
The useful part is understanding why each one belongs on the list.
Step two: turn the research into a working tracker
Once the research was done, I had ChatGPT turn it into an outreach tracker.
The spreadsheet included fields for things like:
organization
website
contact
why we’re reaching out
the specific ask
outreach status
follow-up status
notes
Now the research wasn’t sitting in a chat.
It had become an actual working pipeline.
Someone could open the tracker and immediately see who was being contacted, why they mattered, what the ask was, and where each outreach attempt stood.
Step three: create the emails
Then I wanted to see if the system could go one step further.
There were more than 40 organizations on the list.
I didn’t want 40 copies of the same generic email.
A chamber of commerce shouldn’t necessarily receive the same message as a property management association.
A local newsletter might be asked to feature the offer.
A business organization might be asked whether it shares member resources.
A commercial real estate group might get a different request based on the businesses it serves.
So ChatGPT used the research it had already gathered to draft a different email for each organization.
Then those emails were placed into the email draft folder.
Nothing was automatically sent.
That was intentional.
Because this is where I think the line matters.
The repetitive parts can be handed off.
The final judgment should stay with the person responsible for the outreach.
The whole test took about 20 minutes
That was the part I was most interested in.
In roughly 20 minutes, the hypothetical scenario went from:
“We need to sell more of this.”
to:
more than 40 researched outreach targets
an explanation of why each one was relevant
a customized ask for each organization
websites and contact information
a working outreach tracker
statuses for managing follow-up
more than 40 customized email drafts ready for review
The test wasn’t perfect.
Research still needs to be checked.
Contact information can change.
Some organizations may turn out not to be a fit.
Some emails will need edits.
And I would review every message before sending it.
But that’s the point.
The goal isn’t to remove the human
It’s to change what the human spends time doing.
Without AI, this kind of work can involve hours of:
researching organizations
opening websites
copying contact information
building spreadsheet rows
deciding what to ask
writing individual emails
organizing follow-up
With the right workflow, much of that can be handled before the person even begins reviewing.
That means the human can focus on questions like:
Is this organization actually worth contacting?
Does the ask make sense?
Is this the right person?
Does the email sound right?
Would I actually send this?
Should we change the angle?
Those are judgment calls.
That’s where I want the human involved.
The interesting part isn’t that AI can write emails
That’s easy to reduce this to.
“ChatGPT can write 40 sales emails.”
True.
But that’s not what made the test interesting.
The useful part was that the research, organization, tracking, customization, and drafting all became parts of one connected workflow.
Each step fed the next.
Research became structured information.
That information became a working tracker.
The tracker became customized outreach.
And the outreach landed somewhere a person could review it before anything went out.
That’s a much bigger shift than asking AI to write something faster.
And 40 wasn’t some magic limit
That was simply the number I used for the test.
The same workflow could have been built around 100 organizations.
Or 200.
At that point, the question becomes less about whether ChatGPT can handle more and more about how much you actually want to review, manage, and follow up on.
Because the goal isn’t to automate the biggest possible list.
It’s to build a system where AI handles the repetitive work at scale, while the human stays responsible for judgment, quality, and what actually gets sent.
That, to me, is the more useful question:
What parts of the process need your thinking, and what parts are just taking up your time?
