I Gave AI a Goal. It Found the Audience, Wrote the Emails, and Ran the Campaign.

Most conversations about AI at work still start with a prompt.

Those things are useful. I use AI for them constantly.

But I have become much more interested in a different question:

What happens when you stop giving AI individual tasks and start giving it an objective?

I had an opportunity to test that with one of my own projects, The Tourney List.

The experiment started with a relatively straightforward business problem.

The Tourney List is designed to make it easier for families, coaches, and teams to discover youth sports tournaments. To make the platform more useful, we need tournament directors around the world to know the site exists and, ultimately, get more of their tournaments listed.

Traditionally, solving that problem would require a lot of human work.

Instead, I built a system where AI does essentially all of it.

I gave it the objective.

Find youth sports tournament directors around the world, introduce them to The Tourney List, and help us expand the tournament inventory available on the site.

Then I let the system work.

From AI Assistant to AI Operator

There is an important distinction between what I built and simply asking ChatGPT to write a marketing email.

AI isn’t one step in this workflow.

AI operates the workflow.

The workflow looks something like this:

Discover → Research → Qualify → Write → Send → Track → Measure → Report → Continue

My role moved further upstream.

Instead of manually performing each step, I defined the objective, established the parameters, built the system, and monitored what happened.

That may sound like a subtle difference.

I don’t think it is.

I think it represents one of the more interesting shifts happening with AI right now.

The Campaign

The target audience was intentionally broad geographically but very specific functionally:

Tournament directors running youth sports tournaments anywhere in the world.

The objective wasn’t to blast as many email addresses as possible.

In fact, the system was intentionally designed not to do that.

It sends outreach gradually.

Emails are distributed slowly over days rather than sending thousands of messages simultaneously.

That matters for several reasons.

It makes the outreach behave more like an ongoing business-development operation than a traditional email blast.

It also gives the system an opportunity to continually work through the available audience rather than treating the campaign as a one-time event.

There isn’t a morning where someone on our team has to sit down and decide:

“Who should we email today?”

The system already knows what it is supposed to be doing.

It keeps working.

The Results

So far, the automated campaign has produced:

Emails sent

4,231

Successfully delivered

3,958

Measured unique opens

1,681

Adjusted open rate

44.83%

Unique clickers

568

Adjusted click-through rate

15.15%

Those engagement numbers were encouraging.

But they aren’t actually the result I care about most.

Because an open isn’t the objective.

Neither is a click.

The real objective is making The Tourney List more useful.

And the outreach has now contributed to hundreds of additional youth sports tournaments being listed on the platform.

That is where the experiment becomes much more interesting to me.

AI didn’t simply generate an email.

It performed work that ultimately resulted in the underlying product becoming more valuable.

The Funnel

The complete system is better understood as a funnel:

1. AI discovers tournament directors
2. AI researches and qualifies them
3. AI creates the outreach
4. AI sends it gradually
5. Tournament directors receive the message
6. 568 people actively click through
7. Tournament directors engage with The Tourney List
8. Hundreds of additional tournaments are ultimately listed

That last step changes how I evaluate the experiment.

If the outcome were simply “AI sent 4,231 emails,” I wouldn’t find it particularly impressive.

Sending email isn’t difficult.

Creating a system that continuously discovers the right people, communicates with them, measures their response, and contributes to a measurable improvement in the product is much more interesting.

Why I Didn’t Send 4,231 Emails at Once

Automation creates an interesting temptation.

Once something works, the instinct is often to turn up the volume.

If AI can send 100 emails, why not send 1,000?

If it can send 1,000, why not send 10,000?

I deliberately took the opposite approach.

The system sends slowly over multiple days.

That was intentional.

I wanted something closer to an autonomous business-development process than a giant marketing blast.

The goal isn’t:

Send as much email as possible.

The goal is:

Continuously identify appropriate opportunities and execute the outreach required to pursue them.

That distinction becomes increasingly important as AI systems become capable of doing more work independently.

The limiting factor shouldn’t necessarily be what the technology can do.

It should be what makes sense for the objective.

The Human Role Didn’t Disappear. It Changed.

Whenever I describe systems like this, there is an obvious question:

If AI is doing all of that, what exactly are you doing?

That question gets to the part of AI that I find most interesting.

My work moved from executing the individual tasks to designing the system responsible for those tasks.

AI performs the execution.

That changes the human role from:

Do the work→Design, direct, evaluate, and improve the system doing the work.

For me, that has become one of the biggest changes in how I think about AI.

This Isn’t Really an Email Story

At first glance, this looks like a case study about email marketing.

I don’t think it is.

Email just happens to be the output channel.

The more important experiment is connecting AI capabilities that are normally treated as separate tasks.

Finding information isn’t particularly revolutionary anymore.

Neither is writing an email.

Neither is analyzing campaign statistics.

But connect them together:

Discovery + Research + Qualification + Writing + Execution + Measurement + Reporting

and you have something different.

You have a workflow.

Give that workflow memory, rules, tools, and an objective, and it begins to look less like an AI assistant.

It starts looking more like an operating system for a specific business function.

That’s the part I’m interested in.

AI as Tool → AI as Workflow → AI as Operating Layer

I’ve started thinking about applied AI in three stages.

1. 01

   AI as a Tool

   You ask AI to perform an individual task.

   “Write an email to this tournament director.”

   Useful.

   But still fundamentally a human workflow with AI inserted into one step.

2. 02

   AI as a Workflow

   AI connects several tasks.

   Research this tournament director → understand the organization → draft appropriate outreach.

   Now we’re removing some of the handoffs between tasks.

3. 03

   AI as an Operating Layer

   Instead of providing the individual tasks, you provide the objective.

   Find tournament directors around the world and introduce them to The Tourney List.

   The system figures out and executes the steps required to pursue that objective within the rules I’ve established.

That’s where this experiment sits.

And it’s where I think some of the most interesting applications of AI are going to emerge.

Why I’m Experimenting With This

My professional background isn’t software engineering.

It’s fundraising and nonprofit leadership.

I’ve spent much of my career in environments where there are always more opportunities than there are people available to pursue them.

The limiting resource is almost always human capacity.

That experience has heavily influenced the way I approach AI.

I’m less interested in whether AI can make one task 20 percent faster.

I’m much more interested in whether we can redesign the workflow so that the task doesn’t require someone to continuously execute it in the first place.

The Tourney List gave me a relatively low-risk environment to experiment aggressively with that idea.

And what I’m learning there directly influences how I think about AI in nonprofit organizations.

The Nonprofit Connection

Imagine applying the same architecture to fundraising.

Not necessarily the same level of autonomy.

The context, stakes, privacy considerations, and importance of human relationships are very different.

But the underlying architecture is interesting.

Instead of:

Find tournament directors → Research → Outreach → Track → Report

imagine:

Identify foundations → Research alignment → Prioritize opportunities → Prepare outreach → Track relationships → Surface follow-ups → Report progress

The human fundraiser shouldn’t disappear from that system.

Quite the opposite.

The objective should be to remove as much administrative friction as possible so the fundraiser can spend more time doing the things humans are actually good at:

That is also why I am experimenting with AI systems in my nonprofit work.

The Tourney List gives me a place to push the automation much further and see what happens.

The Metric That Matters

It would be easy to end this case study with:

44.83% open rate.

Or:

15.15% click-through rate.

Those are useful indicators.

But they’re not the metric I find most meaningful.

It’s this:

Hundreds of new tournaments.

Those are real events that are now part of the platform’s inventory.

Which means more searchable opportunities for the families, coaches, and teams using The Tourney List.

That’s the difference between AI producing activity and AI contributing to an outcome.

And that distinction matters.

What I Learned

The biggest lesson wasn’t that AI can write good emails.

We already know that.

It wasn’t that AI can research people online.

We know that too.

It was that the real leverage begins when those capabilities stop being isolated.

When research feeds qualification.

Qualification feeds outreach.

Outreach feeds measurement.

Measurement feeds the next decision.

And the system continues operating without waiting for someone to manually initiate every step.

That’s when AI starts changing the structure of the work itself.

The Experiment Is Still Running

Perhaps the most interesting part of this case study is that there isn’t really an ending yet.

The campaign wasn’t:

Send 4,231 emails and stop.

It’s an operating system.

It continues discovering opportunities.

It continues working through the audience.

It continues sending gradually.

It continues tracking what happens.

And as The Tourney List expands into additional sports, regions, and countries, the potential universe the system can explore expands with it.

The numbers in this article are therefore a snapshot.

Not a final campaign report.

What I’m Trying to Figure Out

I’m increasingly convinced that the most important question organizations should be asking about AI isn’t:

“Where can we use AI?”

That’s too broad.

And it usually leads to a collection of disconnected tools.

I think a better question is:

“What objective currently requires dozens or hundreds of small human actions, and could we design a system capable of pursuing that objective?”

That’s a very different way of thinking.

It changes the conversation from prompts to systems.

From tasks to outcomes.

From AI as something sitting beside the work to AI becoming part of how the work actually gets done.

That’s the experiment I’m running.

And The Tourney List is one of the places where I get to see just how far that idea can go.

AI built into the work, not around it.

James Graham is a nonprofit executive, fundraiser, entrepreneur, and applied AI builder. He writes about experiments at the intersection of fundraising, nonprofit leadership, AI, automation, and building practical systems.