Build Systems First, Automate Second

I used to run a food truck & catering company. We’d do things like make A+ chorizo tacos for bridal showers and corporate events. (wrote more about that here)

AI couldn't automate the tacos, not for another 100 years (minimum). It couldn't replicate the perfect char on the chorizo or the way we set up a beautiful station for a bride's special day.

But it could automate getting back to leads in 30 seconds instead of 3 hours.

And that's the difference between booking 15 events a month versus 8.

What AI actually automates (and what it doesn't)

Here's what people are afraid to say, AI automation isn't a magic solution for everything. It's a tool for specific problems.

The catering leads? Everyone asking about catering needs roughly the same information upfront. Send them a PDF with pricing, menus, and availability, and just like that first contact is made. They feel like they're in discussions with you. On top of that they’re automatically add them to your CRM database, follow up later that day or next morning.

Without automation: I'm typing the same information 20 times a week, getting back to people hours later, losing bookings to faster competitors.

With automation: Instant response, consistent information, more time spent planning a great service or more time put into preparing the food or just enjoying an extra 30 minutes a day outside.

More accurate. Less waste. Zero mental energy.

You need a system before you need automation

You can't automate chaos.

You need to know the input or have a criteria to judge or route the input. Some automation system will automate 80% of the process and then you need that human in the loop for the other 20%.

If your input is completely random, if you don't know what output you need, if the process changes every single time, automation will fail.

You need to know:

• What's coming in (the input)

• What needs to come out (the output)

• How you currently process it (the system)

Example that works: Proposals come into your email. You currently read them manually, summarize key points, forward to your team. AI can read proposals, extract what matters, route them correctly. You knew the system, so automation fits.

Example that doesn't work: You think "Reddit hook mining sounds cool" so you build an automation. But you don't actually mine Reddit regularly. You don't have a process for what to do with hooks once you get them. The automation sits unused because there was no proven system to automate.

I repeat build systems first, automate second

I learned this is the way with my video content process.

Every time I built an automation, I needed to create:

• Long-form walkthrough video script

• Short-form clips

• Social captions

• YouTube descriptions

• Key talking points

I was recreating this from scratch every single week. Well not exactly from scratch. I was asking Claude, but it took too much time up remembering what i needed, rewriting, going back and forth. That's when I knew it was time to automate.

Now I have an automation where I brain dump via voice into Slack about what the automation does. It sends everything to my video system that outputs:

• Short-form content ready

• Key talking points organized

• Social captions written

• YouTube descriptions formatted

• Video titles generated

All stored in my database exactly where I need it.

The pattern: I did it manually until I proved I'd do it weekly. Then I automated.

The Reddit hook problem

I built a Reddit hook mining automation because it seemed useful. But I wasn't actually mining Reddit consistently. I didn't have a system for what to do with hooks once I got them. It wasn't a priority.

The automation exists but sits unused. (still think it’s cool though and useful at some point)

Contrast that with my cold outreach system. I know I need to reach certain businesses. I know I'll do this weekly. So I'm building an automation that compiles business lists based on specific criteria.

Every time I think "I need more prospects," instead of manually searching business by business, I generate a list instantly.

But here's the key: I'm starting with manual outreach first.

I'll reach out personally, tweak messages, see what works. Once I have a template that converts, then I'll automate the outreach itself.

When automation actually saves time

Scenario 1: You're doing it weekly You've proven the task matters and you know the steps. Now it’s time to automate.

Scenario 2: You know you need it but can't manage it manually You understand what the system should look like but don't have bandwidth manage it once its built. Build it automated from the start. If you don’t have the time to even build it email me at [email protected]

Scenario 3: The middle processing is consistent Input comes from one place (email, form, sheet). Output goes to another (colleague, database, report). The processing rules are clear. Perfect for automation.

Even if you can't automate the entire flow, automate the compilation. Instead of checking email every 20 minutes for new requests, have AI compile them into an organized list every 3 hours. Review the list, process in bulk.

Boom, boom, fireworks. Done.

What you're really automating

You're not automating thinking. You're automating the space between known inputs and known outputs.

The formula:

• Consistent input data and data source(s)

• Clear processing criteria

• Defined output & output destination

If you have those three things, automation works.

If you're missing any of those three things, systematize first, automate second.

The actual value proposition

Depending on your business size, you probably have between 2 and 10 processes that eat time:

• 3 hours a week here

• 15 minutes daily there

• 30 minutes weekly somewhere else

Find the repetitive ones. The ones your business keeps coming back to. Those are your automation targets.

Or find the systems you need but don't have yet, marketing funnels, lead qualification, customer onboarding. Build them automated from the ground up.

AI automation isn't magic, it's systematization at speed.

But you need the system first.

Thanks for reading,

James