In 2017, I built an automation for a small service business. A simple email sequence that fired when a lead came in. Name, number, service type. Three emails over five days. I thought it was brilliant.
What worked: leads stopped falling through cracks. The owner had been manually following up, inconsistently, between jobs. Response rate went up in the first month.
What I got wrong: I built it in a platform the owner had never logged into and didn't want to learn. Six months later, a form field changed, the whole thing broke. Nobody caught it for three weeks. We lost leads the whole time.
That one failure taught me more than most successes. Nine years of building automations for small businesses in Louisiana and Texas. Tools changed constantly. The lessons underneath them did not. Here's what I know now. You can see the work behind some of these patterns in our client case studies.
The most valuable automations I've built are boring. An email that goes out the same way every time. A report that populates itself every Monday morning. A task created the moment a form is submitted.
Early on, I wanted to build impressive things. Conditional logic across six steps. AI reading intent and routing leads. Some was useful. Most was a maintenance nightmare inside six months. Owners who got the most value were running simple, reliable systems they never had to babysit.
I can tell within fifteen minutes of a scoping call whether a project will ship. Ones that ship are small. Defined. One problem, one solution. Ones that don't ship started as one problem and grew. Add CRM sync. Add reporting. Add Slack notifications. Add client portal.
I've watched good ideas die under their own requirements. Rule I follow now: build one thing, prove it works, build the next. A finished small automation is worth ten times more than a half-built complex one.
Some platforms I've used don't exist anymore. Some were acquired. Some the client outgrew. Some I can't even access today.
The question that matters isn't "what's the most powerful option?" It's "who owns this when I'm not in the room?" If nobody on the client's team can open the platform and read what's happening, that automation is one disaster away from becoming a black box nobody trusts.
Sophisticated tools with nobody to own them become technical debt. I bias toward tools the client's team can see and understand, even if that means leaving features on the table.
The mistake I see most now: a business owner drops an LLM into their process and expects the model to figure out the rest. Doesn't work that way. The model is one component. The workflow around it is where value lives.
Who triggers the AI? What data does it receive? What does it do with the output? Where does output go? Who reviews it? What happens when it's wrong? If you can't answer those, the AI is doing something but you don't know what.
The custom AI agents we build are only as good as the process design underneath. We spend as much time on workflow as on model configuration. Sometimes more.
Day 1 looks great almost every time. Demo worked. Client excited. Team did walkthrough. Everything made sense.
Day 30 tells the truth. Did the team actually use it? Did it break when a real edge case came in? Is output being acted on, or piling up in a folder nobody opens? Is the owner still logging in, or has it become background noise?
I build 30-day reviews into every engagement now. That's when the real data shows up. The two fields that need adding. The step the team skips because it's awkward. The report nobody reads.
Businesses getting the most value from automation are not the ones who tried to do everything at once. They're the ones who built something small, let it run, built the next thing, let it run, and kept going.
One working automation per quarter over four years is sixteen automations. Each saves time. Each frees a person to do work that actually requires a person. After a few years, operations look fundamentally different. Not from any single big project. From accumulation.
Most clients don't need transformation. They need the next right step. We talk through exactly that in our consulting engagements.
Clients don't renew because the automation is impressive. They renew because when something breaks at 8 AM on a Tuesday, I pick up the phone.
I've seen automation vendors lose long-term clients over a single week of slow response to a broken workflow. The technical work got them in the door. The reliability kept them. Every automation still running today is running because someone maintained the relationship around it, not just the tool.
This is the part of the business I care about most. Work we do for clients at Starfish is long-term by design. We're not building and leaving. We're staying in it.
Nine years in, discovery calls still surface things the business owner couldn't see. The process they think is the problem usually isn't. There's something upstream they've normalized. A handoff that falls apart every Friday. A task that takes three people because nobody mapped it out.
AI doesn't replace the listening. It doesn't walk into a business and figure out what's broken. A good consultant does that first. Technology comes after, and only when we know what we're solving.
I've had clients convinced they needed a chatbot. Turned out they needed a better lead hand-off process. The chatbot would have automated a broken workflow and made it faster at being broken. Listening first changed the answer entirely.
In 2017, I would have told you to start with whatever tool was getting attention and figure it out as you go. The tools were simpler. The cost of getting it wrong was lower. Trial and error worked well enough.
In 2026, tool choices are more consequential. AI is genuinely powerful, but that power cuts both ways. A poorly designed AI workflow can create bad outputs at scale and do it faster than any human error could. Stakes for getting workflow design wrong are higher than when we were just moving data between spreadsheets.
What I tell clients now is the same thing I believed in 2017, with more confidence behind it: start small, prove it works, build on it. Don't automate anything you don't understand. Own your tools. Review results at 30 days.
The principles from nine years ago still hold. The technology moves faster now, which means discipline around it matters more than it ever did. If you're trying to figure out where to start in 2026, you don't need a grand strategy. You need one working automation and someone who will tell you the truth about what comes next.
A 30-minute call is enough to identify the one automation that will actually move the needle for your business right now. No pitch. Just a real conversation.
Schedule Free Consultation Take the Assessment