AI workflows and automation for founder-led businesses
Most of your week is work that does not need a person.
I build the systems that take repeating work off your plate, then run or hand over the result. I built them for my own practice first, which is how I know which ones survive contact with a real business.
Point of view
The bottleneck is almost never ideas.
A business rarely stalls because nobody knows what to do next. It stalls because the people who know what to do spend their week on work that repeats. Pulling the same report. Resizing the same photos. Checking whether the thing that broke last month broke again.
None of that shows up on an org chart. It is still where the capacity went.
So the first job is not picking a tool. It is finding which repeating work is worth removing, what it currently costs you to keep doing by hand, and which parts genuinely need a human deciding. Automate the wrong half and you have bought yourself a faster way to be wrong.
What I build
Four categories, and they are sequenced.
The audit comes first because the expensive mistake in this work is automating something that should have been deleted. After that, the order depends on what the audit found.
- 01
The automation audit
Where most engagements start. Two to three weeks inside how the work actually gets done, not how the process doc says it does. You get a ranked list of the repeating work, what each one costs you in hours and errors, and which are worth automating first. Useful on its own, whether or not we build anything after.
- 02
Reporting and monitoring that run without you
The unattended layer. Data pulled, joined, and formatted on a schedule. Checks that re-verify live systems every morning and stay quiet unless something disagrees. This is the category that answers what happens when nobody is watching, which is usually when things break.
- 03
Production pipelines
Content, video, creative, and image work. The repetitive half of production, built into commands. Transcription and captioning, batch image and video processing, brand-consistent asset generation, draft pipelines that take a source and return something a human can edit rather than start.
- 04
The marketing engine underneath
Paid media, search, tracking, and measurement. Automation on top of broken measurement just produces confident nonsense faster. Where the numbers cannot be trusted yet, this is the work that comes first. Google, Meta, LinkedIn, TikTok, Microsoft, GA4, Tag Manager, server-side, CRM.
I can build a workflow and hand it to your team, or build it and keep running it. Both are normal. Handing it over is cheaper and I will tell you when it is the better call.
Selected work
Four problems, and what fixed them.
Mostly founder-led practices in cosmetic surgery and medical aesthetics, where the advertising rules are strict and the gap between a lead and a paying patient is enormous. The second and third of these I built for my own practice before offering them to anyone.
The ad platform was reporting the wrong number.
Three accounts for one surgical group. Page views, YouTube subscriptions, and map-direction clicks were all counted as primary conversions, and duplicate call tracking counted a single phone lead up to three times. The platform reported 102 conversions at a $373 blended cost per lead. Rebuilt against the actual leads, the estimate came to 35 to 50 leads at $761 to $1,087. Every automated bidding algorithm in the account set had been optimizing toward the wrong signal for weeks.
A photo task that ate an afternoon every week.
Before and after pairs, nine sizes each, four brands with different logos and surgeon names. In Canva that ran 20 to 30 minutes per pair, and three to five new pairs arrive every week. Now it is one command and about three seconds. Two to four hours a week back, permanently. The tool took about three hours to build.
A site lost all of its analytics for a week before anyone noticed.
Tracking breaks quietly. A plugin updates, a tag reroutes, a leftover from a prior agency reactivates, and nothing announces it. So every tracking ID across every account now lives in one registry, and a script re-checks live platform state against it every morning. It stays silent unless something disagrees. The gap between a break and someone knowing about it went from roughly a week to about a day.
Six hundred pages were invisible to Google.
A practice had 721 patient gallery cases published. Its sitemap listed 122. The rest were live and indexable and nothing the search engine could follow ever pointed at them. The sitemap now serves 713 entries. On an earlier site with the same defect, the image sitemap went from 41 images to 1,061.
There is no revenue figure attached to that last one yet. The fix is recent and the search checkpoints run at four, eight, and twelve weeks. When there is a number worth trusting, it goes here. Client names are left out on purpose. Ask and I will walk you through a real account.
Method
How an engagement runs.
- 01
Find the repeat work
I sit inside how the work actually happens, not the documented version. What gets done every week, who does it, how long it takes, and where it goes wrong. Most of what surfaces here is not what the owner expected, and some of it turns out to be work that should stop rather than be automated.
- 02
Take the cheapest one first
Highest frequency, lowest risk, clearest before and after. Not the most impressive one. The point of the first build is to prove the approach on something where being wrong costs an afternoon instead of a quarter.
- 03
Build it to run unattended
Anything that only works while someone watches it is not finished. That means it fails loudly, it can tell the difference between nothing is wrong and I could not check, and it stays silent on a normal day so the alerts still mean something.
- 04
Hand it over, or run it
Documented well enough that your team can own it, or kept on my side with a monthly review. Either way the judgment calls stay with a person. The goal was never to remove the thinking, only the typing.
About
You get me. That is the whole model.
I run A Marketing Practice, a one-person firm. That is a deliberate structure, not a stage I am trying to grow out of.
Most of my work is in cosmetic surgery and medical aesthetics, where the advertising rules are strict, the buying decision is slow and personal, and the gap between a lead and a patient is enormous. That environment made me precise about tracking, careful about claims, and skeptical of any number that cannot be traced back to revenue.
I do the strategy and I do the execution. There is no coordinator relaying your question to a media buyer you never meet, and no junior quietly learning on your budget.
I take a small number of clients at a time. That works because most of what an agency staffs up for does not actually need a person. Reporting, monitoring, production work, the morning check that catches a broken tag on a Sunday. That runs on its own now. What needs judgment gets me, every time.
Thy Barnes, A Marketing Practice
Start
If your week is full of work you could describe to somebody else, that is the conversation.
Tell me what the business does, which work eats the most time, and what you have already tried to automate. I will tell you whether I am the right person for it. If I am not, I will say so and point you somewhere better.
Currently taking a small number of new engagements.