Why does a med spa need guardrails before automation?
The big picture. With a fixed budget you have three poor options. An agency that reports after the bill lands, a freelancer who checks twice a week, or you, between patients.
All three fail the same way. A bad ad set burns a week of budget before anyone notices. Automation fixes that and adds a worse risk, an unattended tool holding the credit card.
What Kintsu's ad automation actually does
So we drew the line first: three things the automation may do on its own, three it may never do, and the owner's limits around both.
What the automation may do
Automation that can only ever save you money
The second row is permanent. It is not a limit we lift once you trust us.
It can
It can never
When you say yesYou can approve a change yourself, and the same guarded actions run for that one job while the automation stays switched off. Every change is written down somewhere nobody can edit it, including us.
A readback after every change is the only reason we knew a platform had switched off one of your ads by itself on September 1, 2026.
- Limits set with the owner. The automation works inside them. Raising a budget, starting a campaign, and changing ad copy stay with a person. Since August 26, 2026 every action writes to a log the owner can read.
- A cap the owner sets. A combined Google and Meta limit of $1,000 a month took effect on August 4, 2026.
- A weekly one-page report. Once a week the system reads the site analytics and writes up what to fix, and a person approves every change.
- Social posts checked against the brand voice. Drafts run a voice check that stops the batch on a hard miss, and nothing publishes by itself.
- Numbers from their own site. Google Analytics went in across the marketing pages in May 2026, so decisions rest on what visitors do.
That report closes a short loop: read the numbers, propose a change, get a yes from the owner, then measure again.
Once a week
Software that can suggest, but never publish
The week runs in a straight line and stops at a person. Nothing on the site changes until someone says yes.
Step one
It reads the numbers
Once a week, on a machine of ours, with nobody sitting at a keyboard.
Step two
It proposes one change, and says why
The suggestion is written up in one page for a person to read, and that is the end of what it can do on its own.
The software stops here
Step three
A person says yes, or does not
Only after that does anything on the live site change.
A separate check watches that the weekly job actually finished, so a job that stops without saying so fails loudly instead.
What we did not build. In a July 2026 review, five of six proposed tools were dropped as unnecessary.
Why the automation can only spend less
The automation writes in one direction. It can pause and it can cut, and that is the whole list.
The usual build also raises budgets when the numbers look good. On September 1, 2026 that would have cost real money. One campaign looked healthy in the ad platform's report, while the site's own data showed the booking intent was not there.
Every spend change is read back after it is made, so what the platform actually did gets checked against what we asked it to do.
What went wrong
The things that went wrong here were all on our side of the work, in how it was built and scheduled.
- A weekly report ran once in ten days. From July 25 to August 3, 2026 it ran on one unattended laptop and completed exactly once. The trap is now written into the schedule.
- We built something more complicated than the job needed. A separate monitoring service came out again in favor of something simpler.
- Three tries at a blog system. Two content tools were adopted and dropped, and a third is chosen but still unbuilt. None of that churn reached the live site.
- One piece of the social pipeline is still manual. Artwork edits are done by hand, because the design file is too large for the tool that would automate them.
Results
Measured
These are spending measurements, not business returns. The bars set August spending against the $1,000 cap.
Monthly advertising spend
A hard ceiling on what your ads can spend
The ceiling has been $1,000 a month across both ad platforms since August 4, 2026. Move it to see the rule: anything above the line needs a person to say yes.
At $1,000 a month, two of these three sit above the line, and the platform cannot raise its own budget to reach them.
Your advertising cannot run away from you, and when it overshoots you hear the number from us.
The catch. The $1,994 pre-dates the cap. Most of it was committed before the limit took effect, and no cost per booking click has been measured since. The one we did measure, $76.70, sits a long way from the $15 we were aiming at.
Cost per booking click
We publish the target we missed
We aimed at $15 or less for every click on a booking button. Move the line to see how far off that was.
At $15 a click, both of these sit above the line. We set that target, we missed it, and you are reading the number rather than an average that would hide it.
A click on a booking button is not a booking, and no booking, revenue or return figure exists for this work yet.
Not yet measured
Not yet measured. No revenue, booking count, conversion lift, search ranking, or return on ad spend figure exists yet, and we will not imply one. The number that will judge this work is the Zenoti booking count set against what the ads cost. The full Kintsu write-up lists everything else we publish.