Here's how most AI and automation proposals end: with a monthly platform subscription. Every fix, every workflow, every clever idea somehow lands on a tool that bills you every month, forever.
I want to explain why my proposals usually don't, because the reason tells you a lot about how this industry works.
Follow the money
When a consultant builds your solution on a subscription platform, something quiet happens: your problem becomes their recurring revenue. Many platforms pay referral commissions. Even when they don't, the consultant benefits, because a business tangled up in a platform needs the person who tangled it.
I'm not saying everyone selling subscriptions is running a con. Plenty believe in their stack. But you should notice the incentive: the seller profits from complexity, and you pay for it. Every month. Whether the thing ran ten thousand times or twice.
The minimum-stack rule
My rule is boring and I apply it to my own business first: use the minimum stack that does the job. No middleware, no automation platform, no monthly anything, unless real recurring volume actually earns it.
Most small business automations are simple. Move this data there. Send this email when that happens. Check this number and yell if it's wrong. A small script can do that. It runs on things you already pay for, or on tiers that cost nothing. It does one job, it doesn't phone home, and nobody bills you for its existence.
The ownership test
Compare two versions of the same fix.
Version one: a $99-a-month automation platform moves your data around. It works. It also keeps working FOR $99 a month, and the day you stop paying, your automation evaporates. The logic lives in their cloud, in their format. Moving away means rebuilding from scratch. That's not a tool, that's a lease with your process as collateral.
Version two: a small script does the same job. It cost something to build, once. It runs at zero dollars a month. It's a file you can see, hand to any future tech person, and move to a new computer. When I finish a build, the client gets documentation of what it does and how to run it. If I got hit by a food truck tomorrow, their stuff keeps working.
That's the ownership test: if you stopped paying everyone tomorrow, what still runs? The version-two answer is "everything I built." The version-one answer is "nothing," and it's the industry default.
When a platform IS the right call
The rule is minimum stack, not zero stack. Middleware earns its seat when the volume and complexity are real:
- Dozens of workflows genuinely running daily, where visual management beats reading code
- A team that needs to see and edit automations without a developer in the loop
- Integrations that are brutal to maintain by hand and rock-solid on the platform
- Volume where the platform's reliability guarantees are worth rent
Notice these are earned thresholds, not starting points. An empty business process hasn't earned a $99 seat. A process running 500 times a day might have. Start minimal, upgrade when the simple version actually breaks, and you'll skip years of paying for capacity you never used.
Run the audit yourself
You don't need me for this part. Pull up your card statement and list every software subscription. For each one, answer a single question: what did this earn me last month?
Not "what could it do." What did it DO. Real answers sound like "invoicing runs through it" or "we booked 40 appointments on it." If the answer is "we might use it more eventually," that's not a tool, that's rent on a good intention.
Most businesses I look at are carrying at least a couple hundred dollars a month in good intentions. That's real money. It's also a signal about how the stuff was sold to you in the first place.
The takeaway
When somebody proposes a fix for your business, ask where the logic lives and what happens when you stop paying. There are only two answers: you own it, or you're renting it. Renting is sometimes right. But it should be a decision you made, not a default somebody sold you.
Minimum stack. Own the simple thing before you rent the complex one.