AI Automation for SMBs: What to Automate First (and What Not To)
Not every process is worth automating. The ones that are: repetitive capture, reporting, follow-up and reconciliation — high volume and clear rules. A practical method for picking the first one, and why "governed" is the word that matters.
The right question is not "can I?"
For the last couple of years, the answer to "can I automate this with AI?" has almost always been yes. The question that separates companies that capture value from companies that accumulate experiments is a different one: which process should you automate first — and which should you deliberately leave with people?
McKinsey Global Institute documented it in "A future that works": a significant share of work activities can be automated with technology that already exists. But the same study makes a point almost nobody repeats: automation rarely replaces whole jobs — it transforms activities within jobs. The practical consequence for an SMB is direct: you are not trying to replace anyone; you are trying to take away from your team the hours a machine does better, so people can do what the machine cannot.
This article is the selection criteria we use ourselves, in writing.
The four process classes that automate well
Automation pays where two conditions intersect: high volume and clear rules. Four classes of process almost always qualify:
1. Repetitive data capture
Moving information from an email into a spreadsheet. From a PDF into the ERP. From a form into the CRM. If your team retypes what is already written somewhere else, that retyping is pure volume under a fixed rule ("copy field X into system Y") — and every manual entry is an error opportunity that automation removes at the root.
2. Recurring reports
The weekly sales report someone assembles every Monday with the same cuts, the same sources and the same format. If this week's report is built the same way as last week's, the rule already exists — it just lives in one person's head instead of in a system.
3. Follow-up
Payment reminders, order confirmations, the "we received your request", the third email to the prospect who didn't answer. Follow-up dies because it competes against urgent work — not because nobody knows how to do it. It is the class of process where automation doesn't just save hours: it executes what today simply doesn't get done.
4. Reconciliation
The bank statement against the invoices. Orders against deliveries. Billed against collected. Reconciliation is systematic comparison of two lists under explicit rules — exactly what a machine does without fatigue and a person does with errors starting at hour two.
What these four classes share: the success criterion fits in one verifiable sentence. "Every supplier invoice is recorded in the ERP the same day, with an amount identical to the PDF's." If you can write that sentence, you can automate the process — and audit the result.
The processes you should not automate
Just as important. Three signs a process is NOT a candidate:
It requires professional judgment
Deciding whether a customer deserves credit, negotiating a discount, drafting the response to a delicate complaint. AI can prepare the draft or assemble the case file — the decision belongs to a person with a name, because when something goes wrong, an auditor or a customer will ask who decided. "The system" is not a defensible answer.
It rarely happens
A process that runs twice a year does not pay back its automation, even if it is perfectly automatable. The cost of building, testing and maintaining the workflow exceeds the hours it saves. Automation is an investment that returns through repetition: without repetition, there is no return.
Its exceptions are undefined
If the answer to "and what happens when it arrives incomplete?" is "it depends", the process is not ready. Not because the technology falls short, but because nobody has decided the rule. Automating a process with ambiguous exceptions does not remove the chaos — it accelerates it.
The right move in these cases is not to discard the process — it is to split it. Many "judgment" processes have a mechanical part (assembling the file, validating the data, preparing the comparison) and a final decision. You automate the mechanical part; the decision stays in front of a person, at an approval gate.
Why "governed" is not a decorative adjective
A script running on its own and a governed automation system appear to do the same thing. The difference shows up the day something goes wrong — or the day someone asks.
Governed automation means four concrete things:
- Approval gates. Nothing irreversible executes without sign-off from a designated person. The system prepares; a person authorizes.
- An audit trail. Every action is recorded: what was done, when, with which data, and who approved it.
- A stop button. The workflow halts the moment you decide, without calling anyone.
- A rollback procedure. Before the first run, the procedure to undo what the workflow did exists — in writing.
Why does it matter in Mexico, today? Two examples with no hypotheticals:
The SAT. If your automation touches invoicing, it touches CFDI 4.0 — a tax document with strict issuance and cancellation requirements. A workflow that issues invoices with no record of what it issued, when, and why, is a tax liability waiting for a review. With an audit trail, every invoice carries its complete history.
The ISO 9001 auditor. The standard requires documented information and traceability for the processes that affect quality. When the auditor asks "who approved this output, and under what criterion?", a loose script has no answer. A governed workflow has it in print.
The general rule: an automation you cannot explain to an auditor, a client, or the tax authority is not an asset — it is a risk with good productivity.
Start with one
The temptation, once you see the potential, is the big project: "let's automate the whole operation." That ambition kills more automation projects than any technical limit, for a simple mechanical reason: a wide scope multiplies undefined exceptions, and every undefined exception is a meeting, a delay, and an overrun.
One process, well bounded, delivered working, achieves something no master plan does: it teaches your team to operate automation — to approve, to read the audit trail, to use the stop button — with bounded risk. The second process gets chosen with what the first one taught you.
The method for picking the first process
Three steps, in order, before building anything:
1. Map the real process — not the ideal one
Sit with the person who runs it and document what actually happens: every input, every output, every "ah, but when it comes from that client we do it differently". The process in the owner's head and the process that happens on the floor are almost never the same — and only the second one can be automated.
2. Measure the hours
Count the real hours per week the process consumes, with the person who works them. Don't estimate in a meeting: measure. That number turns the decision into arithmetic — a many-hours-a-week process with clear rules pays for itself; a minutes-a-month process does not.
3. Write the acceptance criteria BEFORE building
One page: what goes in, what comes out, what counts as "it works", and how each exception is handled. This document does two jobs: it forces the ambiguities to be decided before they cost money, and it becomes the contract against which delivery is accepted — or rejected. If an automation vendor refuses to fix acceptance criteria in writing, that is all the information you need about that vendor.
What it costs, said in full
Our custom automation offer works on a fixed written scope: USD $1,300 to $2,900 depending on the process — a written-quote range, not a list price: the scope document you define before we start fixes the price and the committed date. Delivery takes 1 to 3 weeks, includes training, and includes 30 days of post-delivery adjustments. Every workflow ships with approval gates, an audit trail, a stop button and a documented rollback procedure. Any scope change is quoted separately — there are never surprise hours.
And the point that weighs most over time: you operate the result. Delivery includes documentation and training so the system is yours — with no dependency on us afterwards.
The first step costs nothing
Before quoting anything, measure where you stand. Our free assessment takes 3 minutes and returns a maturity score with recommendations prioritized by impact — no card, no mandatory call.
And if you already know which process is costing you hours every week, the automation page details the full offer: simiriki.com/automatizacion.
---
Sources cited in this article:
- McKinsey Global Institute, "A future that works: Automation, employment, and productivity" (2017) — a significant share of work activities is automatable with currently demonstrated technologies; automation transforms activities within jobs
- SAT (Mexico) — CFDI 4.0, issuance and cancellation requirements for digital tax invoices (Anexo 20)
- ISO 9001:2015 — documented information and process traceability
- Price ranges and timelines match the current offer at /automatizacion and are written-quote ranges
Which process is costing you the most hours?
The free diagnostic takes 3 minutes, scores your operational maturity, and identifies the process worth addressing first. No card and no required call.