How to automate your SMB processes — from inventory to first deployment

Running an SMB often means juggling flaming swords: late emails, invoices to chase, reports to compile, teams to keep in sync. The temptation is to "add another tool". It's almost always the wrong answer.

Here's how I move a client from chaos to a fluid system. Six weeks typically. No revolution, no big bang — just a sequence that works.

Step 0 — Map before coding

The first 3 hours of an automation engagement happen without a screen. I just ask:

When was the last time you (or your team) said "this task again" this week?
What was the last costly error linked to a manual re-entry?
Where do information items get lost between your tools?

These three questions give me 80% of the map. The rest is field observation — I shadow the assistant or sales rep for 2-3 hours, and note the useless clicks.

Step 1 — List the 10 highest-ROI tasks

I build a simple table:

Task	Frequency	Unit time	Total time/month	Complexity	ROI Score
Example: quote entry → accounting	3×/day	8 min	10h/month	Low	★★★★★
Example: unpaid follow-up	1×/week	30 min	2h/month	Medium	★★★

The "ROI Score" is subjective but must be shared with the team — otherwise the team will sabotage the automation it doesn't understand.

Rule I impose on myself: never more than 3 builds in parallel. Beyond that, nothing ships.

Step 2 — Centralise what already exists

Before adding anything, I look at what's already plugged in and could be plugged in better. Nine times out of ten, the SMB already has:

A CRM (HubSpot, Pipedrive, Zoho, or a homemade Airtable).
An accounting tool (Sage, Pennylane, Indy, Qonto).
A project tool (Notion, ClickUp, Asana, Trello).
Shared spreadsheets for everything that doesn't fit.

First action: connect these tools cleanly (API or webhook) rather than replacing them. 70% of the gain comes from there, without changing habits.

Step 3 — Automate recurring tasks

My preferred tools, in order of preference depending on the case:

Case	Tool	Why
Simple SaaS → SaaS chains	Make or Zapier	Fast, reliable, light maintenance
Business logic, complex orchestration	n8n self-hosted	Owned, no variable cost, extensible
Integration with legacy ERP or business database	Custom Node/Python code	Full control, long term
AI extraction (PDF, email, documents)	LLM + n8n	Precision, low cost

In a typical SMB, I rarely have more than 2 tools in production. Simplicity always beats sophistication.

Step 4 — Make automation visible

The worst scenario: automation runs perfectly, and no one knows it. Result: at the first bug (always, one day), everyone panics.

What I systematically put in place:

A simple operational dashboard (Metabase, Grafana, or custom) showing active flows, errors, volumes processed.
Targeted alerts (Slack, email, SMS) — never in bulk, only when action is needed.
A one-page doc readable by a non-dev, describing what each automation does and who to alert if it breaks.

Step 5 — Cut errors, not people

Well-led automation doesn't replace your teams — it replaces the tasks that wear them out. The difference is fundamental. Announce from day one that no one will be let go: you'll gain everyone's collaboration.

What I see in practice after 3 months of automation in an SMB of 15 to 30 people:

5 to 12 hours per week recovered across the whole team.
Error rate divided by 3 to 10 on automated tasks.
Capacity to absorb 30 to 50% more volume without hiring.

These aren't marketing promises — they're the orders of magnitude I've seen come back across my engagements. The exact figure always depends on the starting point.

Step 6 — Reassess, always

An automation that worked in January may be obsolete in June: API change, new internal workflow, replaced tool. I always deliver with:

A short quarterly review (2h) to check everything's running.
A clear entry point to adjust without starting from scratch.
An up-to-date doc any serious dev can pick up.

What's not to automate

Three cases where I recommend not automating, even if it's technically possible:

High-empathy client interactions — first commercial contact, handling an upset client, signing a large account contract.
Strategic decisions — AI can prepare, but you decide.
Tasks done once a year — automation cost is never profitable.

Where to start

If you want me to look at it together, I offer 30 minutes of discussion to map your case and tell you what's worth automating. Book a slot when you want — and come with the list of the three tasks that annoy you the most.