AI automation for small teams: which workflow to automate first (and which to leave alone)
Bu yazıyı Türkçe oku → · Srpski →
Short answer: automate the task that repeats every day, whose mistakes are cheap to catch, and that passes between at least two people. Leave the judgement calls — pricing exceptions, legal wording, medical decisions — with a person, and automate the collecting, drafting and routing around them.
Most teams that ask us about AI automation arrive with a list of twelve things they would like to stop doing by hand. The list is usually right. The order is usually wrong. The first workflow you automate teaches the whole team what automation is, so it has to work on day one and it has to be visible.
The three-question score
For each candidate task, answer three questions with a number from 1 to 5.
| Question | 1 | 5 |
|---|---|---|
| How often does it repeat? | Monthly or rarer | Several times a day |
| How cheap is a mistake? | Costly and invisible | Obvious and reversible in a minute |
| How many handoffs? | One person, start to finish | Three or more people or tools |
Multiply the three numbers. Anything scoring 60 or above is a first-week candidate. Anything with a “mistake” score of 1 goes to the bottom of the list no matter how often it repeats — that is where you keep a human.
What usually scores highest
- Inbound requests that need a reply and a record. A quote request, a booking, a support question. It repeats constantly, the reply is easy to check, and it crosses from inbox to spreadsheet to calendar. This is why we built an AI receptionist before anything else.
- Reminders and follow-ups. Unpaid invoices, unconfirmed appointments, quotes that went quiet. Cheap to get slightly wrong, expensive to forget entirely.
- Data that is retyped between two systems. Orders into accounting, leads into the CRM. The most boring work in the company and the easiest to hand over.
What usually scores lowest
Anything where the person doing it is applying judgement you could not write down: exceptions to pricing, the wording of a legal reply, a clinician's assessment, the final approval on a payment. These are not bad candidates because AI is incapable; they are bad first candidates because a mistake is expensive and quiet. Automate the preparation — gather the documents, draft the reply, put the numbers side by side — and let the person decide in a tenth of the time.
The rule we apply to ourselves
Every automation we ship has a deterministic guard around the language model: a rule in code that checks the output before it leaves. When we built the receptionist, we measured that 38% of unassisted replies ended the conversation without inviting a response, so we added a rule that rejects those. A prompt asking politely for the same thing failed repeatedly in testing; code did not. If a workflow cannot be guarded that way, it is not ready to run unattended.
What it costs and how long it takes
Our packages are public and fixed. Operations is one end-to-end workflow plus a landing site and admin panel: €450 build, €69 a month, three weeks. Growth adds a multi-workflow system and a five-to-seven page site: €750 and €99 a month. Operations OS is the multi-team version with a white-label panel: €850 and €139 a month. Every build fee has a 30-day money-back guarantee, and subscriptions cancel monthly.
If you have the twelve-item list, send it. We will score it with you on a 20-minute call and tell you which one to build first — including if the honest answer is “none of these yet”.
Frequently asked questions
How do I choose the first process to automate with AI?
Score each candidate on three things: how often it repeats, how expensive a mistake is, and how many people it passes through. Automate the one that repeats daily, where an error is cheap to catch, and that crosses at least one handoff. That combination pays back fastest and teaches the team what automation feels like without risk.
Which workflows should not be automated?
Anything where a wrong answer is costly and hard to detect: pricing exceptions, legal wording, medical judgement, final sign-off on money. Automate the work around those steps — collecting, drafting, routing, reminding — and keep the decision with a person.
How long does a first AI workflow take to build?
Our Operations package delivers one end-to-end workflow, a landing site and an admin panel in three weeks. The build itself is a few days; most of the calendar time is mapping the real process with the team and testing on real cases.
What does it cost?
Our packages are public: Operations at €450 build plus €69/month, Growth at €750 plus €99/month, Operations OS at €850 plus €139/month. Every build fee carries a 30-day money-back guarantee.
Do I need to change my tools?
Usually no. A workflow engine sits next to your existing inbox, spreadsheet, CRM or calendar and moves work between them. Replacing tools is a separate decision, and often unnecessary.
— Written by the Bulsu Labs team · Talk to us